feat(app): add experimental vertical session tabs (#45210)

Co-authored-by: Brendonovich <14191578+Brendonovich@users.noreply.github.com>

.github/workflows/test.yml

@@ -42,7 +42,7 @@ jobs:
       - name: Find affected packages
         id: packages
         env:
-          TURBO_SCM_BASE: ${{ github.event.pull_request.base.sha || github.event.before }}
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
           TURBO_SCM_HEAD: ${{ github.sha }}
         run: |
           if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
@@ -71,6 +71,7 @@ jobs:
         uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4.3.1
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
+          fetch-depth: 0
 
       - name: Setup Node
         uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
@@ -110,9 +111,16 @@ jobs:
 
       - name: Run unit tests
         timeout-minutes: 20
-        run: GITHUB_ACTIONS=false bun turbo test
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            GITHUB_ACTIONS=false bun turbo test
+            exit 0
+          fi
+          GITHUB_ACTIONS=false bun turbo test --affected
         env:
           OPENCODE_EXPERIMENTAL_DISABLE_FILEWATCHER: ${{ runner.os == 'Windows' && 'true' || 'false' }}
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
 
       - name: Verify published codemode package
         if: runner.os == 'Linux'
@@ -122,8 +130,15 @@ jobs:
       - name: Verify packed workerd SDK
         if: runner.os == 'Linux'
         timeout-minutes: 15
-        working-directory: packages/sdk
-        run: bun run verify:package
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            bun turbo verify:package --filter=@opencode-ai/sdk
+            exit 0
+          fi
+          bun turbo verify:package --affected --filter=@opencode-ai/sdk
+        env:
+          TURBO_SCM_BASE: ${{ github.event_name == 'pull_request' && format('{0}^1', github.sha) || github.event.before }}
+          TURBO_SCM_HEAD: ${{ github.sha }}
 
       - name: Verify compiled service lifecycle
         if: always()

bun.lock

@@ -346,9 +346,7 @@
         "@ai-sdk/amazon-bedrock": "4.0.112",
         "@ai-sdk/anthropic": "3.0.82",
         "@ai-sdk/azure": "3.0.88",
-        "@ai-sdk/cerebras": "2.0.41",
         "@ai-sdk/cohere": "3.0.27",
-        "@ai-sdk/deepinfra": "2.0.41",
         "@ai-sdk/gateway": "3.0.104",
         "@ai-sdk/google-vertex": "4.0.128",
         "@ai-sdk/groq": "3.0.31",
@@ -357,7 +355,6 @@
         "@ai-sdk/perplexity": "3.0.26",
         "@ai-sdk/provider": "3.0.8",
         "@ai-sdk/provider-utils": "4.0.23",
-        "@ai-sdk/togetherai": "2.0.41",
         "@ai-sdk/vercel": "2.0.39",
         "@aws-sdk/credential-providers": "3.1057.0",
         "@ff-labs/fff-bun": "0.10.5",
@@ -667,6 +664,7 @@
       "dependencies": {
         "@opencode-ai/client": "workspace:*",
         "@opencode-ai/core": "workspace:*",
+        "@opencode-ai/plugin": "workspace:*",
         "@opencode-ai/schema": "workspace:*",
         "@opencode-ai/server": "workspace:*",
         "@opencode-ai/util": "workspace:*",
@@ -973,6 +971,7 @@
       "dependencies": {
         "@effect/opentelemetry": "catalog:",
         "@effect/platform-node": "catalog:",
+        "@effect/platform-node-shared": "catalog:",
         "@npmcli/arborist": "catalog:",
         "@npmcli/config": "10.8.1",
         "@opentelemetry/api": "1.9.0",
@@ -1178,8 +1177,6 @@
 
     "@ai-sdk/deepgram": ["@ai-sdk/deepgram@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8pcrQvEQCbvrrQKnD6hclBbI0hUgSrgyADykRbabxv/g9vPurfMC6n23J7dD+KZ3EcCoW+qz3IUIfySJ58gBOg=="],
 
-    "@ai-sdk/deepinfra": ["@ai-sdk/deepinfra@2.0.41", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-y6RoOP7DGWmDSiSxrUSt5p18sbz+Ixe5lMVPmdE7x+Tr5rlrzvftyHhjWHfqlAtoYERZTGFbP6tPW1OfQcrb4A=="],
-
     "@ai-sdk/deepseek": ["@ai-sdk/deepseek@2.0.47", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@ai-sdk/provider-utils": "4.0.38" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MzcQ321JO8OY+TVLFI81A7cIIuoeLLxrLCDD+8C1E3Ro6UFyfMtRXo9bw9OhTMRSDMo6hgSDOo4Fekz8aJtQYQ=="],
 
     "@ai-sdk/elevenlabs": ["@ai-sdk/elevenlabs@2.0.52", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-ZgkausouWvO9U4ZtowNJ093bSNYOvH8zqls3uLC3+oxzWvbbTZO8SOdmFk0+gGafsXFJvq2yUX9+rEeJPwOJLw=="],
@@ -1206,8 +1203,6 @@
 
     "@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.23", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-z8GlDaCmRSDlqkMF2f4/RFgWxdarvIbyuk+m6WXT1LYgsnGiXRJGTD2Z1+SDl3LqtFuRtGX1aghYvQLoHL/9pg=="],
 
-    "@ai-sdk/togetherai": ["@ai-sdk/togetherai@2.0.41", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-k3p9e3k0/gpDDyTtvafsK4HYR4D/aUQW/kzCwWo1+CzdBU84i4L14gWISC/mv6tgSicMXHcEUd521fPufQwNlg=="],
-
     "@ai-sdk/vercel": ["@ai-sdk/vercel@2.0.39", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.37", "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-8eu3ljJpkCTP4ppcyYB+NcBrkcBoSOFthCSgk5VnjaxnDaOJFaxnPwfddM7wx3RwMk2CiK1O61Px/LlqNc7QkQ=="],
 
     "@ai-sdk/xai": ["@ai-sdk/xai@3.0.123", "", { "dependencies": { "@ai-sdk/openai-compatible": "2.0.69", "@ai-sdk/provider": "3.0.15", "@ai-sdk/provider-utils": "4.0.46" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-WNASvd1C516oh2qYIj9EvAVPdU+Abads8DQWU6p9lQtvFFeGh8QW+3LDOARZd1GCINUFfw5yadEK845SMQKLsA=="],
@@ -5912,10 +5907,6 @@
 
     "@ai-sdk/deepgram/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.46", "", { "dependencies": { "@ai-sdk/provider": "3.0.15", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8", "undici": "^6.28.0" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-tEtld97plCFiYevsJuOkGkeuhQndeMWFBVrJS4AjnbD5AqrNSXRCe0p+BZ3Cju/sxDeeZ9ym3q9YUV8fASA7aQ=="],
 
-    "@ai-sdk/deepinfra/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
-
-    "@ai-sdk/deepinfra/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
-
     "@ai-sdk/deepseek/@ai-sdk/provider": ["@ai-sdk/provider@3.0.14", "", { "dependencies": { "json-schema": "^0.4.0" } }, "sha512-5X1k57JBJ4H7H1QjX7CnJYAB1I19r/trVZTMcSms7/kLNZ8RaU4Nt2agcwZzv82Hfx6Q7/TOLU7agAKeFfc8cA=="],
 
     "@ai-sdk/deepseek/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.38", "", { "dependencies": { "@ai-sdk/provider": "3.0.14", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.8" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-/HHGmtKllqjg1OLc023v9w9kK3laW7Z6TzfZukYQWCsGBbzB9p60zTvvpXFVcs44NZBVXL3viOa1HRKUbeee8g=="],
@@ -5954,10 +5945,6 @@
 
     "@ai-sdk/perplexity/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
 
-    "@ai-sdk/togetherai/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
-
-    "@ai-sdk/togetherai/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],
-
     "@ai-sdk/vercel/@ai-sdk/openai-compatible": ["@ai-sdk/openai-compatible@2.0.37", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@ai-sdk/provider-utils": "4.0.21" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-+POSFVcgiu47BK64dhsI6OpcDC0/VAE2ZSaXdXGNNhpC/ava++uSRJYks0k2bpfY0wwCTgpAWZsXn/dG2Yppiw=="],
 
     "@ai-sdk/vercel/@ai-sdk/provider-utils": ["@ai-sdk/provider-utils@4.0.21", "", { "dependencies": { "@ai-sdk/provider": "3.0.8", "@standard-schema/spec": "^1.1.0", "eventsource-parser": "^3.0.6" }, "peerDependencies": { "zod": "^3.25.76 || ^4.1.8" } }, "sha512-MtFUYI1/8mgDvRmaBDjbLJPFFrMG777AvSgyIFQtZHIMzm88R/12vYBBpnk7pfiWLFE1DSZzY4WDYzGbKAcmiw=="],

nix/hashes.json

@@ -1,8 +1,8 @@
 {
   "nodeModules": {
-    "x86_64-linux": "sha256-3Jx1Q7hl+Y0Log/k2vd5y6dzBpzFKWlhShPESxn1Rm4=",
-    "aarch64-linux": "sha256-EiiI6g01oBIrExCMAUgT3w82P0fvu4FAJhI32C+ze0I=",
-    "aarch64-darwin": "sha256-s+w49HRp1+ewtiTaU65tPWjUiO1NQw3kzfemMEEQZb0=",
-    "x86_64-darwin": "sha256-/Ee5V7pnL/qm3c4ZHeWEjH7FhGVXArXryOugbG5vsz8="
+    "x86_64-linux": "sha256-t7k0Nk9Z7FMkAKt2UReI1jD1KoPB+LhQQx1x1p5Htlc=",
+    "aarch64-linux": "sha256-tU7C2zYKiX+RpwBBBLjX7vrqQNLPvDc5Uz9EbgGJvSA=",
+    "aarch64-darwin": "sha256-A3o+2zNuFHV5FZx/azdXBjcKdS5o41b2e2Q1Qq8C1m0=",
+    "x86_64-darwin": "sha256-NXgvteaxJQqBBZIHnQbivghHTJnL2IMZiFPNssGp2Os="
   }
 }

packages/ai/script/setup-recording-env.ts

@@ -157,9 +157,9 @@ const PROVIDERS: ReadonlyArray<Provider> = [
     id: "togetherai",
     label: "TogetherAI",
     tier: "compatible",
-    note: "Existing OpenAI-compatible text/tool recorded tests",
-    vars: [{ name: "TOGETHER_AI_API_KEY" }],
-    validate: (env) => validateBearer("https://api.together.xyz/v1/models", Redacted.make(env.TOGETHER_AI_API_KEY)),
+    note: "Native Together AI text/tool recorded tests",
+    vars: [{ name: "TOGETHER_API_KEY" }],
+    validate: (env) => validateBearer("https://api.together.xyz/v1/models", Redacted.make(env.TOGETHER_API_KEY)),
   },
   {
     id: "minimax",
@@ -200,8 +200,8 @@ const PROVIDERS: ReadonlyArray<Provider> = [
   {
     id: "cerebras",
     label: "Cerebras",
-    tier: "optional",
-    note: "OpenAI-compatible bridge",
+    tier: "compatible",
+    note: "Native Cerebras text/tool/tool-loop recorded tests",
     vars: [{ name: "CEREBRAS_API_KEY" }],
     validate: (env) => validateBearer("https://api.cerebras.ai/v1/models", Redacted.make(env.CEREBRAS_API_KEY)),
   },

packages/ai/src/cache-policy.ts

@@ -36,7 +36,7 @@ const resolve = (policy: CachePolicy | undefined): CachePolicyObject => {
 // Protocols whose wire format ignores inline cache markers (OpenAI's implicit
 // prefix caching, Gemini's implicit + out-of-band CachedContent). Skip the
 // whole policy pass for these — emitting hints would be harmless but pointless.
-const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "bedrock-converse", "openrouter"])
+const RESPECTS_INLINE_HINTS = new Set(["anthropic-messages", "google-vertex-messages", "bedrock-converse", "openrouter"])
 
 const makeHint = (ttlSeconds: number | undefined): CacheHint =>
   ttlSeconds !== undefined ? new CacheHint({ type: "ephemeral", ttlSeconds }) : new CacheHint({ type: "ephemeral" })

packages/ai/src/protocols/anthropic-messages.ts

@@ -744,9 +744,12 @@ const endsInServerToolUse = (message: LLMRequest["messages"][number]) => {
   return message.role === "assistant" && last?.type === "tool-call" && last.providerExecuted === true
 }
 
-const canUseNativeSystemUpdate = (messages: LLMRequest["messages"], index: number) => {
-  const previous = messages[index - 1]
-  const next = messages[index + 1]
+const canUseNativeSystemUpdate = (request: LLMRequest, index: number) => {
+  const previous = request.messages[index - 1]
+  const next = request.messages[index + 1]
+  // Vertex currently rejects/404s for a system message after local tool results,
+  // so fold it into the user tool-result turn across continuations and history.
+  if (request.model.route.id === "google-vertex-messages" && previous?.role === "tool") return false
   return (
     previous !== undefined &&
     previous.role !== "system" &&
@@ -793,7 +796,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
     if (message.role === "system") {
       if (splitsLocalToolResults(request.messages, index))
         return yield* invalid("Anthropic Messages system updates cannot split a local tool call from its tool result")
-      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request.messages, index)) {
+      if (supportsNativeSystemUpdates(request) && canUseNativeSystemUpdate(request, index)) {
         messages.push(yield* lowerNativeSystemUpdate(message, breakpoints))
         continue
       }

packages/ai/src/protocols/open-responses.ts

@@ -196,7 +196,7 @@ type LoweredInputItem =
 // multiple streamed summary parts into the same item before flushing.
 type OpenResponsesReasoningInput = {
   type: "reasoning"
-  id: string
+  id?: string
   summary: Array<{ type: "summary_text"; text: string }>
   encrypted_content?: string | null
 }
@@ -372,9 +372,6 @@ export const Event = Schema.StructWithRest(
 )
 export type Event = Schema.Schema.Type<typeof Event>
 
-// Which lowered input item a persisted item id is about to be attached to.
-export type ItemKind = "message" | "reasoning" | "function-call" | "hosted-tool"
-
 export interface Extension {
   readonly id: string
   readonly name: string
@@ -384,10 +381,6 @@ export interface Extension {
     readonly request: LLMRequest
   }) => MediaInput | undefined
   readonly lowerHostedToolItem?: (item: unknown) => ExtendedHostedToolItem | undefined
-  // Optional grammar check applied before a persisted item id is resent as
-  // part of replayed history. Returning false drops the id; every lowered
-  // item treats a dropped id the same as an absent one.
-  readonly acceptsItemID?: (kind: ItemKind, id: string) => boolean
 }
 
 const BASE: Extension = { id: ADAPTER, name: NAME }
@@ -447,53 +440,37 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL
     tool: (toolName) => ({ type: "function" as const, name: toolName }),
   })
 
-// Servers validate item ids on replayed history, and a malformed or oversized
-// id can fail an otherwise valid request. Only server-issued tokens are worth
-// resending; anything else is treated as absent so the item is resent without
-// an id (or skipped, for items that cannot be expressed without one).
-const ITEM_ID_PATTERN = /^[A-Za-z0-9_-]{1,64}$/
+// Server-issued item ids need a nonempty prefix and suffix, but the prefix is
+// provider-defined and does not necessarily identify the item's semantic type.
 const itemID = (providerMetadata: ProviderMetadata | undefined, providerMetadataKey: string) => {
   const metadata = providerMetadata?.[providerMetadataKey]
-  return ProviderShared.isRecord(metadata) &&
-    typeof metadata.itemId === "string" &&
-    ITEM_ID_PATTERN.test(metadata.itemId)
-    ? metadata.itemId
-    : undefined
+  if (!ProviderShared.isRecord(metadata) || typeof metadata.itemId !== "string") return undefined
+  const separator = metadata.itemId.indexOf("_")
+  return separator > 0 && separator < metadata.itemId.length - 1 ? metadata.itemId : undefined
 }
 
-const acceptsItemID = (extension: Extension, kind: ItemKind, id: string | undefined): id is string =>
-  id !== undefined && (extension.acceptsItemID?.(kind, id) ?? true)
-
-const lowerToolCall = (
-  part: ToolCallPart,
-  providerMetadataKey: string,
-  extension: Extension,
-): OpenResponsesInputItem => {
+const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
   const id = itemID(part.providerMetadata, providerMetadataKey)
   return {
     type: "function_call",
-    ...(acceptsItemID(extension, "function-call", id) ? { id } : {}),
+    ...(id === undefined ? {} : { id }),
     call_id: part.id,
     name: part.name,
     arguments: ProviderShared.encodeJson(part.input),
   }
 }
 
-const lowerReasoning = (
-  part: ReasoningPart,
-  providerMetadataKey: string,
-  extension: Extension,
-): OpenResponsesReasoningInput | undefined => {
+const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
   const metadata = part.providerMetadata?.[providerMetadataKey]
+  if (!ProviderShared.isRecord(metadata)) return undefined
   const id = itemID(part.providerMetadata, providerMetadataKey)
-  if (!ProviderShared.isRecord(metadata) || !acceptsItemID(extension, "reasoning", id)) return undefined
   const encryptedContent =
     typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
       ? metadata.reasoningEncryptedContent
       : undefined
   return {
     type: "reasoning",
-    id,
+    ...(id === undefined ? {} : { id }),
     summary: part.text.length > 0 ? [{ type: "summary_text", text: part.text }] : [],
     encrypted_content: encryptedContent,
   }
@@ -584,9 +561,7 @@ const lowerToolResultOutput = Effect.fnUntraced(function* (
 })
 
 const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (request: LLMRequest, extension: Extension) {
-  const system: LoweredInputItem[] =
-    request.system.length === 0 ? [] : [{ role: "system", content: ProviderShared.joinText(request.system) }]
-  const input: LoweredInputItem[] = [...system]
+  const input: LoweredInputItem[] = []
   const providerMetadataKey = request.model.route.providerMetadataKey ?? "openresponses"
 
   for (const message of request.messages) {
@@ -616,8 +591,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
           Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
         >((groups, part) => {
           const metadata = part.providerMetadata?.[providerMetadataKey]
-          const rawID = itemID(part.providerMetadata, providerMetadataKey)
-          const id = acceptsItemID(extension, "message", rawID) ? rawID : undefined
+          const id = itemID(part.providerMetadata, providerMetadataKey)
           const phase = ProviderShared.isRecord(metadata) ? messagePhase(metadata.phase) : undefined
           const group = groups.at(-1)
           if (group && group.id === id && group.phase === phase) group.parts.push(part)
@@ -642,23 +616,23 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
         }
         if (part.type === "reasoning") {
           flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey, extension)
+          const reasoning = lowerReasoning(part, providerMetadataKey)
           if (!reasoning) continue
-          const existing = reasoningItems[reasoning.id]
+          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
           if (existing) {
             existing.summary.push(...reasoning.summary)
             if (typeof reasoning.encrypted_content === "string")
               existing.encrypted_content = reasoning.encrypted_content
             continue
           }
-          reasoningItems[reasoning.id] = reasoning
+          if (reasoning.id !== undefined) reasoningItems[reasoning.id] = reasoning
           input.push(reasoning)
           continue
         }
         if (part.type === "tool-call") {
           flushText()
           if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey, extension))
+          input.push(lowerToolCall(part, providerMetadataKey))
           continue
         }
         if (part.type === "tool-result" && part.providerExecuted === true) {
@@ -670,7 +644,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
               : Schema.is(HostedToolItem)(part.result.value)
                 ? part.result.value
                 : extension.lowerHostedToolItem?.(part.result.value)
-          if (acceptsItemID(extension, "hosted-tool", id) && hosted?.id === id) {
+          if (id !== undefined && hosted?.id === id) {
             if (!hostedToolItems.has(id)) {
               input.push(hosted)
               hostedToolItems.add(id)
@@ -716,10 +690,11 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (reques
 
 const lowerOptions = (request: LLMRequest) => {
   const options = OpenResponsesOptions.resolve(request)
+  const instructions = ProviderShared.joinText(request.system)
   const cacheKey = ProviderShared.promptCacheKey(request)
   const parallelToolCalls = resolveParallelToolCalls(request)
   return {
-    ...(options.instructions ? { instructions: options.instructions } : {}),
+    ...(instructions ? { instructions } : {}),
     ...(options.store !== undefined ? { store: options.store } : {}),
     ...(options.metadata ? { metadata: options.metadata } : {}),
     ...(options.safetyIdentifier ? { safety_identifier: options.safetyIdentifier } : {}),
@@ -945,25 +920,23 @@ const onOutputItemAdded = (state: ParserState, event: Event): StepResult => {
       events,
     ]
   }
-  if (item?.type !== "function_call" || !item.id) return [state, NO_EVENTS]
-  const metadata = providerMetadata(state, { itemId: item.id })
+  if (item?.type !== "function_call" || !item.call_id) return [state, NO_EVENTS]
+  const id = item.id ?? item.call_id
+  const metadata = item.id ? providerMetadata(state, { itemId: item.id }) : undefined
   const events: LLMEvent[] = []
   const lifecycle = Lifecycle.stepStart(state.lifecycle, events)
   return [
     {
       ...state,
       lifecycle,
-      tools: ToolStream.start(state.tools, item.id, {
-        id: item.call_id ?? item.id,
+      tools: ToolStream.start(state.tools, id, {
+        id: item.call_id,
         name: item.name ?? "",
         input: item.arguments ?? "",
         providerMetadata: metadata,
       }),
     },
-    [
-      ...events,
-      LLMEvent.toolInputStart({ id: item.call_id ?? item.id, name: item.name ?? "", providerMetadata: metadata }),
-    ],
+    [...events, LLMEvent.toolInputStart({ id: item.call_id, name: item.name ?? "", providerMetadata: metadata })],
   ]
 }
 
@@ -1095,18 +1068,19 @@ const onOutputItemDone = Effect.fn("OpenResponses.onOutputItemDone")(function* (
   }
 
   if (item.type === "function_call") {
-    if (!item.id || !item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
-    const tools = state.tools[item.id]
+    if (!item.call_id || !item.name) return [state, NO_EVENTS] satisfies StepResult
+    const id = item.id ?? item.call_id
+    const tools = state.tools[id]
       ? state.tools
-      : ToolStream.start(state.tools, item.id, {
+      : ToolStream.start(state.tools, id, {
           id: item.call_id,
           name: item.name,
-          providerMetadata: providerMetadata(state, { itemId: item.id }),
+          providerMetadata: item.id ? providerMetadata(state, { itemId: item.id }) : undefined,
         })
     const result =
       item.arguments === undefined
-        ? yield* ToolStream.finish(state.id, tools, item.id)
-        : yield* ToolStream.finishWithInput(state.id, tools, item.id, item.arguments)
+        ? yield* ToolStream.finish(state.id, tools, id)
+        : yield* ToolStream.finishWithInput(state.id, tools, id, item.arguments)
     const events: LLMEvent[] = []
     const resultEvents = result.events ?? []
     const lifecycle = resultEvents.length ? Lifecycle.stepStart(state.lifecycle, events) : state.lifecycle
@@ -1160,10 +1134,11 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
           event.response?.output ?? [],
           () => [state, NO_EVENTS] satisfies StepResult,
           ([current, events], item) => {
+            const id = item.id ?? (item.type === "function_call" ? item.call_id : undefined)
             if (
-              !item.id ||
-              ((item.type !== "function_call" || !current.tools[item.id]) &&
-                (item.type !== "reasoning" || !current.reasoningItems[item.id]))
+              !id ||
+              ((item.type !== "function_call" || !current.tools[id]) &&
+                (item.type !== "reasoning" || !current.reasoningItems[id]))
             )
               return Effect.succeed([current, events] satisfies StepResult)
             return onOutputItemDone(current, { type: "response.output_item.done", item }).pipe(
@@ -1289,10 +1264,11 @@ export const step = (state: ParserState, input: Event) => {
   if (event.type === "response.output_item.added") {
     if (event.item?.type === "message" && !event.item.id)
       return ProviderShared.eventError(state.id, `${event.type} message is missing id`)
+    const id = event.item?.id ?? (event.item?.type === "function_call" ? event.item.call_id : undefined)
     return Effect.succeed(
       onOutputItemAdded(
-        event.output_index !== undefined && event.item?.id
-          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: event.item.id } }
+        event.output_index !== undefined && id
+          ? { ...state, outputItems: { ...state.outputItems, [event.output_index]: id } }
           : state,
         event,
       ),

packages/ai/src/protocols/openai-chat.ts

@@ -364,8 +364,9 @@ const lowerUserMessage = Effect.fn("OpenAIChat.lowerUserMessage")(function* (
 
 const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(function* (
   message: OpenAIChatRequestMessage,
-  configuredField?: string,
-  options: LoweringOptions = {},
+  configuredField: string | undefined,
+  requireReasoning: boolean,
+  options: LoweringOptions,
 ) {
   const content: TextPart[] = []
   const reasoning: ReasoningPart[] = []
@@ -392,15 +393,17 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
   const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
   const fullyStructured = reasoning.every((part) => Array.isArray(part.providerMetadata?.openai?.reasoningDetails))
   const field = (() => {
-    if (configuredField !== undefined) return configuredField
-    if (reasoning.length === 0) return undefined
+    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
+      return configuredField
+    if (reasoning.length === 0) return requireReasoning ? "reasoning_content" : undefined
     if (observedField !== undefined) return observedField
     if (nativeReasoning !== undefined) return "reasoning_content"
-    if (!fullyStructured) return "reasoning_content"
+    if (!fullyStructured || requireReasoning) return "reasoning_content"
   })()
   const reasoningText = (() => {
-    if (configuredField !== undefined) return reasoning.length === 0 ? (nativeReasoning ?? "") : text
-    if (reasoning.length === 0) return nativeReasoning
+    if (configuredField !== undefined)
+      return reasoning.length === 0 ? (nativeReasoning ?? (requireReasoning ? "" : undefined)) : text
+    if (reasoning.length === 0) return nativeReasoning ?? (requireReasoning ? "" : undefined)
     return text
   })()
   const cached = message.content.findLast((part) => "cache" in part && part.cache !== undefined)
@@ -454,11 +457,13 @@ const lowerToolMessages = Effect.fn("OpenAIChat.lowerToolMessages")(function* (
 
 const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
   message: OpenAIChatRequestMessage,
-  reasoningField?: string,
-  options: LoweringOptions = {},
+  reasoningField: string | undefined,
+  requireReasoning: boolean,
+  options: LoweringOptions,
 ) {
   if (message.role === "user") return [yield* lowerUserMessage(message, options)]
-  if (message.role === "assistant") return [yield* lowerAssistantMessage(message, reasoningField, options)]
+  if (message.role === "assistant")
+    return [yield* lowerAssistantMessage(message, reasoningField, requireReasoning, options)]
   return (yield* lowerToolMessages(message, options)).messages
 })
 
@@ -480,6 +485,13 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
         : [{ role: "system", content: ProviderShared.joinText(request.system) }]
   const messages = [...system]
   const modelID = request.model.id.toLowerCase()
+  const requireReasoning =
+    request.model.compatibility?.requireReasoning ??
+    (request.model.compatibility?.reasoningField !== undefined ||
+      request.model.provider === "deepseek" ||
+      request.model.route.endpoint.baseURL?.toLowerCase().includes("deepseek.com") ||
+      modelID.includes("deepseek"))
+  const reasoningField = request.model.compatibility?.reasoningField
   const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
   const lowering = {
     ...options,
@@ -554,7 +566,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
       continue
     }
     flushImages()
-    messages.push(...(yield* lowerMessage(message, request.model.compatibility?.reasoningField, lowering)))
+    messages.push(...(yield* lowerMessage(message, reasoningField, requireReasoning, lowering)))
   }
   flushImages()
   return messages

packages/ai/src/protocols/openai-responses.ts

@@ -86,28 +86,10 @@ const OpenAIResponsesBody = Schema.Struct({
 })
 export type OpenAIResponsesBody = Schema.Schema.Type<typeof OpenAIResponsesBody>
 
-// Replayed items are paired with stored server state by id, so a foreign or
-// synthetic token can fail request validation even when `call_id` pairing is
-// intact. Only resend ids in each item kind's own grammar; hosted tool
-// items keep generic validation because every hosted tool mints its own
-// prefix. The same allowlist approach codex uses before resending history
-// (codex-rs core/src/client.rs, `prepare_response_items_for_request`).
-const ITEM_ID_PREFIXES: Record<OpenResponses.ItemKind, ReadonlyArray<string>> = {
-  message: ["msg_"],
-  reasoning: ["rs_"],
-  "function-call": ["fc_"],
-  // Every hosted tool mints its own id prefix, so items keep generic validation.
-  "hosted-tool": [],
-}
-
 const extension = {
   id: ADAPTER,
   name: NAME,
   lowerHostedToolItem: (item: unknown) => (Schema.is(OpenAIResponsesHostedToolItem)(item) ? item : undefined),
-  acceptsItemID: (kind: OpenResponses.ItemKind, id: string) => {
-    const prefixes = ITEM_ID_PREFIXES[kind]
-    return prefixes.length === 0 || prefixes.some((prefix) => id.startsWith(prefix))
-  },
 } satisfies OpenResponses.Extension
 
 const nativeImageToolInput = (tool: ToolDefinition) => {

packages/ai/src/protocols/shared.ts

@@ -210,10 +210,9 @@ export const errorText = (error: unknown) => {
  * `framing` step for Server-Sent Events. Decodes UTF-8, runs the SSE channel
  * decoder, optionally filters named events, and drops empty / `[DONE]`
  * keep-alive events so the protocol event schema sees one JSON string per
- * element. The SSE channel emits a
- * `Retry` control event on its error channel; we drop it here (we don't
- * implement client-driven retries). Decoder failures become provider output
- * errors so the public error channel stays `AIError`.
+ * element. Retry control events are ignored without interrupting the stream.
+ * Decoder failures become provider output errors so the public error channel
+ * stays `AIError`.
  */
 export const sseFraming = (
   bytes: Stream.Stream<Uint8Array, AIError>,
@@ -221,9 +220,23 @@ export const sseFraming = (
 ): Stream.Stream<string, AIError> =>
   bytes.pipe(
     Stream.decodeText(),
-    Stream.pipeThroughChannel(Sse.decode()),
-    Stream.catchTag("Retry", () => Stream.empty),
-    Stream.catchTag("SseError", (error) => Stream.fail(eventError("sse", error.message))),
+    Stream.mapAccumEffect(
+      () => {
+        const output: Sse.Event[] = []
+        return {
+          output,
+          parser: Sse.makeParser((event) => {
+            if (event._tag === "Event") output.push(event)
+          }),
+        }
+      },
+      (state, chunk) =>
+        Effect.gen(function* () {
+          const error = state.parser.feed(chunk)
+          if (error) return yield* eventError("sse", error.message)
+          return [state, state.output.splice(0)] as const
+        }),
+    ),
     Stream.filter(
       (event) =>
         (events === undefined || events.has(event.event)) &&

packages/ai/src/protocols/utils/open-responses-options.ts

@@ -56,7 +56,6 @@ export const StreamOptions = Schema.Struct({
 })
 
 export const Options = Schema.Struct({
-  instructions: Schema.optional(Schema.String),
   store: Schema.optional(Schema.Boolean),
   metadata: Schema.optional(Schema.Record(Schema.String, Schema.String)),
   safetyIdentifier: Schema.optional(Schema.String),

packages/ai/src/provider-error.ts

@@ -40,15 +40,16 @@ const patterns = [
   /model_context_window_exceeded/i,
   /too many tokens/i,
   /token limit exceeded/i,
+  /request_too_large/i,
 ]
 
-const payloadPatterns = [/request_too_large/i, /request entity too large/i, /payload too large/i, /request too large/i]
+const payloadPatterns = [/request entity too large/i, /payload too large/i, /request too large/i]
 
 const exclusions = [/^(throttling error|service unavailable):/i, /rate limit/i, /too many requests/i]
 
 export const isContextOverflow = (message: string) =>
   !exclusions.some((pattern) => pattern.test(message)) &&
-  (patterns.some((pattern) => pattern.test(message)) || /^400\s*(status code)?\s*\(no body\)/i.test(message))
+  (patterns.some((pattern) => pattern.test(message)) || /^4(?:00|13)\s*(status code)?\s*\(no body\)/i.test(message))
 
 export const isPayloadTooLarge = (message: string) => payloadPatterns.some((pattern) => pattern.test(message))
 
@@ -106,6 +107,7 @@ export function classifyProviderFailure(input: ProviderFailure): AIError["reason
     clientScoped &&
     (codes.includes("context_length_exceeded") ||
       codes.includes("model_context_window_exceeded") ||
+      codes.includes("request_too_large") ||
       isContextOverflow(text))
   )
     return new InvalidRequestReason({ ...common, classification: "context-overflow" })

packages/ai/src/providers/amazon-bedrock-mantle.ts

@@ -1,5 +1,5 @@
 import { Auth } from "../route/auth.js"
-import type { Route as RouteDef, RouteDefaultsInput } from "../route/client.js"
+import { Route, type RouteDefaultsInput } from "../route/client.js"
 import type { ProviderPackage } from "../provider-package.js"
 import { OpenAIChat } from "../protocols/openai-chat.js"
 import { OpenAIResponses } from "../protocols/openai-responses.js"
@@ -26,9 +26,15 @@ export interface Settings extends ProviderPackage.Settings {
   readonly providerOptions?: OpenAIProviderOptionsInput
 }
 
-const responsesRoute = OpenAIResponses.route.with({
+const responsesRoute = Route.make({
   id: "bedrock-mantle-responses",
   provider: id,
+  providerMetadataKey: OpenAIResponses.route.providerMetadataKey,
+  protocol: OpenAIResponses.protocol,
+  endpoint: OpenAIResponses.route.endpoint,
+  auth: OpenAIResponses.route.auth,
+  transport: OpenAIResponses.httpTransport,
+  defaults: OpenAIResponses.route.defaults,
 })
 
 const chatRoute = OpenAIChat.route.with({
@@ -38,7 +44,7 @@ const chatRoute = OpenAIChat.route.with({
 
 export const routes = [responsesRoute, chatRoute]
 
-const configuredRoute = <Body, Prepared>(route: RouteDef<Body, Prepared>, input: Config) => {
+const configuredRoute = <Body, Prepared>(route: Route<Body, Prepared>, input: Config) => {
   const region = input.region ?? input.credentials?.region ?? "us-east-1"
   const credentials = input.credentials === undefined ? undefined : { ...input.credentials, region }
   return route.with({

packages/ai/src/providers/cerebras.ts

@@ -0,0 +1,58 @@
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
+
+export const id = ProviderID.make("cerebras")
+
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly baseURL?: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
+}
+
+export const route = OpenAICompatibleChat.route.with({
+  id: "cerebras-chat",
+  provider: id,
+  endpoint: { baseURL: profiles.cerebras.baseURL },
+})
+
+export const routes = [route]
+
+export const configure = (input: LanguageModelOptions = {}) => {
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
+  const configured = route.with({
+    ...defaults,
+    endpoint: { baseURL: baseURL ?? profiles.cerebras.baseURL },
+    auth: AuthOptions.bearer(input, "CEREBRAS_API_KEY"),
+  })
+  return {
+    id,
+    model: (modelID: string | ModelID) =>
+      configured.model<OpenAIProviderOptionsInput>({
+        id: modelID,
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning", supportsStore: false },
+      }),
+    configure,
+  }
+}
+
+export const provider = configure()
+
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
+  configure({
+    apiKey: settings.apiKey,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    providerOptions: settings.providerOptions,
+  }).model(modelID)

packages/ai/src/providers/deepinfra.ts

@@ -0,0 +1,61 @@
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
+
+export const id = ProviderID.make("deepinfra")
+
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly baseURL?: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
+}
+
+export const route = OpenAICompatibleChat.route.with({
+  id: "deepinfra-chat",
+  provider: id,
+  endpoint: { baseURL: profiles.deepinfra.baseURL },
+})
+
+export const routes = [route]
+
+export const configure = (input: LanguageModelOptions = {}) => {
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
+  const root = baseURL?.replace(/\/+$/, "")
+  const configured = route.with({
+    ...defaults,
+    endpoint: {
+      baseURL: root === undefined ? profiles.deepinfra.baseURL : root.endsWith("/openai") ? root : `${root}/openai`,
+    },
+    auth: AuthOptions.bearer(input, "DEEPINFRA_API_KEY"),
+  })
+  return {
+    id,
+    model: (modelID: string | ModelID) =>
+      configured.model<OpenAIProviderOptionsInput>({
+        id: modelID,
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
+      }),
+    configure,
+  }
+}
+
+export const provider = configure()
+
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
+  configure({
+    apiKey: settings.apiKey,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    providerOptions: settings.providerOptions,
+  }).model(modelID)

packages/ai/src/providers/index.ts

@@ -3,8 +3,10 @@ export * as AnthropicCompatible from "./anthropic-compatible.js"
 export * as AmazonBedrock from "./amazon-bedrock.js"
 export * as AmazonBedrockMantle from "./amazon-bedrock-mantle.js"
 export * as Azure from "./azure.js"
+export * as Cerebras from "./cerebras.js"
 export * as Cloudflare from "./cloudflare.js"
 export { CloudflareAIGateway, CloudflareWorkersAI } from "./cloudflare.js"
+export * as DeepInfra from "./deepinfra.js"
 export * as Google from "./google.js"
 export * as GoogleVertex from "./google-vertex.js"
 export * as GoogleVertexChat from "./google-vertex-chat.js"
@@ -14,5 +16,6 @@ export * as OpenAI from "./openai.js"
 export * as OpenAICompatible from "./openai-compatible.js"
 export * as OpenAICompatibleResponses from "./openai-compatible-responses.js"
 export * as OpenRouter from "./openrouter.js"
+export * as TogetherAI from "./togetherai.js"
 export * as XAI from "./xai.js"
 export * as ZAI from "./zai.js"

packages/ai/src/providers/togetherai.ts

@@ -0,0 +1,58 @@
+import type { ProviderPackage } from "../provider-package.js"
+import { OpenAICompatibleChat } from "../protocols/openai-compatible-chat.js"
+import { AuthOptions, type ProviderAuthOption } from "../route/auth-options.js"
+import type { RouteDefaultsInput } from "../route/client.js"
+import { ProviderID, type ModelID } from "../schema/index.js"
+import { profiles } from "./openai-compatible-profile.js"
+import type { OpenAIProviderOptionsInput } from "./openai-options.js"
+
+export const id = ProviderID.make("togetherai")
+
+export type LanguageModelOptions = Omit<RouteDefaultsInput, "providerOptions"> &
+  ProviderAuthOption<"optional"> & {
+    readonly baseURL?: string
+    readonly providerOptions?: OpenAIProviderOptionsInput
+  }
+
+export interface Settings extends ProviderPackage.Settings {
+  readonly apiKey?: string
+  readonly baseURL?: string
+  readonly providerOptions?: OpenAIProviderOptionsInput
+}
+
+export const route = OpenAICompatibleChat.route.with({
+  id: "togetherai-chat",
+  provider: id,
+  endpoint: { baseURL: profiles.togetherai.baseURL },
+})
+
+export const routes = [route]
+
+export const configure = (input: LanguageModelOptions = {}) => {
+  const { apiKey: _apiKey, auth: _auth, baseURL, ...defaults } = input
+  const configured = route.with({
+    ...defaults,
+    endpoint: { baseURL: baseURL ?? profiles.togetherai.baseURL },
+    auth: AuthOptions.bearer(input, ["TOGETHER_API_KEY", "TOGETHER_AI_API_KEY"]),
+  })
+  return {
+    id,
+    model: (modelID: string | ModelID) =>
+      configured.model<OpenAIProviderOptionsInput>({
+        id: modelID,
+        compatibility: { maxTokensField: "max_tokens", supportsStore: false, supportsStrictMode: false },
+      }),
+    configure,
+  }
+}
+
+export const provider = configure()
+
+export const model: ProviderPackage.Definition<Settings, OpenAIProviderOptionsInput>["model"] = (modelID, settings) =>
+  configure({
+    apiKey: settings.apiKey,
+    baseURL: settings.baseURL,
+    headers: settings.headers === undefined ? undefined : { ...settings.headers },
+    http: settings.body === undefined ? undefined : { body: { ...settings.body } },
+    providerOptions: settings.providerOptions,
+  }).model(modelID)

packages/ai/src/schema/options.ts

@@ -153,6 +153,8 @@ export class LanguageModelCompatibility extends Schema.Class<LanguageModelCompat
 )({
   toolSchema: Schema.optional(LanguageModelToolSchemaCompatibility),
   reasoningField: Schema.optional(Schema.String),
+  /** Require every assistant message to include its reasoning field, even when empty. */
+  requireReasoning: Schema.optional(Schema.Boolean),
   maxTokensField: Schema.optional(LanguageModelMaxTokensFieldCompatibility),
   requireFinishReason: Schema.optional(Schema.Boolean),
   requireAssistantAfterTool: Schema.optional(Schema.Boolean),

packages/ai/test/cache-policy.test.ts

@@ -3,7 +3,7 @@ import { Effect } from "effect"
 import { CacheHint, LLM, Message } from "../src/index.js"
 import { Auth } from "../src/route.js"
 import { compileRequest } from "../src/route/client.js"
-import { AmazonBedrock } from "../src/providers.js"
+import { AmazonBedrock, GoogleVertexMessages } from "../src/providers.js"
 import * as AnthropicMessages from "../src/protocols/anthropic-messages.js"
 import * as Gemini from "../src/protocols/gemini.js"
 import * as OpenAIChat from "../src/protocols/openai-chat.js"
@@ -86,6 +86,27 @@ describe("applyCachePolicy", () => {
     }),
   )
 
+  it.effect("'auto' emits Anthropic cache markers on Vertex", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: GoogleVertexMessages.configure({ accessToken: "test", location: "global", project: "test" }).model(
+            "claude-opus-4-8",
+          ),
+          system: "You are concise.",
+          tools: [{ name: "lookup", description: "Look up a value", inputSchema: { type: "object", properties: {} } }],
+          prompt: "hi",
+        }),
+      )
+
+      expect(prepared.body).toMatchObject({
+        tools: [{ name: "lookup", cache_control: { type: "ephemeral" } }],
+        system: [{ type: "text", text: "You are concise.", cache_control: { type: "ephemeral" } }],
+        messages: [{ role: "user", content: [{ type: "text", text: "hi", cache_control: { type: "ephemeral" } }] }],
+      })
+    }),
+  )
+
   it.effect("'auto' is a no-op on OpenAI (implicit caching protocol)", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(

packages/ai/test/executor.test.ts

@@ -211,6 +211,28 @@ describe("RequestExecutor", () => {
     }).pipe(Effect.provide(responsesLayer([new Response("request too large", { status: 413 })]))),
   )
 
+  it.effect("classifies Anthropic request_too_large as context overflow", () =>
+    Effect.gen(function* () {
+      const executor = yield* RequestExecutor.Service
+      const error = yield* executor.execute(request).pipe(Effect.flip)
+
+      expectAIError(error)
+      expect(error.reason).toMatchObject({
+        _tag: "InvalidRequest",
+        classification: "context-overflow",
+        http: { response: { status: 413 } },
+      })
+    }).pipe(
+      Effect.provide(
+        responsesLayer([
+          new Response('{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}', {
+            status: 413,
+          }),
+        ]),
+      ),
+    ),
+  )
+
   it.effect("does not classify ordinary invalid requests as context overflow", () =>
     Effect.gen(function* () {
       const executor = yield* RequestExecutor.Service

packages/ai/test/fixtures/recordings/cerebras-chat/cerebras-gpt-oss-120b-text.json

@@ -0,0 +1,32 @@
+{
+  "version": 1,
+  "metadata": {
+    "provider": "cerebras",
+    "route": "cerebras-chat",
+    "transport": "http",
+    "model": "gpt-oss-120b",
+    "tags": ["prefix:cerebras-chat", "provider:cerebras", "text", "golden"],
+    "name": "cerebras-chat/cerebras-gpt-oss-120b-text",
+    "recordedAt": "2026-08-25T23:55:27.619Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.cerebras.ai/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-oss-120b\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":256}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"role\":\"assistant\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\"The\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" user says: \\\"Reply exactly with: Hello!\\\" So we must output exactly\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" \\\"Hello!\\\" with no extra characters, no formatting\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\". Ensure no extra spaces or new\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\"lines? Probably just \\\"Hello!\\\".\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"reasoning\":\" Usually we output exactly that.\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{\"content\":\"Hello!\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-cd45cd6e-886a-433e-8ba9-caca78c1f13a\",\"choices\":[{\"delta\":{},\"finish_reason\":\"stop\",\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_e2cabf4999eb0aead3d1\",\"object\":\"chat.completion.chunk\",\"usage\":{\"total_tokens\":142,\"completion_tokens\":58,\"completion_tokens_details\":{\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0,\"reasoning_tokens\":46},\"prompt_tokens\":84,\"prompt_tokens_details\":{\"cached_tokens\":0}},\"time_info\":{\"created\":1787702127.645281,\"queue_time\":0.003817115,\"prompt_time\":0.001587193,\"completion_time\":0.029805929,\"total_time\":0.036823272705078125}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/cerebras-chat/cerebras-gpt-oss-120b-tool-call.json

@@ -0,0 +1,32 @@
+{
+  "version": 1,
+  "metadata": {
+    "provider": "cerebras",
+    "route": "cerebras-chat",
+    "transport": "http",
+    "model": "gpt-oss-120b",
+    "tags": ["prefix:cerebras-chat", "provider:cerebras", "tool", "tool-call", "golden"],
+    "name": "cerebras-chat/cerebras-gpt-oss-120b-tool-call",
+    "recordedAt": "2026-08-25T23:55:28.454Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.cerebras.ai/v1/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"gpt-oss-120b\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":512}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"role\":\"assistant\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\"We\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\" need to call the function get_weather with city \\\"\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"reasoning\":\"Paris\\\".\"},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"tool_calls\":[{\"function\":{\"name\":\"get_weather\",\"arguments\":\"\"},\"type\":\"function\",\"id\":\"3d860cefe\",\"index\":0}]},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{\"tool_calls\":[{\"function\":{\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},\"type\":\"function\",\"index\":0}]},\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\"}\n\ndata: {\"id\":\"chatcmpl-402a09cc-8668-446d-8f9c-3e4121f5fa51\",\"choices\":[{\"delta\":{},\"finish_reason\":\"tool_calls\",\"index\":0}],\"created\":1787702127,\"model\":\"gpt-oss-120b\",\"system_fingerprint\":\"fp_4cfabdd6620dc0120785\",\"object\":\"chat.completion.chunk\",\"usage\":{\"total_tokens\":174,\"completion_tokens\":37,\"completion_tokens_details\":{\"accepted_prediction_tokens\":0,\"rejected_prediction_tokens\":0,\"reasoning_tokens\":13},\"prompt_tokens\":137,\"prompt_tokens_details\":{\"cached_tokens\":0}},\"time_info\":{\"created\":1787702127.8019717,\"queue_time\":0.31196235,\"prompt_time\":0.005234764,\"completion_time\":0.020198402,\"total_time\":0.702225923538208}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-text.json

@@ -0,0 +1,32 @@
+{
+  "version": 1,
+  "metadata": {
+    "provider": "deepinfra",
+    "route": "deepinfra-chat",
+    "transport": "http",
+    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "text", "golden"],
+    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-text",
+    "recordedAt": "2026-08-26T00:34:03.019Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"You are concise.\"},{\"role\":\"user\",\"content\":\"Reply exactly with: Hello!\"}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":40,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Hello\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"!\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RbZ8MyY5pos2MShihSmoXoRe\",\"object\":\"chat.completion.chunk\",\"created\":1787704442,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":25,\"total_tokens\":28,\"completion_tokens\":3,\"estimated_cost\":3.46e-6,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-tool-call.json

@@ -0,0 +1,32 @@
+{
+  "version": 1,
+  "metadata": {
+    "provider": "deepinfra",
+    "route": "deepinfra-chat",
+    "transport": "http",
+    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-call", "golden"],
+    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-call",
+    "recordedAt": "2026-08-26T00:34:04.173Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Call tools exactly as requested.\"},{\"role\":\"user\",\"content\":\"Call get_weather with city exactly Paris.\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"tool_choice\":{\"type\":\"function\",\"function\":{\"name\":\"get_weather\"}},\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_SMfBjXa8eCmHLyjfeyARxe3a\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RFOwhlB2PhZrgMLviGrx5BQf\",\"object\":\"chat.completion.chunk\",\"created\":1787704443,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":230,\"total_tokens\":244,\"completion_tokens\":14,\"estimated_cost\":0.000027480000000000005,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop.json

@@ -0,0 +1,50 @@
+{
+  "version": 1,
+  "metadata": {
+    "provider": "deepinfra",
+    "route": "deepinfra-chat",
+    "transport": "http",
+    "model": "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+    "tags": ["prefix:deepinfra-chat", "provider:deepinfra", "tool", "tool-loop", "golden"],
+    "name": "deepinfra-chat/deepinfra-llama-3-3-70b-tool-loop",
+    "recordedAt": "2026-08-26T00:34:05.656Z"
+  },
+  "interactions": [
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":[{\"index\":0,\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"function\":{\"arguments\":\"{\\\"city\\\": \\\"Paris\\\"}\",\"name\":\"get_weather\"},\"type\":\"function\"}]},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RxlHFSnlLbUUz6XSxqBQj7TC\",\"object\":\"chat.completion.chunk\",\"created\":1787704444,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":243,\"total_tokens\":257,\"completion_tokens\":14,\"estimated_cost\":0.000028780000000000002,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
+      }
+    },
+    {
+      "transport": "http",
+      "request": {
+        "method": "POST",
+        "url": "https://api.deepinfra.com/v1/openai/chat/completions",
+        "headers": {
+          "content-type": "application/json"
+        },
+        "body": "{\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"messages\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":\"What is the weather in Paris?\"},{\"role\":\"assistant\",\"content\":null,\"tool_calls\":[{\"id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"}}],\"reasoning_content\":\"\"},{\"role\":\"tool\",\"tool_call_id\":\"call_W3stxe7FNHozlB4tDxlTVZou\",\"content\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}}],\"stream\":true,\"stream_options\":{\"include_usage\":true},\"max_tokens\":80,\"temperature\":0}"
+      },
+      "response": {
+        "status": 200,
+        "headers": {
+          "content-type": "text/event-stream; charset=utf-8"
+        },
+        "body": "data: {\"service_tier\":null,\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"Paris\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" is\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\" sunny\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\".\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":null}],\"usage\":null}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[{\"index\":0,\"delta\":{\"role\":\"assistant\",\"content\":\"\",\"reasoning_content\":null,\"tool_calls\":null},\"logprobs\":null,\"finish_reason\":\"stop\"}],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":null,\"prompt_tokens_details\":null}}\n\ndata: {\"service_tier\":\"default\",\"id\":\"chatcmpl-RHESVChSFPybgK1eUj2cLLY0\",\"object\":\"chat.completion.chunk\",\"created\":1787704445,\"model\":\"meta-llama/Llama-3.3-70B-Instruct-Turbo\",\"choices\":[],\"usage\":{\"prompt_tokens\":276,\"total_tokens\":281,\"completion_tokens\":5,\"estimated_cost\":0.0000292,\"prompt_tokens_details\":null}}\n\ndata: [DONE]\n\n"
+      }
+    }
+  ]
+}

packages/ai/test/fixtures/recordings/openai-responses-websocket/continues-a-tool-call-over-one-socket.json

@@ -32,7 +32,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Call get_weather once, then reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Call get_weather once, then reply exactly: Paris is sunny.\"}"
         },
         {
           "direction": "server",
@@ -62,7 +62,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"type\":\"function_call_output\",\"call_id\":\"call_ws_weather\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"previous_response_id\":\"resp_ws_tool_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"type\":\"function_call_output\",\"call_id\":\"call_ws_weather\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"max_output_tokens\":50,\"previous_response_id\":\"resp_ws_tool_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Call get_weather once, then reply exactly: Paris is sunny.\"}"
         },
         {
           "direction": "server",

packages/ai/test/fixtures/recordings/openai-responses-websocket/reconstructs-full-context-after-reconnect.json

@@ -32,7 +32,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
         },
         {
           "direction": "server",
@@ -81,7 +81,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_reconnect_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Alpha.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Beta.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Alpha.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_reconnect_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Alpha.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Beta.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
         },
         {
           "direction": "server",

packages/ai/test/fixtures/recordings/openai-responses-websocket/recovers-from-explicit-continuation-rejection.json

@@ -32,7 +32,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
         },
         {
           "direction": "server",
@@ -81,7 +81,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"previous_response_id\":\"resp_ws_rejection_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"previous_response_id\":\"resp_ws_rejection_1\",\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
         },
         {
           "direction": "server",
@@ -91,7 +91,7 @@
         {
           "direction": "client",
           "kind": "text",
-          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Follow the user's exact reply instruction.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_rejection_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Ready.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"}}"
+          "body": "{\"type\":\"response.create\",\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Ready.\"}]},{\"type\":\"message\",\"id\":\"msg_ws_rejection_1\",\"role\":\"assistant\",\"content\":[{\"type\":\"output_text\",\"text\":\"Ready.\"}]},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Reply exactly: Recovered.\"}]}],\"store\":false,\"max_output_tokens\":30,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"instructions\":\"Follow the user's exact reply instruction.\"}"
         },
         {
           "direction": "server",

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-reasoning-continuation.json

@@ -26,7 +26,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true,\"instructions\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-reasoning.json

@@ -18,7 +18,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Think briefly, then reply exactly with: Hello!\"}]}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"low\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":120,\"stream\":true,\"instructions\":\"Show concise reasoning when the provider supports visible reasoning summaries.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/fixtures/recordings/openai-responses/openai-responses-gpt-5-5-tool-loop.json

@@ -25,7 +25,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
       },
       "response": {
         "status": 200,
@@ -43,7 +43,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"system\",\"content\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true}"
+        "body": "{\"model\":\"gpt-5.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"What is the weather in Paris?\"}]},{\"type\":\"reasoning\",\"id\":\"rs_0ad67c31d9ddad95016a869efbd02487d1a51eed850e6f87f5\",\"summary\":[],\"encrypted_content\":\"gAAAAABqhp79U9UPKmTdmo9tmdil0C2KXpkFqUc4MNkYHT53Lzos9omncFPg76QzUmmSOdBcajisWBEo-xiTCvhp135uACUq8TJcdw4DluieYq6dWszijy28PFFfeO-6MmHwi7zeln1Z202zErJUEyuf1bML68VAeam5PqlMLG-a4-pmnWiH2ExWKibTUX37QoMQoArrkccJOCmxwDflV_kWDPMFxQVDfeMg9fd1gVv2u-x1Mjk0b9mJDOq0Fe5Gh-IkpWzfXgZTdptFmCM75cksvs61Rqsx6P33czal-LSixEF0WMizCvbMQmqKGs7MKGMeoa6j6vWOnB3ICIbv6FShnSaTpZWJFwejvOurkfuxa-2q6xVDZsBoQCgMWPHsqLxwAo1JKdfBk0pMvSuvpw2BRxykUZ1ULCYJ-BypST65292-EuSZFIuXPMPir-_raSCTsgsZNMscDG6ll3qksDTDS6_o5NutD7Ra-WZzaUe_HQlSLKLACTc4qv2EK1QoC4aYv4goxkTSx17WhS2D86lILgkUd-TIHjJ6iR3uxSNx7YeBNxiJgddIAEjAaSrdF-WDouSNT9k3efd5HhT3zahIOMKgb3XIQzFOYWfWgea5-SbaIdKwne9hU0QyhcBQs6yoifSg-fJZtahbPb-GCDYnOLlH-bV94vldoccb-2P1JdB3jaLj5tJUecfr2H4qiu8MgkPj0TkwYNbJynYmJo9H5Lm-XJ9gfzIXzJh0arKwsS4gwDLf4J3LOEF3WEW3mknOjjb9PrLmHRYXQQh9tTiX9ILPZpbufkyCurTUMQgWiSCitXBC6FoLXRHilSmb-6_avBnlUMziMfey-FkKvRfiPox6BaJrnOq6SGlOv11y7EKvzrn29la7HKPygYenDAkyq2mq0Zk2nLWNmJcv9sQTBrkBdFMmJYPi2J2im8XD5MmAjEL8R4FCBHoPIIZ6pENQykvH8PhpWKuzF5gJlY3Vwz4iJ0Qb9TrNI0hzBoI1U0LeB5FJ2HgjZQwCFF5x3ubh72xrUsFpuyYyYPa8GDT0Bo-LW_IlJ_mN4EwI5Nk9n-8Bt015yxsfpa5YaDeCeQFcdj8SD0UAd7QWtGACpzKcIj1-vJJU7OiwscV_v1dLvoiEe1ehI9jcvPn28TgHlo_dippe0iMN4FAm1Bf8vtWVMFDvfV1rPv1pAFFnSa9XqFszD5Exo_xzcQEoKXvQv3OnUtoiM4Db4uadClazLjoep2TQgHcJBVbTbLySTVPmok4ROFQZsU_mq4vu2M__d8HOjADfIIYz5VLVQKNpo0Hv_QkT2bn56Q==\"},{\"type\":\"function_call\",\"id\":\"fc_0ad67c31d9ddad95016a869efd126887d1b7e2f17f155cb5dc\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"name\":\"get_weather\",\"arguments\":\"{\\\"city\\\":\\\"Paris\\\"}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_qrzOfKDfzaq8fqbSNNVHlNsV\",\"output\":\"{\\\"temperature\\\":22,\\\"condition\\\":\\\"sunny\\\"}\"}],\"tools\":[{\"type\":\"function\",\"name\":\"get_weather\",\"description\":\"Get current weather for a city.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"city\":{\"type\":\"string\"}},\"required\":[\"city\"],\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"reasoning\":{\"effort\":\"medium\",\"summary\":\"auto\"},\"text\":{\"verbosity\":\"low\"},\"max_output_tokens\":80,\"stream\":true,\"instructions\":\"Use the get_weather tool exactly once. After the tool result, reply exactly: Paris is sunny.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/fixtures/recordings/pdf/openai-tool-result.json

@@ -21,7 +21,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"gpt-4o-mini\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true,\"instructions\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/fixtures/recordings/pdf/xai-tool-result.json

@@ -21,7 +21,7 @@
         "headers": {
           "content-type": "application/json"
         },
-        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"system\",\"content\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"},{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true}"
+        "body": "{\"model\":\"grok-4.5\",\"input\":[{\"role\":\"user\",\"content\":[{\"type\":\"input_text\",\"text\":\"Return only the verification code from the PDF.\"}]},{\"type\":\"function_call\",\"call_id\":\"call_pdf_1\",\"name\":\"read_pdf\",\"arguments\":\"{}\"},{\"type\":\"function_call_output\",\"call_id\":\"call_pdf_1\",\"output\":[{\"type\":\"input_text\",\"text\":\"PDF read successfully\"},{\"type\":\"input_file\",\"filename\":\"verification.pdf\",\"file_data\":\"data:application/pdf;base64,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\"}]}],\"tools\":[{\"type\":\"function\",\"name\":\"read_pdf\",\"description\":\"Read the attached PDF.\",\"parameters\":{\"type\":\"object\",\"properties\":{},\"additionalProperties\":false},\"strict\":false}],\"store\":false,\"include\":[\"reasoning.encrypted_content\"],\"max_output_tokens\":40,\"temperature\":0,\"stream\":true,\"instructions\":\"Read the PDF returned by the tool and follow the user's response format exactly.\"}"
       },
       "response": {
         "status": 200,

packages/ai/test/provider-error.test.ts

@@ -18,13 +18,19 @@ describe("provider error classification", () => {
     expect(messages.every(isContextOverflow)).toBe(true)
   })
 
-  test("classifies request size failures separately from context overflow", () => {
-    const failures = [
-      classifyProviderFailure({ message: "request too large", status: 413 }),
+  test("classifies Anthropic request_too_large as recoverable overflow", () => {
+    expect(
       classifyProviderFailure({
         message: '{"error":{"type":"request_too_large","message":"Request exceeds the maximum size"}}',
         status: 400,
       }),
+    ).toMatchObject({ _tag: "InvalidRequest", classification: "context-overflow" })
+    expect(isContextOverflow("413 status code (no body)")).toBe(true)
+  })
+
+  test("classifies generic request size failures separately from context overflow", () => {
+    const failures = [
+      classifyProviderFailure({ message: "request too large", status: 413 }),
       classifyProviderFailure({ message: "upstream request entity too large", status: 502 }),
     ]
 
@@ -33,7 +39,6 @@ describe("provider error classification", () => {
         expect.objectContaining({ _tag: "InvalidRequest", classification: "payload-too-large" }),
       ),
     )
-    expect(isContextOverflow("413 status code (no body)")).toBe(false)
   })
 
   test("does not classify rate limits as context overflow", () => {

packages/ai/test/provider-package.test.ts

@@ -26,6 +26,9 @@ describe("provider package entrypoints", () => {
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle"),
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle/chat"),
       import("@opencode-ai/ai/providers/amazon-bedrock/mantle/responses"),
+      import("@opencode-ai/ai/providers/togetherai"),
+      import("@opencode-ai/ai/providers/cerebras"),
+      import("@opencode-ai/ai/providers/deepinfra"),
     ])
 
     for (const module of modules) expect(module.model).toBeFunction()
@@ -35,6 +38,24 @@ describe("provider package entrypoints", () => {
     expect(modules[19].model).toBe(modules[20].model)
   })
 
+  test("maps DeepInfra package settings onto its native executable model", async () => {
+    const DeepInfra = await import("@opencode-ai/ai/providers/deepinfra")
+    const settings = {
+      apiKey: "fixture",
+      baseURL: "https://provider.example.test/v1/",
+      headers: { "x-application": "opencode" },
+      body: { service_tier: "priority" },
+      providerOptions: { reasoningEffort: "high" as const },
+    }
+    const deepinfra = DeepInfra.model("google/gemma-3-27b-it", settings)
+
+    expect(deepinfra.route.id).toBe("deepinfra-chat")
+    expect(deepinfra.route.endpoint.baseURL).toBe("https://provider.example.test/v1/openai")
+    expect(deepinfra.route.defaults.providerOptions).toEqual(settings.providerOptions)
+    expect(deepinfra.route.defaults.headers).toEqual(settings.headers)
+    expect(deepinfra.route.defaults.http?.body).toEqual(settings.body)
+  })
+
   test("maps OpenRouter and xAI package settings onto executable models", async () => {
     const OpenRouter = await import("@opencode-ai/ai/providers/openrouter")
     const XAI = await import("@opencode-ai/ai/providers/xai")

packages/ai/test/provider/anthropic-messages.test.ts

@@ -5,6 +5,7 @@ import { CacheHint, LLM, AIError, LLMRequest, Message, ToolCallPart, ToolDefinit
 import { Auth, LLMClient } from "../../src/route.js"
 import { compileRequest } from "../../src/route/client.js"
 import * as AnthropicMessages from "../../src/protocols/anthropic-messages.js"
+import { GoogleVertexMessages } from "../../src/providers.js"
 import { continuationRequest, nativeAnthropicMessagesContinuation } from "../continuation-scenarios.js"
 import { it } from "../lib/effect.js"
 import { dynamicResponse, fixedResponse } from "../lib/http.js"
@@ -27,6 +28,12 @@ const compileUnsignedReasoning = (model: LLMRequest["model"]) =>
     }),
   )
 
+const vertexOpus48 = GoogleVertexMessages.configure({
+  accessToken: "test",
+  location: "global",
+  project: "test",
+}).model("claude-opus-4-8")
+
 const request = LLM.request({
   id: "req_1",
   model,
@@ -286,6 +293,149 @@ describe("Anthropic Messages route", () => {
     }),
   )
 
+  it.effect("keeps a terminal Vertex system update in the tool-result turn", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+            {
+              type: "text",
+              text: "<system-update>\nOperator update.\n</system-update>",
+              cache_control: undefined,
+            },
+          ],
+        },
+      ])
+    }),
+  )
+
+  it.effect("preserves folded tool-result system updates across multi-turn Vertex history", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+            Message.assistant("Acknowledged."),
+            Message.user("Next step."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+            {
+              type: "text",
+              text: "<system-update>\nOperator update.\n</system-update>",
+              cache_control: undefined,
+            },
+          ],
+        },
+        { role: "assistant", content: [{ type: "text", text: "Acknowledged." }] },
+        { role: "user", content: [{ type: "text", text: "Next step." }] },
+      ])
+    }),
+  )
+
+  it.effect("keeps a terminal direct Anthropic system update native", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: opus48,
+          messages: [
+            Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+            Message.tool({ id: "call_1", name: "lookup", result: "Done." }),
+            Message.system("Operator update."),
+          ],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "tool_use", id: "call_1", name: "lookup", input: {} }],
+        },
+        {
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: "call_1",
+              content: '"Done."',
+              is_error: undefined,
+              cache_control: undefined,
+            },
+          ],
+        },
+        {
+          role: "system",
+          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
+        },
+      ])
+    }),
+  )
+
+  it.effect("keeps an ordinary terminal Vertex system update native", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: vertexOpus48,
+          messages: [Message.user("Before."), Message.system("Operator update.")],
+          cache: "none",
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "user", content: [{ type: "text", text: "Before." }] },
+        {
+          role: "system",
+          content: [{ type: "text", text: "Operator update.", cache_control: undefined }],
+        },
+      ])
+    }),
+  )
+
   it.effect("rejects a system update between a local tool call and its result", () =>
     Effect.gen(function* () {
       const error = yield* compileRequest(

packages/ai/test/provider/bedrock-mantle.test.ts

@@ -1,11 +1,12 @@
 import { describe, expect } from "bun:test"
 import { Effect } from "effect"
 import { HttpClientRequest } from "effect/unstable/http"
-import { LLM } from "../../src/index.js"
+import { LLM, Message } from "../../src/index.js"
 import { AmazonBedrockMantle } from "../../src/providers.js"
+import { OpenAIResponses } from "../../src/protocols/openai-responses.js"
 import { compileRequest, LLMClient } from "../../src/route/client.js"
 import { it } from "../lib/effect.js"
-import { dynamicResponse } from "../lib/http.js"
+import { dynamicResponse, fixedResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"
 import { recordedTests } from "../recorded-test.js"
 
@@ -19,6 +20,7 @@ describe("Amazon Bedrock Mantle provider", () => {
   it.effect("uses Chat by default and exposes Responses", () =>
     Effect.gen(function* () {
       const provider = AmazonBedrockMantle.configure({ credentials })
+      expect(provider.responses("openai.gpt-oss-120b").route.transport).toBe(OpenAIResponses.httpTransport)
       const chat = yield* compileRequest(LLM.request({ model: provider.model("openai.gpt-oss-120b"), prompt: "Hi" }))
       const responses = yield* compileRequest(
         LLM.request({ model: provider.responses("openai.gpt-oss-120b"), prompt: "Hi" }),
@@ -83,6 +85,39 @@ describe("Amazon Bedrock Mantle provider", () => {
       expect(seen).toEqual([{ url: "https://mantle.test/v1/chat/completions", authorization: "Bearer test-key" }])
     }),
   )
+
+  it.effect("replays reasoning with Mantle's message-prefixed item ids", () =>
+    Effect.gen(function* () {
+      const model = AmazonBedrockMantle.configure({ apiKey: "test-key" }).responses("openai.gpt-oss-120b")
+      const item = { type: "reasoning", id: "msg_95d4d0af4350432a", encrypted_content: "mantle-state" }
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", item },
+              { type: "response.reasoning_summary_text.delta", item_id: item.id, delta: "Considering." },
+              { type: "response.output_item.done", item },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      const prepared = yield* compileRequest(
+        LLM.request({ model, messages: [response.message, Message.user("Continue.")] }),
+      )
+
+      expect(prepared.body.input).toEqual([
+        {
+          type: "reasoning",
+          id: "msg_95d4d0af4350432a",
+          summary: [{ type: "summary_text", text: "Considering." }],
+          encrypted_content: "mantle-state",
+        },
+        { role: "user", content: [{ type: "input_text", text: "Continue." }] },
+      ])
+    }),
+  )
 })
 
 const recorded = recordedTests({

packages/ai/test/provider/golden.recorded.test.ts

@@ -1,5 +1,6 @@
 import * as Anthropic from "../../src/providers/anthropic.js"
 import * as AnthropicCompatible from "../../src/providers/anthropic-compatible.js"
+import { Cerebras, DeepInfra, TogetherAI } from "../../src/providers/index.js"
 import { CloudflareAIGateway, CloudflareWorkersAI } from "../../src/providers/cloudflare.js"
 import * as Google from "../../src/providers/google.js"
 import * as OpenAI from "../../src/providers/openai.js"
@@ -47,14 +48,16 @@ const cloudflareWorkersAITools = cloudflareWorkers.model("@cf/openai/gpt-oss-20b
 const deepseek = OpenAICompatible.deepseek
   .configure({ apiKey: process.env.DEEPSEEK_API_KEY ?? "fixture" })
   .model("deepseek-chat")
-const together = OpenAICompatible.togetherai
-  .configure({
-    apiKey: process.env.TOGETHER_AI_API_KEY ?? "fixture",
-  })
-  .model("meta-llama/Llama-3.3-70B-Instruct-Turbo")
+const together = TogetherAI.configure({
+  apiKey: process.env.TOGETHER_API_KEY ?? process.env.TOGETHER_AI_API_KEY ?? "fixture",
+}).model("meta-llama/Llama-3.3-70B-Instruct-Turbo")
+const cerebras = Cerebras.configure({ apiKey: process.env.CEREBRAS_API_KEY ?? "fixture" }).model("gpt-oss-120b")
 const groq = OpenAICompatible.groq
   .configure({ apiKey: process.env.GROQ_API_KEY ?? "fixture" })
   .model("llama-3.3-70b-versatile")
+const deepInfra = DeepInfra.configure({ apiKey: process.env.DEEPINFRA_API_KEY ?? "fixture" }).model(
+  "meta-llama/Llama-3.3-70B-Instruct-Turbo",
+)
 const openRouter = OpenRouter.configure({ apiKey: process.env.OPENROUTER_API_KEY ?? "fixture" })
 const openrouter = openRouter.model("openai/gpt-4o-mini")
 const openrouterGpt55 = openRouter.model("openai/gpt-5.5")
@@ -193,8 +196,27 @@ describeRecordedGoldenScenarios([
     name: "TogetherAI Llama 3.3 70B",
     prefix: "openai-compatible-chat",
     model: together,
-    requires: ["TOGETHER_AI_API_KEY"],
-    scenarios: ["text", "tool-call"],
+    requires: ["TOGETHER_API_KEY"],
+    scenarios: [
+      {
+        id: "text",
+        cassette: "openai-compatible-chat/togetherai-streams-text",
+        prompt: "Reply with exactly: Hello!",
+        maxTokens: 20,
+      },
+      { id: "tool-call", cassette: "openai-compatible-chat/togetherai-streams-tool-call" },
+    ],
+  },
+  {
+    name: "Cerebras GPT OSS 120B",
+    prefix: "cerebras-chat",
+    model: cerebras,
+    requires: ["CEREBRAS_API_KEY"],
+    scenarios: [
+      { id: "text", maxTokens: 256, temperature: false },
+      { id: "tool-call", maxTokens: 512, temperature: false },
+      { id: "tool-loop", maxTokens: 512, temperature: false, timeout: 30_000 },
+    ],
   },
   {
     name: "Groq Llama 3.3 70B",
@@ -203,6 +225,13 @@ describeRecordedGoldenScenarios([
     requires: ["GROQ_API_KEY"],
     scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
   },
+  {
+    name: "DeepInfra Llama 3.3 70B",
+    prefix: "deepinfra-chat",
+    model: deepInfra,
+    requires: ["DEEPINFRA_API_KEY"],
+    scenarios: ["text", "tool-call", { id: "tool-loop", timeout: 30_000 }],
+  },
   {
     name: "OpenRouter gpt-4o-mini",
     prefix: "openai-compatible-chat",

packages/ai/test/provider/native-providers.test.ts

@@ -0,0 +1,184 @@
+import { describe, expect } from "bun:test"
+import { ConfigProvider, Effect } from "effect"
+import { HttpClientRequest } from "effect/unstable/http"
+import { LLM, Message, ToolDefinition } from "../../src/index.js"
+import { Cerebras, DeepInfra, TogetherAI } from "../../src/providers/index.js"
+import { compileRequest } from "../../src/route/client.js"
+import { it } from "../lib/effect.js"
+import { dynamicResponse } from "../lib/http.js"
+import { sseEvents } from "../lib/sse.js"
+
+describe("native OpenAI-compatible providers", () => {
+  it.effect("preserves native Together AI and Cerebras provider and route identities", () =>
+    Effect.gen(function* () {
+      const together = TogetherAI.configure({ apiKey: "fixture" }).model("meta-llama/Llama-3.3-70B")
+      const cerebras = Cerebras.configure({ apiKey: "fixture" }).model("qwen-3-235b-a22b")
+
+      expect(together).toMatchObject({
+        provider: "togetherai",
+        compatibility: { maxTokensField: "max_tokens", supportsStore: false, supportsStrictMode: false },
+        route: { id: "togetherai-chat", protocol: "openai-chat" },
+      })
+      expect(together.route.endpoint.baseURL).toBe("https://api.together.xyz/v1")
+      expect(cerebras).toMatchObject({
+        provider: "cerebras",
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning", supportsStore: false },
+        route: { id: "cerebras-chat", protocol: "openai-chat" },
+      })
+      expect(cerebras.route.endpoint.baseURL).toBe("https://api.cerebras.ai/v1")
+    }),
+  )
+
+  it.effect("preserves native DeepInfra provider and route identity", () =>
+    Effect.gen(function* () {
+      const deepinfra = DeepInfra.configure({ apiKey: "fixture" }).model("google/gemma-3-27b-it")
+      expect(deepinfra).toMatchObject({
+        provider: "deepinfra",
+        compatibility: { maxTokensField: "max_tokens", reasoningField: "reasoning_content", supportsStore: false },
+        route: { id: "deepinfra-chat", protocol: "openai-chat" },
+      })
+      expect(deepinfra.route.endpoint.baseURL).toBe("https://api.deepinfra.com/v1/openai")
+    }),
+  )
+
+  it.effect("applies native provider request defaults even with a custom gateway URL", () =>
+    Effect.gen(function* () {
+      const together = yield* compileRequest(
+        LLM.request({
+          model: TogetherAI.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("llama"),
+          prompt: "Use a tool.",
+          generation: { maxTokens: 32 },
+          tools: [
+            ToolDefinition.make({ name: "lookup", description: "Look up data", inputSchema: { type: "object" } }),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(together.body).toMatchObject({
+        max_tokens: 32,
+        stream_options: { include_usage: true },
+        tools: [{ function: { name: "lookup" } }],
+      })
+      expect(together.body).not.toHaveProperty("max_completion_tokens")
+      expect(together.body).not.toHaveProperty("store")
+      expect(together.body.tools?.[0]?.function).not.toHaveProperty("strict")
+
+      const cerebras = yield* compileRequest(
+        LLM.request({
+          model: Cerebras.configure({ apiKey: "fixture", baseURL: "https://gateway.example/v1" }).model("qwen"),
+          generation: { maxTokens: 48 },
+          messages: [
+            Message.user("Think first."),
+            Message.assistant([
+              { type: "reasoning", text: "A deliberate thought." },
+              { type: "text", text: "An answer." },
+            ]),
+            Message.user("Continue."),
+          ],
+          providerOptions: { store: true },
+        }),
+      )
+
+      expect(cerebras.body).toMatchObject({
+        max_tokens: 48,
+        messages: [
+          { role: "user", content: "Think first." },
+          { role: "assistant", content: "An answer.", reasoning: "A deliberate thought." },
+          { role: "user", content: "Continue." },
+        ],
+      })
+      expect(cerebras.body).not.toHaveProperty("max_completion_tokens")
+      expect(cerebras.body).not.toHaveProperty("store")
+      expect(cerebras.body.messages[1]).not.toHaveProperty("reasoning_content")
+    }),
+  )
+
+  it.effect("normalizes DeepInfra API roots without duplicating the OpenAI path", () =>
+    Effect.gen(function* () {
+      for (const baseURL of [
+        "https://gateway.example/v1",
+        "https://gateway.example/v1/",
+        "https://gateway.example/v1/openai",
+        "https://gateway.example/v1/openai/",
+      ]) {
+        expect(DeepInfra.configure({ apiKey: "fixture", baseURL }).model("gemma").route.endpoint.baseURL).toBe(
+          "https://gateway.example/v1/openai",
+        )
+      }
+    }),
+  )
+
+  it.effect("maps package settings onto native executable models", () =>
+    Effect.gen(function* () {
+      for (const native of [TogetherAI, Cerebras]) {
+        const selected = native.model("provider-model", {
+          apiKey: "fixture",
+          baseURL: "https://gateway.example/v1",
+          headers: { "x-application": "opencode" },
+          body: { service_tier: "priority" },
+          providerOptions: { reasoningEffort: "high" },
+        })
+
+        expect(selected.route.endpoint.baseURL).toBe("https://gateway.example/v1")
+        expect(selected.route.defaults.headers).toEqual({ "x-application": "opencode" })
+        expect(selected.route.defaults.http?.body).toEqual({ service_tier: "priority" })
+        expect(selected.route.defaults.providerOptions).toEqual({ reasoningEffort: "high" })
+      }
+    }),
+  )
+
+  it.effect("resolves provider environment credentials and preserves deprecated Together credentials", () =>
+    Effect.gen(function* () {
+      const scenarios = [
+        {
+          model: TogetherAI.configure().model("llama"),
+          env: { TOGETHER_API_KEY: "together-primary", TOGETHER_AI_API_KEY: "together-legacy" },
+          token: "together-primary",
+          url: "https://api.together.xyz/v1/chat/completions",
+        },
+        {
+          model: TogetherAI.configure().model("llama"),
+          env: { TOGETHER_AI_API_KEY: "together-legacy" },
+          token: "together-legacy",
+          url: "https://api.together.xyz/v1/chat/completions",
+        },
+        {
+          model: Cerebras.configure().model("qwen"),
+          env: { CEREBRAS_API_KEY: "cerebras-secret" },
+          token: "cerebras-secret",
+          url: "https://api.cerebras.ai/v1/chat/completions",
+        },
+        {
+          model: DeepInfra.configure().model("gemma"),
+          env: { DEEPINFRA_API_KEY: "deepinfra-secret" },
+          token: "deepinfra-secret",
+          url: "https://api.deepinfra.com/v1/openai/chat/completions",
+        },
+      ]
+
+      yield* Effect.forEach(scenarios, (scenario) =>
+        LLM.generate(LLM.request({ model: scenario.model, prompt: "Say hello." })).pipe(
+          Effect.provide(
+            dynamicResponse((input) =>
+              Effect.gen(function* () {
+                const request = yield* HttpClientRequest.toWeb(input.request).pipe(Effect.orDie)
+                expect(request.url).toBe(scenario.url)
+                expect(request.headers.get("authorization")).toBe(`Bearer ${scenario.token}`)
+                return input.respond(
+                  sseEvents(
+                    { id: "chatcmpl_fixture", choices: [{ delta: { content: "Hello" }, finish_reason: null }] },
+                    { id: "chatcmpl_fixture", choices: [{ delta: {}, finish_reason: "stop" }] },
+                  ),
+                  { headers: { "content-type": "text/event-stream" } },
+                )
+              }),
+            ),
+          ),
+          Effect.provide(ConfigProvider.layer(ConfigProvider.fromEnv({ env: scenario.env }))),
+          Effect.tap((response) => Effect.sync(() => expect(response.text).toBe("Hello"))),
+        ),
+      )
+    }),
+  )
+})

packages/ai/test/provider/openai-chat.test.ts

@@ -169,6 +169,56 @@ describe("OpenAI Chat route", () => {
     }),
   )
 
+  it.effect("preserves observed reasoning fields when reasoning is required", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: LanguageModel.update(model, { compatibility: { requireReasoning: true } }),
+          messages: [
+            Message.assistant([
+              {
+                type: "reasoning",
+                text: "thinking",
+                providerMetadata: { openai: { reasoningField: "reasoning_text" } },
+              },
+              { type: "text", text: "Hello" },
+            ]),
+            Message.assistant("Done"),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
+        { role: "assistant", content: "Done", reasoning_content: "" },
+      ])
+    }),
+  )
+
+  it.effect("omits empty configured reasoning fields when reasoning is explicitly optional", () =>
+    Effect.gen(function* () {
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model: LanguageModel.update(model, {
+            compatibility: { reasoningField: "reasoning_text", requireReasoning: false },
+          }),
+          messages: [
+            Message.assistant([
+              { type: "reasoning", text: "thinking" },
+              { type: "text", text: "Hello" },
+            ]),
+            Message.assistant("Done"),
+          ],
+        }),
+      )
+
+      expect(prepared.body.messages).toEqual([
+        { role: "assistant", content: "Hello", reasoning_text: "thinking" },
+        { role: "assistant", content: "Done" },
+      ])
+    }),
+  )
+
   it.effect("rejects reasoning fields that conflict with assistant message fields", () =>
     Effect.gen(function* () {
       const error = yield* compileRequest(

packages/ai/test/provider/openai-compatible-chat.test.ts

@@ -317,6 +317,56 @@ describe("OpenAI-compatible Chat route", () => {
     }),
   )
 
+  it.effect("requires reasoning for DeepSeek models, providers, and endpoints unless explicitly overridden", () =>
+    Effect.gen(function* () {
+      const cases = [
+        { id: "DeepSeek-V3", provider: "custom", baseURL: "https://api.custom.test/v1", required: true },
+        { id: "custom-model", provider: "deepseek", baseURL: "https://api.custom.test/v1", required: true },
+        { id: "custom-model", provider: "custom", baseURL: "https://API.DeepSeek.COM/v1", required: true },
+        { id: "ordinary-model", provider: "custom", baseURL: "https://api.custom.test/v1", required: false },
+        {
+          id: "ordinary-model",
+          provider: "custom",
+          baseURL: "https://api.custom.test/v1",
+          compatibility: { requireReasoning: true, reasoningField: "reasoning" },
+          required: true,
+          field: "reasoning",
+        },
+        {
+          id: "deepseek-chat",
+          provider: "deepseek",
+          baseURL: "https://api.deepseek.com/v1",
+          compatibility: { requireReasoning: false },
+          required: false,
+        },
+      ] as const
+
+      yield* Effect.forEach(cases, (item) =>
+        Effect.gen(function* () {
+          const selected = OpenAICompatibleChat.route
+            .with({ provider: item.provider, endpoint: { baseURL: item.baseURL } })
+            .model({ id: item.id, compatibility: "compatibility" in item ? item.compatibility : undefined })
+          const prepared = yield* compileRequest(
+            LLM.request({
+              model: selected,
+              messages: [
+                Message.assistant("Hello"),
+                Message.assistant([ToolCallPart.make({ id: "call_1", name: "lookup", input: {} })]),
+                Message.tool({ id: "call_1", name: "lookup", result: "Sunny" }),
+              ],
+            }),
+          )
+          const field = "field" in item ? item.field : "reasoning_content"
+
+          for (const message of prepared.body.messages.filter((message) => message.role === "assistant")) {
+            if (item.required) expect(message).toHaveProperty(field, "")
+            else expect(message).not.toHaveProperty(field)
+          }
+        }),
+      )
+    }),
+  )
+
   it.effect("posts to the configured compatible endpoint and parses text usage", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(

packages/ai/test/provider/openai-compatible-responses.test.ts

@@ -47,10 +47,8 @@ describe("Open Responses-compatible route", () => {
       })
       expect(prepared.body).toEqual({
         model: "example-model",
-        input: [
-          { role: "system", content: "You are concise." },
-          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
-        ],
+        input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
+        instructions: "You are concise.",
         stream: true,
         store: false,
         include: ["reasoning.encrypted_content"],
@@ -84,10 +82,12 @@ describe("Open Responses-compatible route", () => {
       const prepared = yield* compileRequest(
         LLM.request({
           model,
+          system: "Initial instructions.",
           messages: [Message.user("Before."), Message.system("Operator update."), Message.assistant("After.")],
         }),
       )
 
+      expect(prepared.body.instructions).toBe("Initial instructions.")
       expect(prepared.body.input).toEqual([
         { role: "user", content: [{ type: "input_text", text: "Before." }] },
         { role: "developer", content: "Operator update." },
@@ -195,15 +195,19 @@ describe("Open Responses-compatible route", () => {
           model,
           messages: [
             Message.assistant([
-              // The baseline does not enforce a provider id grammar, so a
-              // non-OpenAI but well-formed token is resent as-is.
               { type: "text", text: "Kept.", providerMetadata: { openresponses: { itemId: "history_1" } } },
-              // Shape violations are dropped even without a grammar policy.
               {
                 type: "text",
-                text: "Dropped.",
-                providerMetadata: { openresponses: { itemId: `m${"a".repeat(64)}` } },
+                text: "Long.",
+                providerMetadata: { openresponses: { itemId: `history_${"a".repeat(64)}` } },
               },
+              {
+                type: "text",
+                text: "Opaque.",
+                providerMetadata: { openresponses: { itemId: "provider_value/with+symbols" } },
+              },
+              { type: "text", text: "No suffix.", providerMetadata: { openresponses: { itemId: "msg_" } } },
+              { type: "text", text: "No prefix.", providerMetadata: { openresponses: { itemId: "_item" } } },
             ]),
           ],
         }),
@@ -218,8 +222,23 @@ describe("Open Responses-compatible route", () => {
         },
         {
           type: "message",
+          id: `history_${"a".repeat(64)}`,
           role: "assistant",
-          content: [{ type: "output_text", text: "Dropped." }],
+          content: [{ type: "output_text", text: "Long." }],
+        },
+        {
+          type: "message",
+          id: "provider_value/with+symbols",
+          role: "assistant",
+          content: [{ type: "output_text", text: "Opaque." }],
+        },
+        {
+          type: "message",
+          role: "assistant",
+          content: [
+            { type: "output_text", text: "No suffix." },
+            { type: "output_text", text: "No prefix." },
+          ],
         },
       ])
     }),
@@ -288,6 +307,43 @@ describe("Open Responses-compatible route", () => {
     }),
   )
 
+  it.effect("streams function calls without optional item ids through the shared baseline", () =>
+    Effect.gen(function* () {
+      const model = configure({
+        apiKey: "test-key",
+        baseURL: "https://responses.example.test/v1",
+        provider: "example",
+      }).model("example-model")
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Look it up." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 1, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 1,
+                item_id: "opaque_item",
+                delta: '{"query":"shared"}',
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 1,
+                item: { ...item, arguments: '{"query":"complete"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "complete" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+    }),
+  )
+
   it.effect("finalizes pending function calls from completed response output", () =>
     Effect.gen(function* () {
       const model = configure({

packages/ai/test/provider/openai-responses-websocket.recorded.test.ts

@@ -136,6 +136,7 @@ describe("OpenAI Responses WebSocket recorded", () => {
       expect(channel.opens()).toBe(1)
       expect(channel.sent).toHaveLength(2)
       expect(channel.sent[1]).toMatchObject({
+        instructions: "Call get_weather once, then reply exactly: Paris is sunny.",
         previous_response_id: expect.any(String),
         input: [{ type: "function_call_output", call_id: call.id, output: expect.any(String) }],
       })
@@ -167,8 +168,8 @@ describe("OpenAI Responses WebSocket recorded", () => {
       expect(channel.opens()).toBe(2)
       expect(channel.sent[1]).not.toHaveProperty("previous_response_id")
       expect(channel.sent[1]).toMatchObject({
+        instructions: "Follow the user's exact reply instruction.",
         input: [
-          { role: "system", content: "Follow the user's exact reply instruction." },
           { role: "user", content: [{ type: "input_text", text: "Reply exactly: Alpha." }] },
           { role: "assistant", content: [{ type: "output_text", text: "Alpha." }] },
           { role: "user", content: [{ type: "input_text", text: "Reply exactly: Beta." }] },
@@ -204,8 +205,8 @@ describe("OpenAI Responses WebSocket recorded", () => {
       expect(channel.sent[1]).toHaveProperty("previous_response_id", expect.any(String))
       expect(channel.sent[2]).not.toHaveProperty("previous_response_id")
       expect(channel.sent[2]).toMatchObject({
+        instructions: "Follow the user's exact reply instruction.",
         input: [
-          { role: "system", content: "Follow the user's exact reply instruction." },
           { role: "user", content: [{ type: "input_text", text: "Reply exactly: Ready." }] },
           { role: "assistant", content: [{ type: "output_text", text: "Ready." }] },
           { role: "user", content: [{ type: "input_text", text: "Reply exactly: Recovered." }] },

packages/ai/test/provider/openai-responses.test.ts

@@ -112,10 +112,8 @@ describe("OpenAI Responses route", () => {
 
       expect(prepared.body).toEqual({
         model: "gpt-4.1-mini",
-        input: [
-          { role: "system", content: "You are concise." },
-          { role: "user", content: [{ type: "input_text", text: "Say hello." }] },
-        ],
+        input: [{ role: "user", content: [{ type: "input_text", text: "Say hello." }] }],
+        instructions: "You are concise.",
         store: false,
         include: ["reasoning.encrypted_content"],
         stream: true,
@@ -469,7 +467,7 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("continues a tool call with only the new tool output", () =>
+  it.effect("continues an item-id-less tool call with only the new tool output", () =>
     Effect.gen(function* () {
       const firstRequest = {
         type: "response.create",
@@ -485,7 +483,6 @@ describe("OpenAI Responses route", () => {
           type: "response.output_item.done",
           item: {
             type: "function_call",
-            id: "fc_1",
             status: "completed",
             call_id: "call_1",
             name: "weather",
@@ -1597,8 +1594,8 @@ describe("OpenAI Responses route", () => {
       )
 
       expect(prepared.body).toMatchObject({
+        instructions: "You are concise. Continue from the provided history.",
         input: [
-          { role: "system", content: "You are concise. Continue from the provided history." },
           {
             role: "user",
             content: [
@@ -2120,6 +2117,47 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("routes item-id-less function arguments by output index and prefers item completion", () =>
+    Effect.gen(function* () {
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 2, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 2,
+                item_id: "opaque_delta",
+                delta: '{"query":"streamed"}',
+              },
+              {
+                type: "response.function_call_arguments.done",
+                output_index: 2,
+                item_id: "opaque_done",
+                arguments: '{"query":"arguments-done"}',
+              },
+              {
+                type: "response.output_item.done",
+                output_index: 2,
+                item: { ...item, arguments: '{"query":"output-item-done"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter((event) => event.type === "tool-input-delta")).toMatchObject([
+        { id: "call_1", text: '{"query":"streamed"}' },
+      ])
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "output-item-done" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+    }),
+  )
+
   it.effect("routes reasoning summary events by output index", () =>
     Effect.gen(function* () {
       const response = yield* LLMClient.generate(request).pipe(
@@ -2298,13 +2336,25 @@ describe("OpenAI Responses route", () => {
   it.effect("rejects function argument events without the spec-required item id", () =>
     Effect.gen(function* () {
       const events = [
-        { type: "response.function_call_arguments.delta", delta: "{}" },
-        { type: "response.function_call_arguments.done", arguments: "{}" },
+        { type: "response.function_call_arguments.delta", output_index: 0, delta: "{}" },
+        { type: "response.function_call_arguments.done", output_index: 0, arguments: "{}" },
       ]
 
       for (const event of events) {
         const error = yield* LLMClient.generate(request).pipe(
-          Effect.provide(fixedResponse(sseEvents(event, { type: "response.completed", response: { id: "resp_1" } }))),
+          Effect.provide(
+            fixedResponse(
+              sseEvents(
+                {
+                  type: "response.output_item.added",
+                  output_index: 0,
+                  item: { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" },
+                },
+                event,
+                { type: "response.completed", response: { id: "resp_1" } },
+              ),
+            ),
+          ),
           Effect.flip,
         )
 
@@ -3100,38 +3150,50 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
-  it.effect("drops replayed item ids outside the server's grammar", () =>
+  it.effect("preserves provider-issued item ids and removes malformed ids without dropping items", () =>
     Effect.gen(function* () {
       const prepared = yield* compileRequest(
         LLM.request({
           model,
           messages: [
             Message.assistant([
-              // Fails the message id prefix.
               {
                 type: "text",
                 text: "Hello",
                 providerMetadata: { openai: { itemId: "history_1" } },
               },
-              // Oversized for the Responses item id limit.
               {
                 type: "text",
                 text: "World",
-                providerMetadata: { openai: { itemId: `m${"a".repeat(64)}` } },
+                providerMetadata: { openai: { itemId: `message_${"a".repeat(64)}` } },
               },
-              // Fails the reasoning id prefix, so the whole item is unreplayable
-              // statelessly and is skipped rather than sent malformed.
               {
                 type: "reasoning",
                 text: "Checked the diff.",
                 providerMetadata: { openai: { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" } },
               },
+              {
+                type: "reasoning",
+                text: "Missing suffix.",
+                providerMetadata: { openai: { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
+              },
+              {
+                type: "reasoning",
+                text: "No prefix separator.",
+                providerMetadata: { openai: { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
+              },
               ToolCallPart.make({
                 id: "call_1",
                 name: "lookup",
                 input: { query: "weather" },
                 providerMetadata: { openai: { itemId: "toolu_01A" } },
               }),
+              ToolCallPart.make({
+                id: "call_2",
+                name: "lookup",
+                input: { query: "news" },
+                providerMetadata: { openai: { itemId: "fc_" } },
+              }),
             ]),
           ],
         }),
@@ -3140,18 +3202,44 @@ describe("OpenAI Responses route", () => {
       expect(prepared.body.input).toEqual([
         {
           type: "message",
+          id: "history_1",
           role: "assistant",
-          content: [
-            { type: "output_text", text: "Hello" },
-            { type: "output_text", text: "World" },
-          ],
+          content: [{ type: "output_text", text: "Hello" }],
+        },
+        {
+          type: "message",
+          id: `message_${"a".repeat(64)}`,
+          role: "assistant",
+          content: [{ type: "output_text", text: "World" }],
+        },
+        {
+          type: "reasoning",
+          id: "thinking_1",
+          summary: [{ type: "summary_text", text: "Checked the diff." }],
+          encrypted_content: "encrypted-state",
+        },
+        {
+          type: "reasoning",
+          summary: [{ type: "summary_text", text: "Missing suffix." }],
+          encrypted_content: "another-state",
+        },
+        {
+          type: "reasoning",
+          summary: [{ type: "summary_text", text: "No prefix separator." }],
         },
         {
           type: "function_call",
+          id: "toolu_01A",
           call_id: "call_1",
           name: "lookup",
           arguments: '{"query":"weather"}',
         },
+        {
+          type: "function_call",
+          call_id: "call_2",
+          name: "lookup",
+          arguments: '{"query":"news"}',
+        },
       ])
     }),
   )
@@ -3457,6 +3545,43 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("finalizes and replays a completed function call without an optional item id", () =>
+    Effect.gen(function* () {
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              {
+                type: "response.output_item.done",
+                item: { type: "function_call", call_id: "call_1", name: "lookup", arguments: '{"query":"weather"}' },
+              },
+              { type: "response.completed", response: { id: "resp_1" } },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "weather" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+
+      const prepared = yield* compileRequest(
+        LLM.request({
+          model,
+          messages: [
+            response.message,
+            Message.tool({ id: "call_1", name: "lookup", resultType: "json", result: { forecast: "sunny" } }),
+          ],
+        }),
+      )
+      expect(prepared.body.input).toEqual([
+        { type: "function_call", call_id: "call_1", name: "lookup", arguments: '{"query":"weather"}' },
+        { type: "function_call_output", call_id: "call_1", output: '{"forecast":"sunny"}' },
+      ])
+    }),
+  )
+
   it.effect("emits only missing function arguments from the arguments done event", () =>
     Effect.gen(function* () {
       const body = sseEvents(
@@ -3681,6 +3806,37 @@ describe("OpenAI Responses route", () => {
     }),
   )
 
+  it.effect("reconciles an item-id-less pending function call from completed response output", () =>
+    Effect.gen(function* () {
+      const item = { type: "function_call", call_id: "call_1", name: "lookup", arguments: "" }
+      const response = yield* LLMClient.generate(request).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "response.output_item.added", output_index: 0, item },
+              {
+                type: "response.function_call_arguments.delta",
+                output_index: 0,
+                item_id: "opaque_delta",
+                delta: '{"query":"partial',
+              },
+              {
+                type: "response.completed",
+                response: { id: "resp_1", output: [{ ...item, arguments: '{"query":"complete"}' }] },
+              },
+            ),
+          ),
+        ),
+      )
+
+      expect(response.events.filter(LLMEvent.is.toolCall)).toEqual([
+        expect.objectContaining({ id: "call_1", name: "lookup", input: { query: "complete" } }),
+      ])
+      expect(response.events.find(LLMEvent.is.toolCall)?.providerMetadata).toBeUndefined()
+      expect(response.events.filter(LLMEvent.is.toolInputEnd)).toHaveLength(1)
+    }),
+  )
+
   it.effect("lets completed response output override arguments done", () =>
     Effect.gen(function* () {
       const body = sseEvents(

packages/ai/test/recorded-golden.ts

@@ -20,6 +20,7 @@ type ScenarioInput =
       readonly name?: string
       readonly cassette?: string
       readonly tags?: ReadonlyArray<string>
+      readonly prompt?: string
       readonly maxTokens?: number
       readonly temperature?: number | false
       readonly timeout?: number
@@ -87,6 +88,7 @@ const runTarget = (target: TargetInput) => {
             yield* runGoldenScenario(input.id, {
               id: `recorded_${kebab(target.name).replaceAll("-", "_")}_${input.id.replaceAll("-", "_")}`,
               model: target.model,
+              prompt: input.prompt,
               maxTokens: input.maxTokens,
               temperature: input.temperature,
             })

packages/ai/test/recorded-scenarios.ts

@@ -164,6 +164,7 @@ export const expectGoldenWeatherToolLoop = (events: ReadonlyArray<LLMEvent>) =>
 export interface GoldenScenarioContext {
   readonly id: string
   readonly model: LanguageModel
+  readonly prompt?: string
   readonly maxTokens?: number
   readonly temperature?: number | false
 }
@@ -298,7 +299,7 @@ const runGeneratedConversation = (context: GoldenScenarioContext, steps: Readonl
 
 const runTextScenario = (context: GoldenScenarioContext) =>
   runGeneratedConversation(context, [
-    user("Reply exactly with: Hello!"),
+    user(context.prompt ?? "Reply exactly with: Hello!"),
     assistant.expectText(/^Hello!?$/, {
       system: "You are concise.",
       maxTokens: context.maxTokens ?? 40,

packages/ai/test/schema.test.ts

@@ -102,6 +102,38 @@ describe("AI.Usage", () => {
     expect(error.reason._tag).toBe("InvalidProviderOutput")
   })
 
+  test("sseFraming ignores retry directives without ending the stream", async () => {
+    const encoder = new TextEncoder()
+    const frames = await Effect.runPromise(
+      ProviderShared.sseFraming(
+        Stream.make(
+          encoder.encode("retry: 1000\n\n"),
+          encoder.encode('data: {"first":true}\n\n'),
+          encoder.encode("retry: 2000\n\n"),
+          encoder.encode('data: {"second":true}\n\n'),
+        ).pipe(Stream.rechunk(1)),
+      ).pipe(Stream.runCollect),
+    )
+
+    expect(Array.from(frames)).toEqual(['{"first":true}', '{"second":true}'])
+  })
+
+  test("sseFraming preserves event data around retry directives", async () => {
+    const encoder = new TextEncoder()
+    const frames = await Effect.runPromise(
+      ProviderShared.sseFraming(
+        Stream.make(
+          encoder.encode("event: update\ndata: first\n"),
+          encoder.encode("retry: 1000\n"),
+          encoder.encode("data: second\n\n"),
+        ).pipe(Stream.rechunk(1)),
+        new Set(["update"]),
+      ).pipe(Stream.runCollect),
+    )
+
+    expect(Array.from(frames)).toEqual(["first\nsecond"])
+  })
+
   test("visibleOutputTokens clamps reasoning > output to zero", () => {
     expect(new Usage({ outputTokens: 10, reasoningTokens: 4 }).visibleOutputTokens).toBe(6)
     expect(new Usage({ outputTokens: 10 }).visibleOutputTokens).toBe(10)

packages/app/e2e/regression/remote-session-settings.spec.ts

@@ -24,11 +24,17 @@ test("session settings use the remote server context", async ({ page }) => {
   await configureServers(page)
 
   await page.goto(`/server/${base64Encode(serverB)}/session/${sessionB.id}`)
-  await expect(page.getByRole("heading", { name: sessionB.title, exact: true })).toBeVisible()
+  const sessionHeading = page.getByRole("heading", { name: sessionB.title, exact: true, includeHidden: true })
+  await expect(sessionHeading).toBeVisible()
   await page.keyboard.press("Control+,")
 
-  const dialog = page.locator(".settings-dialog")
-  const autoAccept = dialog.locator('[data-action="settings-auto-accept-permissions"]')
+  const settings = page.getByTestId("settings-screen")
+  await expect(settings).toBeVisible()
+  await expect(page.getByRole("dialog")).toHaveCount(0)
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "328px")
+  await expect(sessionHeading).toBeAttached()
+  await expect(sessionHeading).toBeHidden()
+  const autoAccept = settings.locator('[data-action="settings-auto-accept-permissions"]')
   const input = autoAccept.getByRole("switch")
   await expect(autoAccept).toBeVisible()
   await expect(input).toBeEnabled()
@@ -55,9 +61,12 @@ test("session settings use the remote server context", async ({ page }) => {
       },
     ])
 
-  await dialog.getByRole("tab", { name: "Models" }).click()
-  await expect(dialog.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
-  await expect(dialog.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
+  await settings.getByRole("tab", { name: "Models" }).click()
+  await expect(settings.getByRole("switch", { name: "Server B Model" })).toBeEnabled()
+  await expect(settings.getByRole("switch", { name: "Server A Model" })).toHaveCount(0)
+  await settings.getByRole("button", { name: "Back to app" }).click()
+  await expect(settings).toBeHidden()
+  await expect(sessionHeading).toBeVisible()
 })
 
 test("auto-accept responds for an unfocused server session", async ({ page }) => {
@@ -78,7 +87,7 @@ test("auto-accept responds for an unfocused server session", async ({ page }) =>
   await page.goto(`/server/${base64Encode(serverA)}/session/${sessionA.id}`)
   await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()
   await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
   await autoAccept.locator('[data-slot="switch-control"]').click()
   await expect(autoAccept.getByRole("switch")).toBeChecked()
   await expect
@@ -178,7 +187,7 @@ test("auto-accept sweeps again after a reconnect", async ({ page }) => {
   const first = await transport.waitForConnection()
 
   await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
   await autoAccept.locator('[data-slot="switch-control"]').click()
   await expect(autoAccept.getByRole("switch")).toBeChecked()
   await expect
@@ -234,7 +243,7 @@ test("auto-accept approves a request discovered by opening a session", async ({
   await expect(page.getByRole("heading", { name: sessionA.title, exact: true })).toBeVisible()
 
   await page.keyboard.press("Control+,")
-  const autoAccept = page.locator(".settings-dialog").locator('[data-action="settings-auto-accept-permissions"]')
+  const autoAccept = page.getByTestId("settings-screen").locator('[data-action="settings-auto-accept-permissions"]')
   await autoAccept.locator('[data-slot="switch-control"]').click()
   await expect(autoAccept.getByRole("switch")).toBeChecked()
 

packages/app/e2e/regression/server-dialog-focus.spec.ts

@@ -2,7 +2,7 @@ import { expect, test, type Route } from "@playwright/test"
 
 const server = "http://127.0.0.1:4097"
 
-test("nested server dialog keeps focus inside the top layer", async ({ page }) => {
+test("server dialog keeps focus above fullscreen settings", async ({ page }) => {
   await page.addInitScript((server) => {
     localStorage.setItem("opencode.global.dat:server", JSON.stringify({ list: [server] }))
   }, server)
@@ -24,8 +24,9 @@ test("nested server dialog keeps focus inside the top layer", async ({ page }) =
 
   await page.goto("/")
   await page.keyboard.press("Control+,")
-  const settings = page.locator(".settings-dialog")
+  const settings = page.getByTestId("settings-screen")
   await expect(settings).toBeVisible()
+  await expect(page.getByRole("dialog")).toHaveCount(0)
   await settings.getByRole("tab", { name: "Servers" }).click()
   await settings.getByRole("button", { name: "Add server" }).click()
 
@@ -41,6 +42,9 @@ test("nested server dialog keeps focus inside the top layer", async ({ page }) =
   await expect(password).toBeFocused()
   await password.fill("secret")
   await expect(password).toHaveValue("secret")
+  await page.keyboard.press("Escape")
+  await expect(editor).toBeHidden()
+  await expect(settings).toBeVisible()
 })
 
 function json(route: Route, body: unknown, status = 200) {

packages/app/e2e/regression/session-timeline-accessibility.spec.ts

@@ -9,7 +9,7 @@ test("space activates a focused timeline button instead of scrolling", async ({
     reducedMotion: true,
   })
   const scroller = page.locator(".scroll-view__viewport", { has: page.locator("[data-timeline-row]") })
-  const trigger = page.locator(`[data-timeline-part-id="${shellID}"] [data-slot="collapsible-trigger"]`)
+  const trigger = page.getByRole("button", { name: "Used Shell" })
   await trigger.focus()
   const before = await scroller.evaluate((element) => element.scrollTop)
   await trigger.press("Space")

packages/app/e2e/regression/session-timeline-context-resize.spec.ts

@@ -40,7 +40,7 @@ test.describe("regression: session timeline context group resize", () => {
     expect(samples.at(-1)?.expanded).toBe("true")
   })
 
-  test("paints a stable exploring to explored transition", async ({ page }) => {
+  test("keeps a grouped tool summary stable as its calls complete", async ({ page }) => {
     const events: OpenCodeEvent[] = []
     await page.setViewportSize({ width: 1400, height: 900 })
     await mockServer(page, events, [
@@ -55,13 +55,12 @@ test.describe("regression: session timeline context group resize", () => {
     await devtools.send("Emulation.setCPUThrottlingRate", { rate: 4 })
     const context = page.locator(`[data-timeline-part-ids="${contextIDs.join(",")}"]`).first()
     await expectAppVisible(context)
-    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Exploring")
+    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")
 
     const contextSelector = `[data-timeline-part-ids="${contextIDs.join(",")}"]`
     const regions = defineVisualRegions({
       status: {
-        selector: `${contextSelector} [data-component="tool-status-title"]`,
-        opacitySelectors: ['[data-slot="tool-status-active"]', '[data-slot="tool-status-done"]'],
+        selector: `${contextSelector} [data-component="context-tool-group-trigger"]`,
       },
       context: { selector: contextSelector, closest: '[data-timeline-row="AssistantPart"]' },
       following: {
@@ -89,7 +88,7 @@ test.describe("regression: session timeline context group resize", () => {
       await page.waitForTimeout(delay)
     }
 
-    await expect(context.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", "Explored")
+    await expect(context.getByRole("button")).toHaveAccessibleName("Used Read, Glob, Grep, List")
     await page.waitForTimeout(700)
     const trace = await stopVisualProbe<keyof typeof regions>(page)
     const labels = trace.samples
@@ -108,7 +107,7 @@ test.describe("regression: session timeline context group resize", () => {
       ]),
     )
 
-    expect(labels).toEqual(["Exploring", "Explored"])
+    expect(labels).toEqual(["Used Read, Glob, Grep, List"])
     expect(issues, JSON.stringify(trace.samples, null, 2)).toEqual([])
   })
 })
@@ -209,13 +208,7 @@ function turn(index: number, target: boolean, status: "running" | "completed" =
   const content: SessionMessageAssistant["content"] = target
     ? [
         toolContent(
-          contextTool(
-            contextIDs[0]!,
-            assistantID,
-            "read",
-            { path: "src/recent-a.ts", offset: 0, limit: 120 },
-            status,
-          ),
+          contextTool(contextIDs[0]!, assistantID, "read", { path: "src/recent-a.ts", offset: 0, limit: 120 }, status),
         ),
         toolContent(contextTool(contextIDs[1]!, assistantID, "glob", { path: directory, pattern: "**/*.ts" }, status)),
         toolContent(

packages/app/e2e/regression/session-timeline-file-projection.spec.ts

@@ -83,6 +83,7 @@ test("keeps an expanded file diff header at the same viewport position", async (
   const before = Array.from({ length: 80 }, (_, index) => `export const value${index} = ${index}\n`).join("")
   const after = before.replaceAll(" = ", " = compute(").replaceAll("\n", ")\n")
   await setupTimeline(page, {
+    settings: { editToolPartsExpanded: true },
     messages: [
       userMessage([userText("Preceding context ".repeat(120))]),
       assistantMessage([

packages/app/e2e/regression/session-timeline-lifecycle-state.spec.ts

@@ -26,7 +26,9 @@ for (const expanded of [false, true]) {
       messages: [userMessage(), assistantMessage([shell(id, "completed", lines(3))])],
       settings: { shellToolPartsExpanded: expanded },
     })
-    const trigger = page.locator(`[data-timeline-part-id="${id}"] [data-slot="collapsible-trigger"]`)
+    const trigger = expanded
+      ? page.locator(`[data-timeline-part-id="${id}"] [data-slot="collapsible-trigger"]`)
+      : page.getByRole("button", { name: "Used Shell" })
     await expect(trigger).toHaveAttribute("aria-expanded", String(expanded))
     await trigger.click()
     await expect(trigger).toHaveAttribute("aria-expanded", String(!expanded))

packages/app/e2e/regression/session-timeline-locale-projection.spec.ts

@@ -1,12 +1,9 @@
 import { expect, test } from "@playwright/test"
 import { assistantMessage, setupTimeline, toolPart, userMessage } from "../performance/timeline-stability/fixture"
 
-for (const profile of [
-  { locale: "de", label: "Erkundung abgeschlossen" },
-  { locale: "ar", label: "تم الاستكشاف" },
-] as const) {
-  test(`projects translated context status in ${profile.locale}`, async ({ page }) => {
-    const ids = [`prt_locale_${profile.locale}_01_read`, `prt_locale_${profile.locale}_02_glob`]
+for (const locale of ["de", "ar"] as const) {
+  test(`projects localized tool names with an English fallback in ${locale}`, async ({ page }) => {
+    const ids = [`prt_locale_${locale}_01_read`, `prt_locale_${locale}_02_glob`]
     await setupTimeline(page, {
       messages: [
         userMessage(),
@@ -15,11 +12,12 @@ for (const profile of [
           toolPart(ids[1]!, "glob", "completed", { path: ".", pattern: "**/*.ts" }),
         ]),
       ],
-      locale: profile.locale,
+      locale,
     })
 
     const group = page.locator(`[data-timeline-part-ids="${ids.join(",")}"]`)
-    await expect(group.locator('[data-component="tool-status-title"]')).toHaveAttribute("aria-label", profile.label)
-    await expect(page.locator("html")).toHaveAttribute("lang", profile.locale)
+    await expect(group.getByRole("button")).toHaveAccessibleName(/^Used /)
+    await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+    await expect(page.locator("html")).toHaveAttribute("lang", locale)
   })
 }

packages/app/e2e/regression/session-timeline-projection.spec.ts

@@ -53,9 +53,14 @@ test.describe("session timeline projection", () => {
     ]
     await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
 
-    await expect(
-      page.locator('[data-timeline-part-ids="prt_01_read,prt_02_glob,prt_03_grep,prt_04_list"]'),
-    ).toBeVisible()
+    const first = page.locator(
+      '[data-timeline-part-ids="prt_01_read,prt_02_glob,prt_03_grep,prt_04_list,prt_webfetch,prt_websearch,prt_task,prt_bash,prt_edit,prt_write,prt_patch"]',
+    )
+    const second = page.locator('[data-timeline-part-ids="prt_skill,prt_custom"]')
+    await expect(first).toBeVisible()
+    await expect(second).toBeVisible()
+    await first.getByRole("button").click()
+    await second.getByRole("button").click()
     for (const id of [
       "prt_webfetch",
       "prt_websearch",
@@ -78,8 +83,7 @@ test.describe("session timeline projection", () => {
     await expect(patch.locator('[data-slot="message-part-title-filename"]')).toHaveCount(0)
     await expect(patch.locator('[data-slot="message-part-actions"]')).toHaveCount(0)
     const edit = page.locator('[data-timeline-part-id="prt_edit"]')
-    await expect(edit.locator('[data-component="apply-patch-tool"]')).toBeVisible()
-    await expect(edit.locator('[data-slot="basic-tool-tool-title"]')).toContainText("Edit")
+    await expect(edit).toContainText("Edit")
     await expect(page.locator('[data-timeline-part-id="prt_todo"]')).toHaveCount(0)
   })
 
@@ -87,6 +91,7 @@ test.describe("session timeline projection", () => {
     const first = "prt_patch_first"
     const second = "prt_patch_second"
     const timeline = await setupTimeline(page, {
+      settings: { editToolPartsExpanded: true },
       messages: [
         userMessage(),
         assistantMessage([

packages/app/e2e/regression/session-timeline-reducer-projection.spec.ts

@@ -13,7 +13,7 @@ import {
   userMessage,
 } from "../performance/timeline-stability/fixture"
 
-test("groups singleton and separated context operations at correct boundaries", async ({ page }) => {
+test("groups every collapsed tool until visible text separates the stack", async ({ page }) => {
   const parts = [
     toolPart("prt_boundary_01_read", "read", "completed", { path: "src/a.ts" }),
     textPart("prt_boundary_02_text", "Boundary text"),
@@ -25,9 +25,112 @@ test("groups singleton and separated context operations at correct boundaries",
   await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
 
   await expect(page.locator('[data-timeline-part-ids="prt_boundary_01_read"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-part-ids="prt_boundary_03_glob,prt_boundary_04_grep"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-part-ids="prt_boundary_06_list"]')).toBeVisible()
-  await expect(page.locator('[data-timeline-row="AssistantPart"]')).toHaveCount(5)
+  const group = page.locator(
+    '[data-timeline-part-ids="prt_boundary_03_glob,prt_boundary_04_grep,prt_boundary_05_shell,prt_boundary_06_list"]',
+  )
+  await expect(group).toBeVisible()
+  await expect(group.getByRole("button")).toHaveAccessibleName("Used Glob, Grep, Shell, List")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("4")
+  await expect(page.locator('[data-timeline-row="AssistantPart"]')).toHaveCount(3)
+  await expect(page.locator('[data-timeline-spacing="content"]')).toHaveCount(2)
+  await expect(page.locator('[data-timeline-spacing="content"]').nth(0)).toHaveCSS("padding-top", "16px")
+})
+
+test("expands a mixed collapsed tool stack without expanding its individual calls", async ({ page }) => {
+  const parts = [
+    shell("prt_stack_shell_1", "completed", "first"),
+    toolPart("prt_stack_explore", "subagent", "completed", {
+      agent: "explore",
+      description: "Inspect the project",
+      prompt: "Explore the project",
+    }),
+    toolPart("prt_stack_patch", "patch", "completed", { patchText: "Update src/value.ts" }),
+    shell("prt_stack_shell_2", "completed", "second"),
+  ]
+  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+
+  const group = page.locator(
+    '[data-timeline-part-ids="prt_stack_shell_1,prt_stack_explore,prt_stack_patch,prt_stack_shell_2"]',
+  )
+  const summary = group.getByRole("button", { name: "Used Shell, Explore, Patch" })
+  await expect(summary).toHaveAttribute("aria-expanded", "false")
+  await expect(summary).toHaveCSS("height", "28px")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("4")
+  await summary.click()
+  await expect(summary).toHaveAttribute("aria-expanded", "true")
+  await expect(group.locator('[data-slot="context-tool-group-item"]')).toHaveCount(4)
+  await expect(group.locator('[data-timeline-part-id="prt_stack_shell_1"]')).toBeVisible()
+  await expect(group.locator('[data-timeline-part-id="prt_stack_patch"]')).toBeVisible()
+  await expect(group.locator('[data-component="context-tool-group-list"]')).toHaveCSS("row-gap", "8px")
+  const content = group.locator(':scope > [data-component="collapsible"] > [data-slot="collapsible-content"]')
+  await expect(content).toHaveCSS("margin-left", "0px")
+  await expect(content).toHaveCSS("padding-left", "12px")
+  await expect.poll(() => content.evaluate((element) => getComputedStyle(element, "::before").content)).toBe("none")
+})
+
+test("leaves tools expanded by settings outside the collapsed stack", async ({ page }) => {
+  const parts = [
+    shell("prt_expanded_shell", "completed", "expanded"),
+    toolPart("prt_collapsed_patch", "patch", "completed", { patchText: "Update src/value.ts" }),
+    toolPart("prt_collapsed_read", "read", "completed", { path: "src/value.ts" }),
+  ]
+  await setupTimeline(page, {
+    messages: [userMessage(), assistantMessage(parts)],
+    settings: { shellToolPartsExpanded: true },
+  })
+
+  await expect(page.locator('[data-timeline-part-id="prt_expanded_shell"]')).toBeVisible()
+  const group = page.locator('[data-timeline-part-ids="prt_collapsed_patch,prt_collapsed_read"]')
+  await expect(group.getByRole("button", { name: "Used Patch, Read" })).toBeVisible()
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await expect(page.locator('[data-timeline-spacing="tool"]')).toHaveCSS("padding-top", "8px")
+})
+
+test("keeps failed search calls and their error cards inside the collapsed stack", async ({ page }) => {
+  const parts = [
+    toolPart(
+      "prt_error_glob",
+      "glob",
+      "error",
+      { path: "C:/Users", pattern: "*.ts" },
+      {
+        error: "Invalid tool input",
+      },
+    ),
+    toolPart(
+      "prt_error_grep",
+      "grep",
+      "error",
+      { path: "C:/Users", pattern: "value" },
+      {
+        error: "Search timed out after 30 seconds",
+      },
+    ),
+  ]
+  await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+
+  const group = page.locator('[data-timeline-part-ids="prt_error_glob,prt_error_grep"]')
+  const summary = group.getByRole("button", { name: "Used Glob, Grep" })
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await summary.click()
+  await expect(group.locator('[data-kind="tool-error-card"]')).toHaveCount(2)
+  const glob = group.locator('[data-timeline-part-id="prt_error_glob"]')
+  await expect(glob).toContainText("Invalid tool input")
+  await expect(glob.locator('[data-component="tool-error-card-icon"]')).toBeVisible()
+  await expect(glob.locator('[data-component="tool-error-card-icon"] use')).toHaveAttribute(
+    "href",
+    "#opencode-v2-icon-circle-exclamation",
+  )
+  await expect
+    .poll(() =>
+      glob
+        .locator('[data-kind="tool-error-card"]')
+        .evaluate((element) => getComputedStyle(element, "::before").display),
+    )
+    .toBe("none")
+  await expect(group.locator('[data-timeline-part-id="prt_error_grep"]')).toContainText(
+    "Search timed out after 30 seconds",
+  )
 })
 
 test("reducer-hardening: converges when idle arrives before final part and message completion", async ({ page }) => {

packages/app/e2e/regression/session-timeline-tool-projection.spec.ts

@@ -21,6 +21,9 @@ test("renders every tool error outcome without leaking hidden tools", async ({ p
   )
   await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
 
+  const group = page.locator(`[data-timeline-part-ids="${ordinary.map((_, index) => `prt_error_${index}`).join(",")}"]`)
+  await expect(group.locator('[data-component="tag"]')).toHaveText(String(ordinary.length))
+  await group.getByRole("button").click()
   await expect(page.locator('[data-kind="tool-error-card"]')).toHaveCount(ordinary.length + 1)
   await expect(page.getByText(/dismissed/i)).toBeVisible()
   await expect(page.locator('[data-timeline-part-id="prt_todo_error"]')).toHaveCount(0)
@@ -33,6 +36,7 @@ test("transitions shell and question through running error outcomes", async ({ p
   const shellID = "prt_transition_error_shell"
   const questionID = "prt_transition_error_question"
   const timeline = await setupTimeline(page, {
+    settings: { shellToolPartsExpanded: true },
     messages: [
       userMessage(),
       assistantMessage(
@@ -44,7 +48,6 @@ test("transitions shell and question through running error outcomes", async ({ p
       ),
     ],
   })
-  await timeline.waitForPart(shellID)
   await expect(page.locator(`[data-timeline-part-id="${questionID}"]`)).toHaveCount(0)
   await timeline.send(partUpdated(toolPart(shellID, "shell", "running", { command: "exit 1" })), 120)
   await timeline.send(partUpdated(toolPart(questionID, "question", "running", questionInput())), 180)
@@ -68,6 +71,7 @@ test("preserves surviving grouped patch state when its first patch fails", async
   const failed = "prt_grouped_patch_failed"
   const surviving = "prt_grouped_patch_surviving"
   const timeline = await setupTimeline(page, {
+    settings: { editToolPartsExpanded: true },
     messages: [
       userMessage(),
       assistantMessage(
@@ -147,10 +151,12 @@ test("labels all web search provider variants", async ({ page }) => {
     toolPart("prt_search_generic", "websearch", "completed", { query: "generic" }),
   ]
   await setupTimeline(page, { messages: [userMessage(), assistantMessage(parts)] })
+  await page.getByRole("button", { name: "Used Parallel Web Search, Exa Web Search, Web Search" }).click()
 
-  await expect(page.getByRole("button", { name: /Parallel Web Search/ })).toBeVisible()
-  await expect(page.getByRole("button", { name: /Exa Web Search/ })).toBeVisible()
-  await expect(page.getByRole("button", { name: /^Web Search/ })).toBeVisible()
+  const tools = page.locator('[data-component="context-tool-group-list"]')
+  await expect(tools.getByRole("button", { name: /Parallel Web Search/ })).toBeVisible()
+  await expect(tools.getByRole("button", { name: /Exa Web Search/ })).toBeVisible()
+  await expect(tools.getByRole("button", { name: /^Web Search/ })).toBeVisible()
 })
 
 test("labels completed searches with result counts", async ({ page }) => {
@@ -201,6 +207,11 @@ test("labels skill tools from IDs and result metadata", async ({ page }) => {
     ],
   })
 
+  const group = page.locator(`[data-timeline-part-ids="${pending},${completed}"]`)
+  await expect(group.getByRole("button")).toHaveAccessibleName("Used Skill")
+  await expect(group.locator('[data-component="tag"]')).toHaveText("2")
+  await group.getByRole("button").click()
+
   for (const [id, name] of [
     [pending, "frontend-design"],
     [completed, "OpenCode"],

packages/app/e2e/regression/subagent-child-navigation.spec.ts

@@ -42,8 +42,7 @@ test("shows parent lineage while the child timeline loads", async ({ page }) =>
   const release = Promise.withResolvers<void>()
   await page.route(
     (url) =>
-      url.pathname === `/api/session/${childID}/message` &&
-      url.port === (process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"),
+      url.pathname === `/api/session/${childID}/message` && url.port === (process.env.PLAYWRIGHT_SERVER_PORT ?? "4096"),
     async (route) => {
       requested.resolve()
       await release.promise
@@ -53,6 +52,7 @@ test("shows parent lineage while the child timeline loads", async ({ page }) =>
 
   await page.goto(sessionHref(parentID))
   await expectSessionTitle(page, parentTitle)
+  await page.getByRole("button", { name: "Used Explore" }).click()
   await page.locator(`a[href="${sessionHref(childID)}"]`).click()
   await Promise.all([requested.promise, expect(page).toHaveURL(sessionHref(childID))])
   await Promise.all([
@@ -77,6 +77,7 @@ test("keeps the parent visible while the child session resolves", async ({ page
   await page.goto(sessionHref(parentID))
   await expectSessionTitle(page, parentTitle)
 
+  await page.getByRole("button", { name: "Used Explore" }).click()
   await page.locator(`a[href="${sessionHref(childID)}"]`).click()
   await requested.promise
   await Promise.all([expect(page).toHaveURL(sessionHref(parentID)), expectSessionTitle(page, parentTitle)]).finally(
@@ -194,6 +195,7 @@ async function setup(page: Page, events?: () => OpenCodeEvent[]) {
 async function openChildFromParent(page: Page) {
   await page.goto(sessionHref(parentID))
   await expectSessionTitle(page, parentTitle)
+  await page.getByRole("button", { name: "Used Explore" }).click()
 
   const card = page.locator(`a[href="${sessionHref(childID)}"]`)
   await expect(card).toBeVisible()

packages/app/e2e/regression/tab-navigate-mousedown.spec.ts

@@ -103,6 +103,127 @@ test("cramped tabs only show the close button for the active tab", async ({ page
   await expect(tabB.locator('[data-slot="tab-close"]')).toBeVisible()
 })
 
+test("vertical tabs show project details, resize, and navigate", async ({ page }) => {
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA, sessionB }) => {
+      localStorage.setItem("settings.v3", JSON.stringify({ appearance: { tabLayout: "vertical" } }))
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([
+          { type: "session", server, sessionId: sessionA },
+          { type: "session", server, sessionId: sessionB },
+        ]),
+      )
+    },
+    { server, sessionA: sessionA.id, sessionB: sessionB.id },
+  )
+
+  const hrefA = `/server/${base64Encode(server)}/session/${sessionA.id}`
+  const hrefB = `/server/${base64Encode(server)}/session/${sessionB.id}`
+  await page.goto(hrefA)
+
+  const sidebar = page.locator('[data-slot="vertical-tabs-sidebar"]')
+  const tabA = sidebar.locator(`[data-titlebar-tab-link][href="${hrefA}"]`)
+  const tabB = sidebar.locator(`[data-titlebar-tab-link][href="${hrefB}"]`)
+  await expect(sidebar).toHaveCSS("width", "260px")
+  await expect(tabA).toContainText(sessionA.title)
+  await expect(tabB).toContainText(sessionB.title)
+  await expect(tabB.locator('[data-slot="tab-project"]')).toHaveText("tab-project")
+  await expect(sidebar.getByRole("button", { name: "New session" })).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+
+  const handle = sidebar.locator('[data-component="resize-handle"]')
+  await expect(handle).toHaveCSS("cursor", "col-resize")
+  const box = await handle.boundingBox()
+  if (!box) throw new Error("vertical tab resize handle has no bounding box")
+  await page.mouse.move(box.x + box.width / 2, box.y + box.height / 2)
+  await page.mouse.down()
+  await page.mouse.move(box.x + box.width / 2 - 80, box.y + box.height / 2)
+  await page.mouse.up()
+  await expect(sidebar).toHaveCSS("width", "180px")
+  await expect(tabB.locator('[data-slot="tab-project"]')).toHaveText("tab-project")
+
+  const resized = await handle.boundingBox()
+  if (!resized) throw new Error("resized vertical tab handle has no bounding box")
+  await page.mouse.move(resized.x + resized.width / 2, resized.y + resized.height / 2)
+  await page.mouse.down()
+  await page.mouse.move(resized.x - 200, resized.y + resized.height / 2)
+  await page.mouse.up()
+  await expect(sidebar).toHaveCSS("width", "130px")
+
+  await tabB.click()
+  await expect(page).toHaveURL(new RegExp(`${hrefB.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}$`))
+  await expect(tabB).toBeVisible()
+})
+
+test("appearance experimental setting switches tab orientation", async ({ page }) => {
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA }) => {
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "session", server, sessionId: sessionA }]),
+      )
+    },
+    { server, sessionA: sessionA.id },
+  )
+
+  await page.goto("/")
+  await expect(page.locator('[data-slot="titlebar-tabs"] [data-titlebar-tab-link]')).toBeVisible()
+  await page.keyboard.press("Control+,")
+
+  const settings = page.getByTestId("settings-screen")
+  await expect(settings).toBeVisible()
+  await settings.getByRole("tab", { name: "Appearance" }).click()
+  await expect(settings.getByRole("heading", { name: "Experimental" })).toBeVisible()
+
+  const layout = settings.locator('[data-action="settings-tab-layout"]')
+  await expect(layout).toContainText("Horizontal")
+  await layout.click()
+  await page.getByRole("option", { name: "Vertical" }).click()
+
+  await expect(layout).toContainText("Vertical")
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "240px")
+
+  await page.setViewportSize({ width: 920, height: 720 })
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toHaveCSS("width", "260px")
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "160px")
+
+  await page.setViewportSize({ width: 800, height: 720 })
+  await expect(settings.getByRole("tablist")).toHaveCSS("width", "160px")
+})
+
+test("vertical tab preference falls back to horizontal on mobile", async ({ page }) => {
+  await page.setViewportSize({ width: 390, height: 720 })
+  await mockServer(page)
+  await page.addInitScript(
+    ({ server, sessionA }) => {
+      localStorage.setItem("settings.v3", JSON.stringify({ appearance: { tabLayout: "vertical" } }))
+      localStorage.setItem(
+        "opencode.window.browser.dat:tabs",
+        JSON.stringify([{ type: "session", server, sessionId: sessionA }]),
+      )
+    },
+    { server, sessionA: sessionA.id },
+  )
+
+  const href = `/server/${base64Encode(server)}/session/${sessionA.id}`
+  await page.goto(href)
+
+  const tabs = page.locator('[data-slot="titlebar-tabs"]')
+  await expect(tabs.locator(`[data-titlebar-tab-link][href="${href}"]`)).toContainText(sessionA.title)
+  await expect(page.locator('[data-slot="vertical-tabs-sidebar"]')).toHaveCount(0)
+
+  await page.setViewportSize({ width: 1280, height: 720 })
+  await expect(
+    page.locator('[data-slot="vertical-tabs-sidebar"]').locator(`[data-titlebar-tab-link][href="${href}"]`),
+  ).toBeVisible()
+  await expect(page.locator('[data-slot="titlebar-tabs"]')).toHaveCount(0)
+})
+
 function session(id: string, title: string) {
   return {
     id,

packages/app/src/app.tsx

@@ -58,7 +58,7 @@ export function AppBaseProviders(
   props: ParentProps<{
     locale?: Locale
     onNativeTranslations?: Parameters<typeof LanguageProvider>[0]["onNativeTranslations"]
-    onThemeApplied?: () => void
+    onThemeApplied?: (mode: "light" | "dark", scheme: "system" | "light" | "dark") => void
   }>,
 ) {
   return (
@@ -67,7 +67,7 @@ export function AppBaseProviders(
       <ThemeProvider
         onThemeApplied={(_, mode, scheme) => {
           void window.api?.setTitlebar?.({ mode, scheme })
-          props.onThemeApplied?.()
+          props.onThemeApplied?.(mode, scheme)
         }}
       >
         <LanguageProvider locale={props.locale} onNativeTranslations={props.onNativeTranslations}>

packages/app/src/new-session/composer-adapter.ts

@@ -133,7 +133,7 @@ export function createNewSessionComposerAdapter(props: {
 
   return {
     adapter,
-    project: createComposerProjectControls({ draftId: props.draftID }),
+    project: createComposerProjectControls({ draftId: props.draftID, worktree: props.worktree }),
     model,
     ready: prompt.ready,
   }

packages/app/src/new-session/project/controller.ts

@@ -4,10 +4,11 @@ import { useGlobal, useServerCtx } from "@/runtime/server/runtime"
 import { useServerSDK } from "@/runtime/server/client"
 import { serverName, ServerConnection, useServers } from "@/runtime/server/registry"
 import { useWorkspaceLocation } from "@/workspaces/location"
+import { workspaceSelectionDestination } from "@/workspaces/paths"
 import { useTabs } from "@/shell/tabs/tabs"
 import type { PromptProjectControls } from "./selector"
 
-export function createComposerProjectControls(props: { draftId: string }) {
+export function createComposerProjectControls(props: { draftId: string; worktree: () => string }) {
   const servers = useServers()
   const serverSDK = useServerSDK()
   const location = useWorkspaceLocation()
@@ -38,7 +39,7 @@ export function createComposerProjectControls(props: { draftId: string }) {
     tabs.updateDraft(props.draftId, {
       server: ServerConnection.key(connection),
       directory: worktree,
-      worktree: undefined,
+      worktree: workspaceSelectionDestination(props.worktree(), location().directory),
       branch: undefined,
     })
   }

packages/app/src/new-session/workspace/controller.ts

@@ -5,6 +5,8 @@ import { useWorkspaceLocation } from "@/workspaces/location"
 import { useServerSDK } from "@/runtime/server/client"
 import { useData } from "@/runtime/server/current"
 import { useSettings } from "@/settings/model"
+import { useTabs } from "@/shell/tabs/tabs"
+import { ServerConnection } from "@/runtime/server/registry"
 import { normalizeProjectInfo } from "@/runtime/server/global-sync/utils"
 import {
   isWorkspaceDirectory,
@@ -12,6 +14,7 @@ import {
   sameDirectory,
   workspaceDefaultSelection,
   workspaceDirectories,
+  workspaceSelectionDestination,
 } from "@/workspaces/paths"
 
 export function resolveNewSessionWorktree(input: {
@@ -57,6 +60,7 @@ export function createNewSessionWorkspaceController(input: {
   const serverSDK = useServerSDK()
   const data = useData()
   const settings = useSettings()
+  const tabs = useTabs()
   const [state, setState] = createStore({ search: "" })
   const searchBranches = debounce((search: string) => setState("search", search.trim()), 100)
   const currentProject = createMemo(() => {
@@ -121,7 +125,8 @@ export function createNewSessionWorkspaceController(input: {
   const remember = (worktree = value()) => {
     const project = currentProject()
     if (!project) return
-    const local = worktree === "main" || sameDirectory(worktree, project.worktree)
+    tabs.initializeDraftWorktrees(ServerConnection.key(serverSDK.server), sdk().directory, fallback())
+    const local = workspaceSelectionDestination(worktree, project.worktree) === "main"
     settings.workspaces.setLastUsed(serverSDK.scope, project.id, local ? "local" : "workspace")
   }
 
@@ -141,6 +146,7 @@ export function createNewSessionWorkspaceController(input: {
       set: (worktree: string) => {
         input.setSelectedBranch(undefined)
         input.setSelectedWorktree(normalizeNewSessionWorktree(worktree, sdk().directory, currentProject()?.worktree))
+        remember(worktree)
       },
       create: (branch: string) => {
         input.setSelectedBranch(branch)
@@ -160,7 +166,10 @@ export function createNewSessionWorkspaceController(input: {
         const loaded = branches.latest
         const list = loaded?.directory === projectRoot() ? loaded.data : []
         return [
-          ...new Set([...list, ...(current && current.toLowerCase().includes(state.search.toLowerCase()) ? [current] : [])]),
+          ...new Set([
+            ...list,
+            ...(current && current.toLowerCase().includes(state.search.toLowerCase()) ? [current] : []),
+          ]),
         ].slice(0, 50)
       },
       searchBranches,

packages/app/src/runtime/i18n/en.ts

@@ -884,6 +884,7 @@ export const dict = {
 
   "settings.section.desktop": "Desktop",
   "settings.section.server": "Server",
+  "settings.backToApp": "Back to app",
   "settings.tab.general": "General",
   "settings.tab.preferences": "Preferences",
   "settings.tab.shortcuts": "Shortcuts",
@@ -892,6 +893,11 @@ export const dict = {
   "settings.tab.extensions": "Extensions",
   "settings.preferences.description": "Customize preferences and theme and default behavior",
   "settings.appearance.description": "Customize theme and fonts",
+  "settings.appearance.section.experimental": "Experimental",
+  "settings.appearance.row.tabs.title": "Tabs",
+  "settings.appearance.row.tabs.description": "Choose how session tabs are arranged",
+  "settings.appearance.row.tabs.horizontal": "Horizontal",
+  "settings.appearance.row.tabs.vertical": "Vertical",
   "settings.notifications.description": "Choose when to receive notifications and hear sounds",
   "settings.shortcuts.description": "Customize shortcuts for common actions",
   "settings.servers.description": "Manage server connections",

packages/app/src/session/composer/queue.test.ts

@@ -0,0 +1,54 @@
+import { describe, expect, test } from "bun:test"
+import type { SessionInboxInfo } from "@opencode-ai/client/promise"
+import { queuedPromptRows } from "./queue"
+
+const queued = [
+  {
+    id: "msg_original",
+    sessionID: "ses_1",
+    timeCreated: 1,
+    type: "user",
+    delivery: "queue",
+    payload: { text: "original" },
+  },
+  {
+    id: "msg_replacement",
+    sessionID: "ses_1",
+    timeCreated: 2,
+    type: "user",
+    delivery: "queue",
+    payload: { text: "edited" },
+  },
+] satisfies SessionInboxInfo[]
+
+describe("queuedPromptRows", () => {
+  test("keeps the edited prompt to one row while its replacement is admitted", () => {
+    expect(queuedPromptRows(queued, { original: "msg_original", replacement: "msg_replacement" })).toEqual([
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+
+  test("keeps the original visible until its replacement appears", () => {
+    expect(queuedPromptRows([queued[0]], { original: "msg_original", replacement: "msg_replacement" })).toEqual([
+      { id: "msg_original", text: "original", attachments: false },
+    ])
+  })
+
+  test("retains unrelated queue entries", () => {
+    expect(queuedPromptRows(queued)).toEqual([
+      { id: "msg_original", text: "original", attachments: false },
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+
+  test("keeps other prompts visible while a mutation replaces the edited prompt", () => {
+    const other = { ...queued[0], id: "msg_other", payload: { text: "other" } }
+
+    expect(
+      queuedPromptRows([queued[0], other, queued[1]], { original: "msg_original", replacement: "msg_replacement" }),
+    ).toEqual([
+      { id: "msg_other", text: "other", attachments: false },
+      { id: "msg_replacement", text: "edited", attachments: false },
+    ])
+  })
+})

packages/app/src/session/composer/queue.ts

@@ -1,6 +1,8 @@
 import { createEffect, createMemo, onCleanup, type Accessor } from "solid-js"
 import { createStore } from "solid-js/store"
+import { useMutation } from "@tanstack/solid-query"
 import type { SessionInboxInfo } from "@opencode-ai/client/promise"
+import { SessionMessage } from "@opencode-ai/schema/session-message"
 import type { ComposerDelivery } from "@/composer/adapter"
 import type { ComposerModel } from "@/composer/model"
 import type { ComposerStateTarget } from "@/composer/submission-state"
@@ -34,39 +36,67 @@ export function createSessionQueue(input: {
   const server = useServerSDK()
   const location = useWorkspaceLocation()
   const language = useLanguage()
-  const [state, setState] = createStore<{ editing?: { id: string; stash: EditStash }; busy: boolean }>({ busy: false })
+  const [state, setState] = createStore<{ editing?: { id: string; stash: EditStash } }>({})
+  const notify = () => showToast({ title: language.t("common.requestFailed") })
+  const mutation = useMutation(() => ({
+    mutationFn: async (
+      change:
+        | { type: "reorder"; inboxIDs: string[] }
+        | {
+            type: "edit"
+            inboxIDs: string[]
+            original: string
+            replacement: string
+            item: QueuedPrompt | undefined
+            prompt: Prompt
+            text: string
+            delivery: ComposerDelivery
+          },
+    ) => {
+      if (change.type === "reorder") return rewrite(change.inboxIDs)
+      const replacement = await editedPromptInput(
+        input.sessionID,
+        location().directory,
+        change.item,
+        change.prompt,
+        change.text,
+      )
+      // Admit before cancelling so a failed replacement never discards the original.
+      const admitted = await data.session.prompt({
+        ...replacement,
+        id: change.replacement,
+        delivery: change.delivery,
+        ...(change.delivery === "queue" ? { resume: false } : {}),
+      })
+      await server.api.session.inbox.cancel({ sessionID: input.sessionID, inboxID: change.original })
+      cancelEdit()
+      if (change.delivery === "queue")
+        await rewrite(change.inboxIDs.map((id) => (id === change.original ? admitted.id : id)))
+    },
+    onError: notify,
+    onSettled: () => data.session.pending.sync(input.sessionID).catch(() => undefined),
+  }))
 
   const queued = createMemo(() =>
     data.session.pending
       .list(input.sessionID)
       .filter((item): item is QueuedPrompt => item.type === "user" && item.delivery === "queue"),
   )
-  const rows = createMemo(() =>
-    queued().map((item) => ({
-      id: item.id,
-      text: queuedPromptText(item),
-      attachments: (item.payload.files?.length ?? 0) > 0,
-    })),
-  )
+  const rows = createMemo(() => {
+    const replacement = mutation.isPending ? mutation.variables : undefined
+    return queuedPromptRows(
+      queued(),
+      replacement?.type === "edit" && replacement.delivery === "queue" ? replacement : undefined,
+    )
+  })
 
   createEffect(() => {
     const editing = state.editing
-    if (!editing || state.busy || queued().some((item) => item.id === editing.id)) return
+    if (!editing || mutation.isPending || queued().some((item) => item.id === editing.id)) return
     setState("editing", undefined)
   })
   onCleanup(() => cancelEdit())
 
-  const notify = () => showToast({ title: language.t("common.requestFailed") })
-  const run = (work: () => Promise<unknown>) => {
-    setState("busy", true)
-    return work()
-      .catch(() => notify())
-      .finally(async () => {
-        await data.session.pending.sync(input.sessionID).catch(() => undefined)
-        setState("busy", false)
-      })
-  }
-
   const rewrite = async (inboxIDs: string[]) => {
     const pending = await server.api.session.inbox.list({ sessionID: input.sessionID })
     if (pending.some((item) => item.delivery === "queue" && item.type !== "user"))
@@ -108,12 +138,12 @@ export function createSessionQueue(input: {
     return server.api.session.inbox.cancel({ sessionID: input.sessionID, inboxID: id }).catch(() => notify())
   }
   const reorder = (inboxIDs: string[]) => {
-    if (state.busy) return Promise.resolve()
-    return run(() => rewrite(inboxIDs))
+    if (mutation.isPending) return Promise.resolve()
+    return mutation.mutateAsync({ type: "reorder", inboxIDs }).catch(() => undefined)
   }
 
   const edit = (id: string) => {
-    if (state.busy) return false
+    if (mutation.isPending) return false
     if (state.editing?.id === id) return true
     const item = queued().find((entry) => entry.id === id)
     if (!item) return false
@@ -147,25 +177,22 @@ export function createSessionQueue(input: {
   }
   const confirmEdit = (delivery: ComposerDelivery) => {
     const editing = state.editing
-    if (!editing || state.busy) return
+    if (!editing || mutation.isPending) return
     const prompt = clonePrompt(input.draft.current())
     const text = prompt.map((part) => ("content" in part ? part.content : "")).join("")
     if (!text.trim() && !prompt.some((part) => part.type === "image")) return cancelEdit()
     const item = queued().find((entry) => entry.id === editing.id)
     const pristine = item && text.trim() === queuedPromptText(item) && !prompt.some((part) => part.type === "image")
     if (pristine && delivery === "queue") return cancelEdit()
-    const inboxIDs = queued().map((entry) => entry.id)
-    void run(async () => {
-      const replacement = await editedPromptInput(input.sessionID, location().directory, item, prompt, text)
-      // Admit before cancelling so a failed replacement never discards the original.
-      const admitted = await data.session.prompt({
-        ...replacement,
-        delivery,
-        ...(delivery === "queue" ? { resume: false } : {}),
-      })
-      await server.api.session.inbox.cancel({ sessionID: input.sessionID, inboxID: editing.id })
-      cancelEdit()
-      if (delivery === "queue") await rewrite(inboxIDs.map((id) => (id === editing.id ? admitted.id : id)))
+    mutation.mutate({
+      type: "edit",
+      inboxIDs: queued().map((entry) => entry.id),
+      original: editing.id,
+      replacement: SessionMessage.ID.create(),
+      item,
+      prompt,
+      text,
+      delivery,
     })
   }
   const editFirst = () => {
@@ -187,7 +214,7 @@ export function createSessionQueue(input: {
     cancelEdit,
     editFirst,
     rows,
-    busy: () => state.busy,
+    busy: () => mutation.isPending,
     working: input.working,
     steer,
     remove,
@@ -204,6 +231,17 @@ export type SessionQueueView = Pick<
   "rows" | "editing" | "working" | "busy" | "steer" | "remove" | "edit" | "reorder"
 >
 
+export function queuedPromptRows(items: QueuedPrompt[], replacement?: { original: string; replacement: string }) {
+  const replaced = replacement && items.some((item) => item.id === replacement.replacement)
+  return items
+    .filter((item) => !replaced || item.id !== replacement.original)
+    .map((item) => ({
+      id: item.id,
+      text: queuedPromptText(item),
+      attachments: (item.payload.files?.length ?? 0) > 0,
+    }))
+}
+
 export function queuedPromptText(item: QueuedPrompt) {
   const display = item.payload.metadata?.["displayText"]
   return typeof display === "string" && display.length > 0 ? display : item.payload.text

packages/app/src/session/timeline/controller.tsx

@@ -105,6 +105,8 @@ export function createTimelineController(input: { session: TimelineSessionSource
     sessionMessages: projectedMessages,
     status: input.session.data.status,
     showReasoningSummaries: settings.general.showReasoningSummaries,
+    shellToolDefaultOpen: settings.general.shellToolPartsExpanded,
+    editToolDefaultOpen: settings.general.editToolPartsExpanded,
     pendingUserMessageIDs,
   })
   const [pending, setPending] = createStore({ rename: false })

packages/app/src/session/timeline/message-timeline.tsx

@@ -342,12 +342,11 @@ function MessageTimelineView(
   const projection = props.data.projection
   const sessionDirectory = createMemo(() => props.session.data.info()?.location.directory ?? sdk().directory)
   const project = createMemo(() => {
-    const projectID = props.session.data.info()?.projectID
-    const value = projectID
-      ? data.project.get(projectID)
-      : data.project.list().find((item) => containsDirectory(item.canonical, sessionDirectory()))
-    if (!value) return undefined
-    return { ...value, worktree: value.canonical, worktrees: [] }
+    const session = props.session.data.info()
+    const projects = server.ctx.sync.data.project
+    return session
+      ? projectForSession(session, projects)
+      : projects.find((item) => containsDirectory(item.worktree, sessionDirectory()))
   })
   const workspaceSession = createMemo(() => isWorkspaceDirectory(project(), sessionDirectory()))
   const showProjectIcon = () =>
@@ -388,7 +387,7 @@ function MessageTimelineView(
   const pinned = createMemo(() => props.pinned)
   const messageByID = projection.messageByID
   const virtualized = createTimelineVirtualizer({
-    sessionKey: props.data.sessionKey,
+    sessionKey: () => `${server.key}/${props.data.sessionID()}`,
     projection,
     showHeader,
     pinned,

packages/app/src/session/timeline/projection.ts

@@ -8,6 +8,8 @@ export function createTimelineProjection(input: {
   sessionMessages: Accessor<SessionMessageInfo[]>
   status: Accessor<SessionStatus>
   showReasoningSummaries: Accessor<boolean>
+  shellToolDefaultOpen: Accessor<boolean>
+  editToolDefaultOpen: Accessor<boolean>
   pendingUserMessageIDs: Accessor<ReadonlySet<string>>
 }) {
   const sessionMessageByID = createMemo(
@@ -81,6 +83,8 @@ export function createTimelineProjection(input: {
       input.showReasoningSummaries(),
       input.status(),
       input.pendingUserMessageIDs(),
+      input.shellToolDefaultOpen(),
+      input.editToolDefaultOpen(),
     ),
   )
   const activeMessageID = createMemo(() => projection().activeMessageID)

packages/app/src/session/timeline/virtualizer.tsx

@@ -92,6 +92,15 @@ export function createTimelineVirtualizer(input: Input) {
     initialOffset: () => (input.pinned() ? Number.MAX_SAFE_INTEGER : 0),
     initialMeasurementsCache: initialMeasurements,
     estimateSize: () => fallbackItemSize,
+    // Do not replace this with TanStack's default measurer: without a ResizeObserver entry,
+    // it returns the cached height instead of reading the element (TanStack/virtual#1183).
+    // Restored sessions, deferred tools, and rewrapped content can then keep stale heights;
+    // our fixed-height, overflow-clipped rows will hide their content. Keep observer entries
+    // on the cheap precomputed path, but make explicit measurements read the real height.
+    measureElement: (element, entry) => {
+      const box = entry?.borderBoxSize[0]
+      return box ? Math.round(box.blockSize) : element.offsetHeight
+    },
     scrollToFn: (offset, options, instance) => {
       if (virtualContent) virtualContent.style.height = `${instance.getTotalSize()}px`
       elementScroll(offset, options, instance)

packages/app/src/settings/appearance/appearance.tsx

@@ -9,6 +9,7 @@ import { createAppearanceSettingsController, type AppearanceSettingsController }
 import "@/settings/settings.css"
 
 const schemeOptions: ("system" | "light" | "dark")[] = ["system", "light", "dark"]
+const tabLayoutOptions: ("horizontal" | "vertical")[] = ["horizontal", "vertical"]
 const fontSettings = {
   ui: {
     action: "settings-ui-font",
@@ -126,6 +127,30 @@ export const SettingsAppearance: Component = () => {
             <FontSetting kind="terminal" fonts={appearance.fonts} />
           </SettingsList>
         </div>
+
+        <div class="settings-section">
+          <h3 class="settings-section-title">{language.t("settings.appearance.section.experimental")}</h3>
+          <SettingsList>
+            <SettingsRow
+              title={language.t("settings.appearance.row.tabs.title")}
+              description={language.t("settings.appearance.row.tabs.description")}
+            >
+              <Select
+                data-action="settings-tab-layout"
+                options={tabLayoutOptions}
+                current={tabLayoutOptions.find((option) => option === appearance.tabs.current())}
+                placement="bottom-end"
+                gutter={6}
+                label={(option) =>
+                  option === "horizontal"
+                    ? language.t("settings.appearance.row.tabs.horizontal")
+                    : language.t("settings.appearance.row.tabs.vertical")
+                }
+                onSelect={(option) => option && appearance.tabs.select(option)}
+              />
+            </SettingsRow>
+          </SettingsList>
+        </div>
       </div>
     </>
   )

packages/app/src/settings/command.tsx

@@ -1,24 +1,10 @@
-import { onCleanup } from "solid-js"
 import { useCommand } from "@/shell/commands/command"
 import { useLanguage } from "@/runtime/i18n/language"
-import { useDialog } from "@opencode-ai/ui/context/dialog"
+import { useSettingsSurface } from "./surface"
 
 export function useSettingsDialog(defaultValue?: string) {
-  const dialog = useDialog()
-  let run = 0
-  let dead = false
-
-  onCleanup(() => {
-    dead = true
-  })
-
-  return () => {
-    const current = ++run
-    void import("@/settings/shell").then((module) => {
-      if (dead || run !== current) return
-      void dialog.show(() => <module.DialogSettings defaultValue={defaultValue} />)
-    })
-  }
+  const settings = useSettingsSurface()
+  return () => settings.open(defaultValue)
 }
 
 export function useSettingsCommand() {

packages/app/src/settings/general/controllers.ts

@@ -81,6 +81,10 @@ export function createAppearanceSettingsController() {
       setCode: (value: string) => settings.appearance.setFont(value),
       setTerminal: (value: string) => settings.appearance.setTerminalFont(value),
     },
+    tabs: {
+      current: settings.appearance.tabLayout,
+      select: settings.appearance.setTabLayout,
+    },
   }
 }
 

packages/app/src/settings/model.tsx

@@ -8,6 +8,7 @@ export type WorkspaceDefaultDestination = "last-used" | "local" | "new"
 export type WorkspaceLastUsed = "local" | "workspace"
 export type TerminalPlacement = "side" | "bottom"
 export type FollowUpBehavior = "queue" | "steer"
+export type TabLayout = "horizontal" | "vertical"
 
 export interface NotificationSettings {
   agent: boolean
@@ -47,6 +48,7 @@ export interface Settings {
     mono: string
     sans: string
     terminal: string
+    tabLayout: TabLayout
   }
   keybinds: Record<string, string>
   permissions: {
@@ -135,6 +137,7 @@ const defaultSettings: Settings = {
     mono: "",
     sans: "",
     terminal: "",
+    tabLayout: "horizontal",
   },
   keybinds: {},
   permissions: {
@@ -287,6 +290,10 @@ export const { use: useSettings, provider: SettingsProvider } = createSimpleCont
         setTerminalFont(value: string) {
           setStore("appearance", "terminal", value.trim() ? value : "")
         },
+        tabLayout: withFallback(() => store.appearance?.tabLayout, defaultSettings.appearance.tabLayout),
+        setTabLayout(value: TabLayout) {
+          setStore("appearance", "tabLayout", value)
+        },
       },
       keybinds: {
         get: (action: string) => store.keybinds?.[action],

packages/app/src/settings/settings.css

@@ -5,6 +5,90 @@
   height: 100%;
 }
 
+.settings-screen {
+  display: flex;
+  width: 100%;
+  height: 100%;
+  min-width: 0;
+  min-height: 0;
+  justify-content: center;
+  overflow: hidden;
+  background: var(--v2-background-bg-deep);
+  outline: none;
+  container: settings-screen / inline-size;
+}
+
+.settings-screen > .settings {
+  display: flex;
+  width: min(100%, 1048px);
+  min-width: 0;
+}
+
+.settings-screen
+  > .settings[data-component="tabs-v2"][data-variant="settings"][data-orientation="vertical"]
+  > [data-slot="tabs-v2-list"] {
+  width: 328px;
+  min-width: 328px;
+  padding-block: 48px;
+  padding-inline-start: 24px;
+  padding-inline-end: 104px;
+  border-inline-end: none;
+  background: transparent;
+}
+
+.settings-screen > .settings > .settings-panel {
+  flex: 1;
+  min-width: 0;
+  max-width: 720px;
+}
+
+.settings-screen .settings-tab-header {
+  padding: 48px 0 32px;
+  background: linear-gradient(to bottom, var(--v2-background-bg-deep) calc(100% - 24px), transparent);
+}
+
+.settings-screen .settings-tab-body {
+  padding-inline: 0;
+}
+
+.settings-nav {
+  display: flex;
+  width: 200px;
+  flex-direction: column;
+  gap: 16px;
+}
+
+.settings-back {
+  display: flex;
+  height: 28px;
+  align-items: center;
+  gap: 6px;
+  padding-inline: 6px;
+  border: 0;
+  border-radius: 4px;
+  background: transparent;
+  color: var(--v2-text-text-muted);
+  font: inherit;
+  line-height: var(--line-height-compact);
+  text-align: start;
+  cursor: pointer;
+}
+
+.settings-back:hover,
+.settings-back:focus-visible {
+  background: var(--v2-background-bg-layer-03);
+  color: var(--v2-text-text-base);
+}
+
+.settings-back:focus-visible {
+  outline: 2px solid var(--v2-border-border-focus);
+  outline-offset: 2px;
+}
+
+[dir="rtl"] .settings-back-icon {
+  transform: scaleX(-1);
+}
+
 [data-component="dialog-v2"][data-variant="settings"] [data-slot="dialog-container"] {
   background: var(--v2-background-bg-base);
 }
@@ -21,6 +105,7 @@
   overflow-y: auto;
   scrollbar-width: none;
   user-select: none;
+  container: settings-panel / inline-size;
 }
 
 .settings-panel :is(input, textarea, [contenteditable="true"]) {
@@ -172,12 +257,16 @@
   width: 100%;
 }
 
-[data-component="dialog-v2"][data-variant="settings"] [data-component="select-v2-root"] {
+:is([data-component="dialog-v2"][data-variant="settings"], .settings-screen) [data-component="select-v2-root"] {
   width: fit-content;
   max-width: 100%;
 }
 
-[data-component="dialog-v2"][data-variant="settings"] [data-component="button-v2"] {
+:is([data-component="dialog-v2"][data-variant="settings"], .settings-screen) [data-component="select-v2"] {
+  max-width: 100%;
+}
+
+:is([data-component="dialog-v2"][data-variant="settings"], .settings-screen) [data-component="button-v2"] {
   width: fit-content;
   max-width: 100%;
 }
@@ -187,27 +276,93 @@
 }
 
 @media (max-width: 639px) {
+  .settings-screen .settings-nav {
+    width: 100%;
+  }
+
+  .settings-screen > .settings > .settings-panel {
+    padding-inline-end: 16px;
+  }
+
+  .settings-screen .settings-tab-header {
+    padding-top: 24px;
+  }
+
   .settings[data-component="tabs-v2"][data-variant="settings"][data-orientation="vertical"]
     > [data-slot="tabs-v2-list"] {
     width: 144px;
     min-width: 144px;
     padding-inline: 8px;
   }
+
+  .settings-screen
+    > .settings[data-component="tabs-v2"][data-variant="settings"][data-orientation="vertical"]
+    > [data-slot="tabs-v2-list"] {
+    width: 160px;
+    min-width: 160px;
+    padding-block: 24px;
+    padding-inline: 12px;
+  }
 }
 
-.settings-nav-footer {
-  display: flex;
-  flex-direction: column;
-  gap: 8px;
-  padding: 4px 0 4px 4px;
-  user-select: none;
+@container settings-screen (min-width: 800px) and (max-width: 1047px) {
+  .settings-screen > .settings {
+    padding-inline: 24px;
+  }
+
+  .settings-screen
+    > .settings[data-component="tabs-v2"][data-variant="settings"][data-orientation="vertical"]
+    > [data-slot="tabs-v2-list"] {
+    width: 240px;
+    min-width: 240px;
+    padding-inline-start: 16px;
+    padding-inline-end: 24px;
+  }
 }
 
-.settings-nav-footer > span {
-  font-size: 11px;
-  font-weight: 440;
-  line-height: 1;
-  color: var(--v2-text-text-faint);
+@container settings-screen (max-width: 799px) {
+  .settings-screen .settings-nav {
+    width: 100%;
+  }
+
+  .settings-screen > .settings > .settings-panel {
+    padding-inline-end: 16px;
+  }
+
+  .settings-screen .settings-tab-header {
+    padding-top: 24px;
+  }
+
+  .settings-screen
+    > .settings[data-component="tabs-v2"][data-variant="settings"][data-orientation="vertical"]
+    > [data-slot="tabs-v2-list"] {
+    width: 160px;
+    min-width: 160px;
+    padding-block: 24px;
+    padding-inline: 12px;
+  }
+}
+
+@container settings-panel (max-width: 520px) {
+  .settings-tab-header-row {
+    flex-wrap: wrap;
+  }
+
+  .settings-tab-header-row [data-component="select-v2-root"] {
+    width: 100%;
+  }
+
+  .settings-tab-header-row [data-component="select-v2"] {
+    width: 100%;
+  }
+
+  [data-component="settings-row"] {
+    flex-wrap: wrap;
+  }
+
+  [data-slot="settings-row-control"] {
+    width: 100%;
+  }
 }
 
 .settings-provider-row {
@@ -230,6 +385,12 @@
   }
 }
 
+@container settings-panel (max-width: 520px) {
+  .settings-provider-row {
+    flex-wrap: wrap;
+  }
+}
+
 .settings-providers [data-component="provider-icon"] {
   color: var(--v2-icon-icon-base);
 }
@@ -904,6 +1065,40 @@
   }
 }
 
+@container settings-panel (max-width: 520px) {
+  .settings-workspaces-toolbar,
+  .settings-workspaces-main {
+    align-items: flex-start;
+  }
+
+  .settings-workspaces-toolbar {
+    flex-wrap: wrap;
+  }
+
+  .settings-workspaces-toolbar-actions {
+    width: 100%;
+    flex-wrap: wrap;
+    justify-content: space-between;
+  }
+
+  .settings-workspaces-inventory [data-component="settings-list"] {
+    max-height: none;
+    overflow-y: visible;
+    padding: 14px;
+  }
+
+  .settings-workspaces-path {
+    overflow: visible;
+    text-overflow: clip;
+    white-space: normal;
+    overflow-wrap: anywhere;
+  }
+
+  .settings-workspaces-active {
+    display: none;
+  }
+}
+
 [data-component="dialog-v2"].settings-server-dialog [data-slot="dialog-container"] {
   width: 480px;
   max-width: calc(100vw - 32px);
@@ -949,7 +1144,7 @@
 }
 
 .settings-extensions-tabs[data-component="tabs-v2"][data-variant="pill"] > [data-slot="tabs-v2-list"] {
-  width: 280px;
+  width: min(280px, 100%);
   padding-inline: 0 !important;
 }
 

packages/app/src/settings/shell.tsx

@@ -1,9 +1,7 @@
-import { Component, createEffect, createMemo, createSignal, startTransition } from "solid-js"
-import { Dialog } from "@opencode-ai/ui/dialog"
+import { Component, createEffect, createMemo, createSignal, onCleanup, onMount, startTransition } from "solid-js"
 import { Tabs } from "@opencode-ai/ui/tabs"
 import { Icon } from "@opencode-ai/ui/icon"
 import { useLanguage } from "@/runtime/i18n/language"
-import { usePlatform } from "@/runtime/platform/platform"
 import { SettingsGeneral } from "./general/general"
 import { SettingsAppearance } from "./appearance/appearance"
 import { SettingsKeybinds } from "./keybinds/keybinds"
@@ -20,19 +18,32 @@ import { useLayout } from "@/shell/state/layout"
 import { useTabs } from "@/shell/tabs/tabs"
 import { useGlobal, useServerCtx } from "@/runtime/server/runtime"
 import { ServerConnection, useServers } from "@/runtime/server/registry"
+import { useCommand } from "@/shell/commands/command"
+import { useSettingsSurface } from "./surface"
 import "@/settings/settings.css"
 
-export const DialogSettings: Component<{
+export const SettingsScreen: Component<{
   defaultValue?: string
 }> = (props) => {
   const language = useLanguage()
-  const platform = usePlatform()
   const dialog = useDialog()
+  const command = useCommand()
+  const surface = useSettingsSurface()
   const layout = useLayout()
   const servers = useServers()
   const tabs = useTabs()
   const global = useGlobal()
   const [tab, setTab] = createSignal(props.defaultValue ?? "general")
+  let root: HTMLDivElement | undefined
+
+  onMount(() => {
+    command.keybinds(false)
+    root?.focus({ preventScroll: true })
+  })
+  onCleanup(() => command.keybinds(true))
+
+  createEffect(() => setTab(props.defaultValue ?? "general"))
+
   const server = createMemo(() => {
     const route = layout.route()
     switch (route.type) {
@@ -67,11 +78,22 @@ export const DialogSettings: Component<{
   })
 
   const showProviders = () => {
-    void dialog.show(() => <DialogSettings defaultValue="providers" />)
+    dialog.close()
+    setTab("providers")
   }
 
   return (
-    <Dialog size="x-large" variant="settings" class="settings-dialog">
+    <div
+      ref={root}
+      data-testid="settings-screen"
+      class="settings-screen"
+      tabIndex={-1}
+      onKeyDown={(event) => {
+        if (event.key !== "Escape" || event.defaultPrevented || dialog.active) return
+        event.preventDefault()
+        surface.close()
+      }}
+    >
       <Tabs
         orientation="vertical"
         variant="settings"
@@ -80,7 +102,11 @@ export const DialogSettings: Component<{
         class="settings"
       >
         <Tabs.List>
-          <div class="flex flex-col justify-between h-full w-full">
+          <div class="settings-nav">
+            <button type="button" class="settings-back" onClick={surface.close}>
+              <Icon name="arrow-left" size="small" class="settings-back-icon" />
+              <span>{language.t("settings.backToApp")}</span>
+            </button>
             <div class="flex flex-col gap-4 w-full">
               {/* Group 1: Preferences */}
               <div class="flex flex-col gap-1 w-full">
@@ -134,11 +160,6 @@ export const DialogSettings: Component<{
                 </Tabs.Trigger>
               </div>
             </div>
-
-            <div class="settings-nav-footer">
-              <span>{language.t("app.name.desktop")}</span>
-              <span>v{platform.version}</span>
-            </div>
           </div>
         </Tabs.List>
 
@@ -175,6 +196,6 @@ export const DialogSettings: Component<{
           </Tabs.Content>
         </SettingsServerScope>
       </Tabs>
-    </Dialog>
+    </div>
   )
 }

packages/app/src/settings/surface.tsx

@@ -0,0 +1,32 @@
+import { useLocation } from "@solidjs/router"
+import { createEffect, on } from "solid-js"
+import { createStore } from "solid-js/store"
+import { createSimpleContext } from "@opencode-ai/ui/context"
+
+export const { use: useSettingsSurface, provider: SettingsSurfaceProvider } = createSimpleContext({
+  name: "SettingsSurface",
+  gate: false,
+  init: () => {
+    const location = useLocation()
+    const [store, setStore] = createStore({ open: false, tab: "general" })
+    let focus: HTMLElement | undefined
+
+    const close = () => {
+      if (!store.open) return
+      setStore("open", false)
+      if (focus?.isConnected) focus.focus({ preventScroll: true })
+      focus = undefined
+    }
+
+    createEffect(on(() => `${location.pathname}${location.search}`, close, { defer: true }))
+
+    return {
+      store,
+      open(tab = "general") {
+        if (!store.open && document.activeElement instanceof HTMLElement) focus = document.activeElement
+        setStore({ open: true, tab })
+      },
+      close,
+    }
+  },
+})

packages/app/src/settings/workspaces/project-dialog.tsx

@@ -11,7 +11,8 @@ import { useLanguage } from "@/runtime/i18n/language"
 import { getProjectAvatarVariant, type LocalProject } from "@/shell/state/layout"
 import { ServerConnection } from "@/runtime/server/registry"
 import { LocationProvider } from "@/workspaces/location"
-import { displayName, getProjectAvatarSource } from "@/shell/layout/helpers"
+import { displayName } from "@/shell/layout/helpers"
+import { ProjectIcon } from "@/shell/layout/project-icon"
 import { createEditProjectModel } from "./project-model"
 import { ProjectSettingsExtensions } from "./project-extensions"
 import { SettingsServerDataScope } from "@/settings/server-scope"
@@ -57,11 +58,7 @@ function ProjectSettingsDialog(props: { project: LocalProject; server: ServerCon
         <Tabs.List>
           <div class="project-settings-nav">
             <Tabs.Trigger value="general">
-              <ProjectAvatar
-                fallback={projectName()}
-                variant={getProjectAvatarVariant(props.project.icon?.color)}
-                class="!size-4 shrink-0"
-              />
+              <ProjectIcon project={props.project} class="!size-4 shrink-0" />
               <span class="truncate">{projectName()}</span>
             </Tabs.Trigger>
             <Tabs.Trigger value="scripts">
@@ -114,14 +111,14 @@ function ProjectSettingsDialog(props: { project: LocalProject; server: ServerCon
                     onDragLeave={model.dragLeave}
                     onClick={model.iconClick}
                   >
-                    <ProjectAvatar
+                    <ProjectIcon
+                      project={props.project}
                       fallback={model.store.name || model.defaultName()}
-                      src={getProjectAvatarSource(props.project.id, {
+                      icon={{
                         color: model.store.color,
                         url: props.project.icon?.url,
                         override: model.store.iconOverride,
-                      })}
-                      variant={getProjectAvatarVariant(model.store.color)}
+                      }}
                       class="!size-16 [&_[data-slot=project-avatar-surface]]:!rounded-[6px] [&_[data-slot=project-avatar-surface]]:!text-[32px]"
                     />
                     <span

packages/app/src/settings/workspaces/projects.tsx

@@ -1,13 +1,12 @@
 import { Component, For, Show, createMemo, createSignal } from "solid-js"
 import { IconButton } from "@opencode-ai/ui/icon-button"
 import { Icon } from "@opencode-ai/ui/icon"
-import { ProjectAvatar } from "@opencode-ai/ui/project-avatar"
 import { useDialog } from "@opencode-ai/ui/context/dialog"
 import { useLanguage } from "@/runtime/i18n/language"
 import { useGlobal } from "@/runtime/server/runtime"
-import { getProjectAvatarVariant } from "@/shell/state/layout"
 import { ServerConnection, serverName } from "@/runtime/server/registry"
 import { displayName } from "@/shell/layout/helpers"
+import { ProjectIcon } from "@/shell/layout/project-icon"
 import { InlineServerSelect } from "@/settings/server-select"
 import { DialogEditProject } from "./project-dialog"
 import "@/settings/settings.css"
@@ -18,6 +17,7 @@ export const SettingsProjects: Component = () => {
   const global = useGlobal()
   const [allServers, setAllServers] = createSignal(true)
   const selected = global.settings.server.selected
+  const multiple = createMemo(() => global.servers.list().length > 1)
   const projects = createMemo(() => {
     const server = selected()
     if (!server) return []
@@ -40,14 +40,13 @@ export const SettingsProjects: Component = () => {
 
   const ProjectRow: Component<{ project: ProjectItem; server: ServerConnection.Any }> = (props) => {
     const name = () => displayName(props.project)
-    const color = () => getProjectAvatarVariant(props.project.icon?.color)
     return (
       <div
-        class="group flex items-center justify-between gap-5 px-4 py-2.5 rounded-lg bg-v2-background-bg-base shadow-[var(--v2-elevation-raised)] cursor-pointer transition-all hover:bg-v2-background-bg-layer-01"
+        class="group flex items-center justify-between gap-5 px-4 py-2.5 rounded-lg bg-v2-background-bg-base shadow-[var(--v2-elevation-raised)] transition-all hover:bg-v2-background-bg-layer-01"
         onClick={() => openProjectSettings(props.project, props.server)}
       >
         <div class="flex items-center gap-2.5 min-w-0 flex-1">
-          <ProjectAvatar fallback={name()} variant={color()} class="shrink-0" />
+          <ProjectIcon project={props.project} class="shrink-0" />
           <span class="text-13-medium text-v2-text-text-base truncate">{name()}</span>
         </div>
         <div class="flex items-center gap-2 shrink-0 opacity-0 group-hover:opacity-100 transition-opacity">
@@ -74,14 +73,16 @@ export const SettingsProjects: Component = () => {
             <h2 class="settings-tab-title">{language.t("settings.projects.title")}</h2>
             <span class="text-11-regular text-v2-text-text-muted">{language.t("settings.projects.description")}</span>
           </div>
-          <InlineServerSelect
-            all={{
-              label: language.t("settings.projects.server.all"),
-              selected: allServers,
-              onSelect: () => setAllServers(true),
-            }}
-            onServerSelect={() => setAllServers(false)}
-          />
+          <Show when={multiple()}>
+            <InlineServerSelect
+              all={{
+                label: language.t("settings.projects.server.all"),
+                selected: allServers,
+                onSelect: () => setAllServers(true),
+              }}
+              onServerSelect={() => setAllServers(false)}
+            />
+          </Show>
         </div>
       </div>
 
@@ -100,7 +101,14 @@ export const SettingsProjects: Component = () => {
               >
                 <Show when={selected()} keyed>
                   {(server) => (
-                    <For each={projects()}>{(project) => <ProjectRow project={project} server={server} />}</For>
+                    <div class="settings-section">
+                      <Show when={multiple()}>
+                        <h3 class="settings-section-title">{serverName(server) || ServerConnection.key(server)}</h3>
+                      </Show>
+                      <div class="flex flex-col gap-2 w-full">
+                        <For each={projects()}>{(project) => <ProjectRow project={project} server={server} />}</For>
+                      </div>
+                    </div>
                   )}
                 </Show>
               </Show>
@@ -119,9 +127,11 @@ export const SettingsProjects: Component = () => {
               <For each={groups()}>
                 {(group) => (
                   <div class="settings-section">
-                    <h3 class="settings-section-title">
-                      {serverName(group.server) || ServerConnection.key(group.server)}
-                    </h3>
+                    <Show when={multiple()}>
+                      <h3 class="settings-section-title">
+                        {serverName(group.server) || ServerConnection.key(group.server)}
+                      </h3>
+                    </Show>
                     <div class="flex flex-col gap-2 w-full">
                       <For each={group.projects}>
                         {(project) => <ProjectRow project={project} server={group.server} />}

packages/app/src/shell/errors/error.tsx

@@ -278,7 +278,7 @@ export const ErrorPage: Component<ErrorPageProps> = (props) => {
 
   return (
     <div
-      class="relative flex-1 h-screen w-screen min-h-0 overflow-y-auto flex flex-col items-center justify-start sm:justify-center p-4 sm:p-8 font-sans"
+      class="relative flex-1 h-full w-full min-h-0 min-w-0 overflow-y-auto flex flex-col items-center justify-start sm:justify-center p-4 sm:p-8 font-sans"
       data-tauri-drag-region
     >
       <div class="w-full max-w-3xl flex flex-col items-center justify-center gap-6 sm:gap-8 my-auto">

packages/app/src/shell/layout/helpers.test.ts

@@ -12,6 +12,7 @@ import {
   homeProjectDirectories,
   homeSessionServerStatus,
   latestRootSession,
+  projectForSession,
   sortedRootSessions,
   toggleHomeProjectSelection,
 } from "./helpers"
@@ -167,6 +168,27 @@ describe("layout workspace helpers", () => {
     expect(childSessionOnPath(list, "root", "other")).toBeUndefined()
   })
 
+  test("keeps the enriched workspace inventory when matching a session by project id", () => {
+    const project = {
+      id: "project",
+      worktree: "/repo",
+      sandboxes: ["/workspaces/feature"],
+      worktrees: [{ directory: "/repo" }, { directory: "/workspaces/feature", strategy: "git" }],
+    }
+
+    expect(projectForSession(session({ id: "feature", directory: "/workspaces/feature/packages/app" }), [project])).toBe(
+      project,
+    )
+  })
+
+  test("finds the enriched project for a nested workspace when its session project id is stale", () => {
+    const project = { id: "updated", worktree: "/repo", sandboxes: ["/workspaces/feature"] }
+
+    expect(projectForSession(session({ id: "feature", directory: "/workspaces/feature/packages/app" }), [project])).toBe(
+      project,
+    )
+  })
+
   test("formats fallback project display name", () => {
     expect(displayName({ worktree: "/tmp/app" })).toBe("app")
     expect(displayName({ worktree: "/tmp/app", name: "My App" })).toBe("My App")

packages/app/src/shell/layout/helpers.ts

@@ -1,6 +1,7 @@
 import { getFilename } from "@opencode-ai/util/path"
 import type { SessionInfo } from "@opencode-ai/client/promise"
 import { pathKey } from "@/workspaces/path-key"
+import { isProjectDirectory } from "@/workspaces/paths"
 import type { ServerConnection } from "@/runtime/server/registry"
 import type { HomeProjectSelection } from "@/shell/state/layout"
 
@@ -100,11 +101,7 @@ export function projectForSession<T extends { id?: string; worktree: string; san
 ) {
   const direct = byID.get(session.projectID)
   if (direct) return direct
-  const directory = pathKey(session.location.directory)
-  return projects.find(
-    (project) =>
-      pathKey(project.worktree) === directory || project.sandboxes?.some((sandbox) => pathKey(sandbox) === directory),
-  )
+  return projects.find((project) => isProjectDirectory(project, session.location.directory))
 }
 
 export const errorMessage = (err: unknown, fallback: string) => {

packages/app/src/shell/layout/project-icon.tsx

@@ -0,0 +1,24 @@
+import { ProjectAvatar, type ProjectAvatarProps } from "@opencode-ai/ui/project-avatar"
+import { splitProps } from "solid-js"
+import { displayName, getProjectAvatarSource } from "@/shell/layout/helpers"
+import { getProjectAvatarVariant, type LocalProject } from "@/shell/state/layout"
+
+type ProjectIconProps = Omit<ProjectAvatarProps, "fallback" | "src" | "variant"> & {
+  project: LocalProject
+  fallback?: string
+  icon?: LocalProject["icon"]
+}
+
+export function ProjectIcon(props: ProjectIconProps) {
+  const [local, rest] = splitProps(props, ["project", "fallback", "icon"])
+  const icon = () => local.icon ?? local.project.icon
+
+  return (
+    <ProjectAvatar
+      {...rest}
+      fallback={local.fallback ?? displayName(local.project)}
+      src={getProjectAvatarSource(local.project.id, icon())}
+      variant={getProjectAvatarVariant(icon()?.color)}
+    />
+  )
+}

packages/app/src/shell/routes/routes.tsx

@@ -6,6 +6,7 @@ import { useGlobal } from "@/runtime/server/runtime"
 import { ServerConnection } from "@/runtime/server/registry"
 import { SessionPanelFrame, SessionRouteFrame } from "@/session/session-frame"
 import { LayoutProvider } from "@/shell/state/layout"
+import { SettingsSurfaceProvider } from "@/settings/surface"
 import Shell from "@/shell/shell"
 import { requireServerKey } from "./session"
 
@@ -69,7 +70,9 @@ function TargetServerRoute(props: ParentProps) {
 function AppLayout(props: ParentProps) {
   return (
     <LayoutProvider>
-      <Shell>{props.children}</Shell>
+      <SettingsSurfaceProvider>
+        <Shell>{props.children}</Shell>
+      </SettingsSurfaceProvider>
     </LayoutProvider>
   )
 }

packages/app/src/shell/routes/session-ui-provider.tsx

@@ -6,6 +6,7 @@ import { LocalProvider } from "@/providers/models/selection"
 import type { ServerConnection } from "@/runtime/server/registry"
 import { sessionHref } from "@/shell/routes/session"
 import { useData } from "@/runtime/server/current"
+import { useServerSDK } from "@/runtime/server/client"
 import { useTabs } from "@/shell/tabs/tabs"
 
 export function SessionUIProvider(
@@ -17,6 +18,7 @@ export function SessionUIProvider(
   const navigate = useNavigate()
   const params = useParams()
   const data = useData()
+  const serverSDK = useServerSDK()
   const tabs = useTabs()
   const directory = () => props.directory
   const href = (sessionID: string) => sessionHref(props.server, sessionID)
@@ -53,6 +55,7 @@ export function SessionUIProvider(
       data={sessionUIData()}
       directory={directory()}
       sessionID={params.id}
+      shellOutput={(input) => serverSDK.api.shell.output(input)}
       onNavigateToSession={navigateToSession}
       onSessionHref={href}
     >

packages/app/src/shell/shell.tsx

@@ -1,15 +1,28 @@
 import { lazy, Show, Suspense, type ParentProps } from "solid-js"
 import { createStore } from "solid-js/store"
+import { createMediaQuery } from "@solid-primitives/media"
+import { ResizeHandle } from "@opencode-ai/ui/resize-handle"
 import { Titlebar, type TitlebarUpdate } from "@/shell/titlebar/titlebar"
 import { usePlatform } from "@/runtime/platform/platform"
 import { ToastRegion } from "@/shell/notifications/toast"
 import { TitlebarRightProvider } from "@/shell/titlebar/right-slot"
+import { useSettingsSurface } from "@/settings/surface"
+import { useSettings } from "@/settings/model"
 
 const DebugBar = lazy(() => import("@/shell/debug/debug-bar").then((module) => ({ default: module.DebugBar })))
+const SettingsScreen = lazy(() => import("@/settings/shell").then((module) => ({ default: module.SettingsScreen })))
 
 export default function Layout(props: ParentProps) {
   const platform = usePlatform()
-  const [state, setState] = createStore({ debugTools: false })
+  const settings = useSettingsSurface()
+  const preferences = useSettings()
+  const mobile = createMediaQuery("(max-width: 767px)")
+  const [state, setState] = createStore({
+    debugTools: false,
+    tabsWidth: 260,
+    tabsMount: undefined as HTMLElement | undefined,
+  })
+  const verticalTabs = () => preferences.appearance.tabLayout() === "vertical" && !mobile()
 
   const update: TitlebarUpdate = {
     get version() {
@@ -34,15 +47,47 @@ export default function Layout(props: ParentProps) {
       >
         <Titlebar
           update={update}
+          verticalTabs={verticalTabs() ? { mount: state.tabsMount } : undefined}
           debugTools={
             import.meta.env.DEV
               ? { visible: state.debugTools, toggle: () => setState("debugTools", (value) => !value) }
               : undefined
           }
         />
-        <main class="flex-1 min-h-0 min-w-0 overflow-x-hidden flex flex-col items-start contain-strict">
-          <Suspense>{props.children}</Suspense>
-        </main>
+        <div class="flex flex-1 min-h-0 min-w-0 flex-row">
+          <Show when={verticalTabs()}>
+            <aside
+              ref={(element) => setState("tabsMount", element)}
+              data-slot="vertical-tabs-sidebar"
+              class="relative flex min-h-0 shrink-0 flex-col bg-v2-background-bg-deep px-2.5 py-2"
+              style={{ width: `${state.tabsWidth}px` }}
+            >
+              <ResizeHandle
+                class="-end-2"
+                direction="horizontal"
+                size={state.tabsWidth}
+                min={130}
+                max={520}
+                onResize={(width) => setState("tabsWidth", width)}
+              />
+            </aside>
+          </Show>
+          <main class="flex-1 min-h-0 min-w-0 overflow-x-hidden flex flex-col items-start contain-strict">
+            <div
+              class="flex size-full min-h-0 min-w-0 flex-col"
+              hidden={settings.store.open}
+              inert={settings.store.open}
+              aria-hidden={settings.store.open}
+            >
+              <Suspense>{props.children}</Suspense>
+            </div>
+            <Show when={settings.store.open}>
+              <Suspense>
+                <SettingsScreen defaultValue={settings.store.tab} />
+              </Suspense>
+            </Show>
+          </main>
+        </div>
         <Show when={import.meta.env.DEV && state.debugTools}>
           <Suspense>
             <DebugBar inline />

packages/app/src/shell/tabs/tabs.tsx

@@ -245,6 +245,14 @@ export const { use: useTabs, provider: TabsProvider } = createSimpleContext({
           )
         })
       },
+      initializeDraftWorktrees(server: ServerConnection.Key, directory: string, worktree: string) {
+        setStore(
+          (tab) => tab.type === "draft" && tab.server === server && tab.directory === directory && !tab.worktree,
+          produce((tab) => {
+            if (tab.type === "draft") tab.worktree = worktree
+          }),
+        )
+      },
       promoteDraft(draftID: string, session: Omit<SessionTab, "type">) {
         // Keep the replacement and navigation atomic so /new-session never renders
         // after its backing draft tab has been removed from the store.

packages/app/src/shell/titlebar/tab-nav.css

@@ -15,6 +15,48 @@
   background: linear-gradient(var(--tab-overlay), var(--tab-overlay)), var(--tab-base);
 }
 
+[data-titlebar-tab][data-orientation="vertical"]:has([data-slot="project-avatar-slot"]) {
+  height: 45px;
+}
+
+[data-titlebar-tab][data-orientation="vertical"]:has([data-slot="project-avatar-slot"]) [data-slot="tab-link"] {
+  display: grid;
+  grid-template-columns: 16px minmax(0, 1fr);
+  grid-template-rows: repeat(2, var(--line-height-compact));
+  align-content: center;
+  column-gap: 6px;
+  row-gap: 0;
+  font-size: 13px;
+  line-height: var(--line-height-compact);
+}
+
+[data-titlebar-tab][data-orientation="vertical"] [data-slot="project-avatar-slot"] {
+  grid-column: 1;
+  grid-row: 1;
+}
+
+[data-titlebar-tab][data-orientation="vertical"] [data-slot="tab-title"] {
+  grid-column: 2;
+  grid-row: 1;
+}
+
+[data-titlebar-tab][data-orientation="vertical"] [data-slot="tab-project"] {
+  grid-column: 2;
+  grid-row: 2;
+  min-width: 0;
+  overflow: hidden;
+  color: var(--v2-text-text-muted);
+  font-size: 13px;
+  line-height: var(--line-height-compact);
+  white-space: nowrap;
+}
+
+[data-titlebar-tab][data-orientation="vertical"] [data-slot="tab-close"] {
+  top: 0;
+  bottom: 0;
+  margin-block: auto;
+}
+
 [data-titlebar-tab]:is(:hover, :has(> [data-slot="tab-link"]:focus-visible)):not([data-state="pressed"]):not(
     [data-dragging="true"]
   ):not([data-editing="true"]) {
@@ -39,6 +81,10 @@
   gap: 6px;
 }
 
+[data-titlebar-tab-list][data-orientation="vertical"] {
+  gap: 2px;
+}
+
 [data-titlebar-tab-slot] {
   --tab-separator: var(--v2-background-bg-layer-03);
   position: relative;
@@ -68,6 +114,11 @@
   display: none;
 }
 
+[data-titlebar-tab-list][data-orientation="vertical"] [data-titlebar-tab-slot]::before,
+[data-titlebar-tab-slot]:has([data-titlebar-tab][data-orientation="vertical"])::before {
+  display: none;
+}
+
 [data-titlebar-tab][data-title-overflow="true"]:not([data-editing="true"]) [data-slot="tab-link"],
 [data-titlebar-tab]:is(:hover, [data-active="true"]):not([data-editing="true"]) [data-slot="tab-link"] {
   --tab-title-fade-offset: 4px;
@@ -133,21 +184,21 @@
 }
 
 @container (max-width: 64px) {
-  [data-titlebar-tab-link] {
+  [data-titlebar-tab]:not([data-orientation="vertical"]) [data-titlebar-tab-link] {
     justify-content: center;
     gap: 0;
     padding-inline: 0;
   }
 
-  [data-titlebar-tab-title] {
+  [data-titlebar-tab]:not([data-orientation="vertical"]) [data-titlebar-tab-title] {
     display: none;
   }
 
-  [data-titlebar-tab]:not([data-active="true"]) [data-slot="tab-close"] {
+  [data-titlebar-tab]:not([data-orientation="vertical"]):not([data-active="true"]) [data-slot="tab-close"] {
     display: none;
   }
 
-  [data-slot="tab-close"] {
+  [data-titlebar-tab]:not([data-orientation="vertical"]) [data-slot="tab-close"] {
     right: auto;
     left: 50%;
     transform: translateX(-50%);

packages/app/src/shell/titlebar/tab-nav.tsx

@@ -33,6 +33,7 @@ export function TabNavItem(props: {
   dragging?: boolean
   pressed?: boolean
   hidden?: boolean
+  orientation?: "horizontal" | "vertical"
 }) {
   const [editing, setEditing] = createSignal(false)
   const [titleOverflowing, setTitleOverflowing] = createSignal(false)
@@ -180,6 +181,7 @@ export function TabNavItem(props: {
       }}
       data-titlebar-tab
       data-slot="titlebar-tab-item"
+      data-orientation={props.orientation ?? "horizontal"}
       data-title-overflow={titleOverflowing()}
       data-editing={editing()}
       class="group relative flex h-7 w-full min-w-0 select-none flex-row items-center gap-1.5 overflow-hidden whitespace-nowrap rounded-[6px] px-1.5 [container-type:inline-size]"
@@ -278,6 +280,13 @@ export function TabNavItem(props: {
             event.preventDefault()
           }}
         />
+        <Show when={props.orientation === "vertical" && projectName()}>
+          {(name) => (
+            <span data-slot="tab-project" dir="auto">
+              {name()}
+            </span>
+          )}
+        </Show>
       </a>
 
       <div data-slot="tab-close">
@@ -299,6 +308,7 @@ export function TabNavItem(props: {
   return (
     <TabPreviewPopover
       trigger={tab}
+      orientation={props.orientation}
       open={popoverOpen() && !previewBlocked()}
       onOpenChange={(value) => {
         if (value && previewBlocked()) return
@@ -325,6 +335,7 @@ export function DraftTabItem(props: {
   dragging?: boolean
   pressed?: boolean
   hidden?: boolean
+  orientation?: "horizontal" | "vertical"
 }) {
   const language = useLanguage()
   const closeTab = (event: MouseEvent) => {
@@ -337,6 +348,7 @@ export function DraftTabItem(props: {
       ref={(el) => forwardTabRef(props.ref, el)}
       data-titlebar-tab
       data-slot="titlebar-tab-item"
+      data-orientation={props.orientation ?? "horizontal"}
       data-active={props.active}
       data-dragging={props.dragging}
       data-state={props.active || props.pressed ? "pressed" : undefined}

packages/app/src/shell/titlebar/tab-popover.css

@@ -8,6 +8,7 @@
   z-index: 50;
   box-sizing: border-box;
   width: 256px;
+  max-width: calc(100dvw - 24px);
   display: flex;
   flex-direction: column;
   gap: 6px;

packages/app/src/shell/titlebar/tab-popover.tsx

@@ -1,9 +1,10 @@
 import { HoverCard } from "@kobalte/core/hover-card"
 import { createSignal, Show, type JSXElement } from "solid-js"
+import { useLanguage } from "@/runtime/i18n/language"
 import "./tab-popover.css"
 
 // Initial hover delay before the preview appears, per design.
-const OPEN_DELAY = 2_000
+const OPEN_DELAY = 750
 // Mouse-out delay: begin closing immediately (a brief exit animation plays).
 const CLOSE_DELAY = 0
 // After a preview closes, hovering a neighbouring tab within this window skips
@@ -24,7 +25,9 @@ export function TabPreviewPopover(props: {
   open: boolean
   onOpenChange: (open: boolean) => void
   data: TabPreviewData
+  orientation?: "horizontal" | "vertical"
 }) {
+  const language = useLanguage()
   let triggerEl: HTMLDivElement | undefined
   // When opened during a rapid tab-hopping streak, this preview appears and
   // disappears instantly (no repeated enter/exit animation) — only the first,
@@ -50,7 +53,13 @@ export function TabPreviewPopover(props: {
       // The preview is non-interactive (pointer-events: none), so there is no
       // safe area to traverse — leaving the tab hides it immediately.
       ignoreSafeArea
-      placement="bottom-start"
+      placement={
+        props.orientation === "vertical"
+          ? language.direction() === "rtl"
+            ? "left-start"
+            : "right-start"
+          : "bottom-start"
+      }
       gutter={6}
     >
       <HoverCard.Trigger ref={triggerEl} as="div" data-component="session-tab-popover-trigger" tabIndex={-1}>

packages/app/src/shell/titlebar/tab-strip.tsx

@@ -4,7 +4,7 @@ import { createResizeObserver } from "@solid-primitives/resize-observer"
 import { DragDropProvider, PointerSensor } from "@dnd-kit/solid"
 import { isSortable, useSortable } from "@dnd-kit/solid/sortable"
 import { Accessibility, AutoScroller, Feedback, PointerActivationConstraints } from "@dnd-kit/dom"
-import { RestrictToHorizontalAxis } from "@dnd-kit/abstract/modifiers"
+import { RestrictToHorizontalAxis, RestrictToVerticalAxis } from "@dnd-kit/abstract/modifiers"
 import { RestrictToElement } from "@dnd-kit/dom/modifiers"
 import { arrayMove } from "@dnd-kit/helpers"
 import { tabHref, tabKey, type SessionTab, type Tab } from "@/shell/tabs/tabs"
@@ -27,6 +27,7 @@ function SessionTabSlot(props: {
   index: number
   active: boolean
   forceTruncate: boolean
+  orientation: "horizontal" | "vertical"
   session: SessionInfo | undefined
   fallbackTitle?: string
   onRename: (title: string) => Promise<void>
@@ -49,7 +50,12 @@ function SessionTabSlot(props: {
       data-titlebar-tab-slot
       data-tab-key={props.id}
       data-active={props.active}
-      class="relative flex w-56 min-w-7 max-w-56 flex-shrink"
+      data-orientation={props.orientation}
+      class="relative flex"
+      classList={{
+        "w-56 min-w-7 max-w-56 flex-shrink": props.orientation === "horizontal",
+        "w-full shrink-0": props.orientation === "vertical",
+      }}
     >
       <TabNavItem
         ref={(el) => {
@@ -65,6 +71,7 @@ function SessionTabSlot(props: {
         active={props.active}
         forceTruncate={props.forceTruncate}
         dragging={sortable.isDragSource()}
+        orientation={props.orientation}
       />
     </div>
   )
@@ -76,6 +83,7 @@ function SessionTabEntry(props: {
   index: number
   active: boolean
   forceTruncate: boolean
+  orientation: "horizontal" | "vertical"
   serverCtx: ServerCtx | undefined
   onVisibleChange: (visible: boolean) => void
   onNavigate: (element: HTMLDivElement) => void
@@ -160,6 +168,7 @@ function SessionTabEntry(props: {
         index={props.index}
         active={props.active}
         forceTruncate={props.forceTruncate}
+        orientation={props.orientation}
         session={session()}
         fallbackTitle={persisted()?.title ?? (missingSession() ? language.t("session.tab.unknown") : undefined)}
         onRename={rename}
@@ -175,6 +184,7 @@ function DraftTabSlot(props: {
   id: string
   index: number
   active: boolean
+  orientation: "horizontal" | "vertical"
   title: string
   onNavigate: (element: HTMLDivElement) => void
   onClose: () => void
@@ -195,7 +205,12 @@ function DraftTabSlot(props: {
       data-titlebar-tab-slot
       data-tab-key={props.id}
       data-active={props.active}
-      class="relative flex w-56 min-w-7 max-w-56 flex-shrink"
+      data-orientation={props.orientation}
+      class="relative flex"
+      classList={{
+        "w-56 min-w-7 max-w-56 flex-shrink": props.orientation === "horizontal",
+        "w-full shrink-0": props.orientation === "vertical",
+      }}
     >
       <DraftTabItem
         ref={(el) => {
@@ -207,12 +222,14 @@ function DraftTabSlot(props: {
         onClose={props.onClose}
         active={props.active}
         dragging={sortable.isDragSource()}
+        orientation={props.orientation}
       />
     </div>
   )
 }
 
 export function TitlebarTabStrip(props: {
+  orientation?: "horizontal" | "vertical"
   tabs: Tab[]
   currentTab: Tab | undefined
   forceTruncate: boolean
@@ -224,6 +241,7 @@ export function TitlebarTabStrip(props: {
   const global = useGlobal()
   const language = useLanguage()
   const command = useCommand()
+  const vertical = () => props.orientation === "vertical"
   let scrollRef!: HTMLDivElement
   let listRef!: HTMLDivElement
   let resizeFrame: number | undefined
@@ -259,7 +277,9 @@ export function TitlebarTabStrip(props: {
 
   function refreshOverflow() {
     if (!scrollRef) return
-    props.onOverflowChange(scrollRef.scrollWidth > scrollRef.clientWidth)
+    props.onOverflowChange(
+      vertical() ? scrollRef.scrollHeight > scrollRef.clientHeight : scrollRef.scrollWidth > scrollRef.clientWidth,
+    )
   }
 
   createResizeObserver(
@@ -288,10 +308,19 @@ export function TitlebarTabStrip(props: {
   })
 
   return (
-    <div data-slot="titlebar-tabs" class="relative min-w-0">
+    <div
+      data-slot={vertical() ? "vertical-tabs" : "titlebar-tabs"}
+      data-orientation={vertical() ? "vertical" : "horizontal"}
+      class="relative min-w-0"
+      classList={{ "min-h-0 overflow-hidden": vertical() }}
+    >
       <div
-        data-slot="titlebar-tabs-scroll"
-        class="flex min-w-0 flex-row items-center gap-1.5 overflow-x-auto no-scrollbar [app-region:no-drag]"
+        data-slot={vertical() ? "vertical-tabs-scroll" : "titlebar-tabs-scroll"}
+        class="flex min-w-0 no-scrollbar [app-region:no-drag]"
+        classList={{
+          "flex-row items-center gap-1.5 overflow-x-auto": !vertical(),
+          "max-h-full flex-col overflow-y-auto overflow-x-hidden": vertical(),
+        }}
         ref={scrollRef}
       >
         <DragDropProvider
@@ -304,10 +333,13 @@ export function TitlebarTabStrip(props: {
                 (event.target instanceof Element && !!event.target.closest('[contenteditable="true"]')),
             }),
           ]}
-          modifiers={[RestrictToHorizontalAxis, RestrictToElement.configure({ element: () => listRef })]}
+          modifiers={[
+            vertical() ? RestrictToVerticalAxis : RestrictToHorizontalAxis,
+            RestrictToElement.configure({ element: () => listRef }),
+          ]}
           plugins={(defaults) => [
             ...defaults.filter((plugin) => plugin !== Accessibility),
-            AutoScroller.configure({ acceleration: 8, threshold: { x: 0.05, y: 0 } }),
+            AutoScroller.configure({ acceleration: 8, threshold: vertical() ? { x: 0, y: 0.05 } : { x: 0.05, y: 0 } }),
             Feedback.configure({ dropAnimation: null }),
           ]}
           onDragStart={(event) => {
@@ -335,7 +367,13 @@ export function TitlebarTabStrip(props: {
             }
           }}
         >
-          <div data-titlebar-tab-list class="flex w-full min-w-0 flex-row items-center" ref={listRef}>
+          <div
+            data-titlebar-tab-list
+            data-orientation={vertical() ? "vertical" : "horizontal"}
+            class="flex w-full min-w-0"
+            classList={{ "flex-row items-center": !vertical(), "flex-col items-stretch": vertical() }}
+            ref={listRef}
+          >
             <For each={props.tabs}>
               {(tab) => {
                 const id = tabKey(tab)
@@ -355,6 +393,7 @@ export function TitlebarTabStrip(props: {
                       index={visibleIndex()}
                       active={props.currentTab === tab}
                       forceTruncate={props.forceTruncate}
+                      orientation={vertical() ? "vertical" : "horizontal"}
                       serverCtx={serverCtx()}
                       onVisibleChange={(visible) => setVisibility(id, visible)}
                       onNavigate={(element) => {
@@ -372,6 +411,7 @@ export function TitlebarTabStrip(props: {
                     id={id}
                     index={visibleIndex()}
                     active={props.currentTab === tab}
+                    orientation={vertical() ? "vertical" : "horizontal"}
                     title={language.t("command.session.new")}
                     onNavigate={(element) => {
                       ref = element
@@ -385,16 +425,18 @@ export function TitlebarTabStrip(props: {
           </div>
         </DragDropProvider>
       </div>
-      <div
-        data-slot="titlebar-tabs-fade-left"
-        aria-hidden="true"
-        class="pointer-events-none absolute inset-y-0 left-0 z-10 w-6 bg-[linear-gradient(to_right,var(--v2-background-bg-deep),transparent)]"
-      />
-      <div
-        data-slot="titlebar-tabs-fade-right"
-        aria-hidden="true"
-        class="pointer-events-none absolute inset-y-0 right-0 z-10 w-6 bg-[linear-gradient(to_left,var(--v2-background-bg-deep),transparent)]"
-      />
+      <Show when={!vertical()}>
+        <div
+          data-slot="titlebar-tabs-fade-left"
+          aria-hidden="true"
+          class="pointer-events-none absolute inset-y-0 left-0 z-10 w-6 bg-[linear-gradient(to_right,var(--v2-background-bg-deep),transparent)]"
+        />
+        <div
+          data-slot="titlebar-tabs-fade-right"
+          aria-hidden="true"
+          class="pointer-events-none absolute inset-y-0 right-0 z-10 w-6 bg-[linear-gradient(to_left,var(--v2-background-bg-deep),transparent)]"
+        />
+      </Show>
     </div>
   )
 }

packages/app/src/shell/titlebar/titlebar.tsx

@@ -1,5 +1,6 @@
 import { createEffect, createMemo, createResource, Match, createSignal, Show, Switch, untrack } from "solid-js"
 import { createStore } from "solid-js/store"
+import { Portal } from "solid-js/web"
 import { useLocation, useNavigate } from "@solidjs/router"
 import { IconButton } from "@opencode-ai/ui/icon-button"
 import { Icon } from "@opencode-ai/ui/icon"
@@ -36,7 +37,11 @@ export type TitlebarUpdate = {
   install: () => void
 }
 
-export function Titlebar(props: { update?: TitlebarUpdate; debugTools?: { visible: boolean; toggle: () => void } }) {
+export function Titlebar(props: {
+  update?: TitlebarUpdate
+  debugTools?: { visible: boolean; toggle: () => void }
+  verticalTabs?: { mount?: HTMLElement }
+}) {
   const platform = usePlatform()
   const command = useCommand()
   const language = useLanguage()
@@ -54,7 +59,7 @@ export function Titlebar(props: { update?: TitlebarUpdate; debugTools?: { visibl
   const titlebarZoom = () => (windows() ? Math.max(zoom(), minTitlebarZoom) : zoom())
   const minHeight = () => {
     if (mac()) return `${titlebarHeight / zoom()}px`
-    if (windows()) return `${titlebarHeight / Math.min(titlebarZoom(), 1)}px`
+    if (windows()) return `env(titlebar-area-height, ${titlebarHeight / Math.min(titlebarZoom(), 1)}px)`
     return undefined
   }
   const windowsControlsWidth = () => `${windowsControlsBaseWidth / Math.max(titlebarZoom(), 1)}px`
@@ -324,7 +329,7 @@ export function Titlebar(props: { update?: TitlebarUpdate; debugTools?: { visibl
               <div
                 class="h-full flex-1 overflow-hidden flex flex-row items-center gap-1.5 px-2 md:pr-3"
                 classList={{
-                  "pt-2": !bottom(),
+                  "pt-2": !bottom() && !windows(),
                   "pb-2": bottom(),
                   "md:pl-2": macTrafficLights(),
                   "md:pl-4": !macTrafficLights(),
@@ -357,40 +362,82 @@ export function Titlebar(props: { update?: TitlebarUpdate; debugTools?: { visibl
                   />
                 </Tooltip>
 
-                <TitlebarTabStrip
-                  tabs={tabsStore}
-                  currentTab={currentTab()}
-                  forceTruncate={tabsAreOverflowing()}
-                  onOverflowChange={setTabsAreOverflowing}
-                  onNavigate={(tab, el) => {
-                    tabs.select(tab)
-                    el?.scrollIntoView({ behavior: "instant" })
-                  }}
-                  onClose={(tab) => {
-                    const index = tabsStore.findIndex((item) => tabKey(item) === tabKey(tab))
-                    if (index !== -1) tabsStoreActions.closeTab(index)
-                  }}
-                  onReorder={(keys) => tabsStoreActions.reorder(keys)}
-                />
-                <Tooltip
-                  placement="bottom"
-                  value={
+                <Show
+                  when={props.verticalTabs}
+                  fallback={
                     <>
-                      {language.t("command.session.new")}
-                      <Keybind keys={newTabTooltipKeybind(command)} variant="neutral" />
+                      <TitlebarTabStrip
+                        tabs={tabsStore}
+                        currentTab={currentTab()}
+                        forceTruncate={tabsAreOverflowing()}
+                        onOverflowChange={setTabsAreOverflowing}
+                        onNavigate={(tab, el) => {
+                          tabs.select(tab)
+                          el?.scrollIntoView({ behavior: "instant" })
+                        }}
+                        onClose={(tab) => {
+                          const index = tabsStore.findIndex((item) => tabKey(item) === tabKey(tab))
+                          if (index !== -1) tabsStoreActions.closeTab(index)
+                        }}
+                        onReorder={(keys) => tabsStoreActions.reorder(keys)}
+                      />
+                      <Tooltip
+                        placement="bottom"
+                        value={
+                          <>
+                            {language.t("command.session.new")}
+                            <Keybind keys={newTabTooltipKeybind(command)} variant="neutral" />
+                          </>
+                        }
+                      >
+                        <IconButton
+                          type="button"
+                          variant="ghost-muted"
+                          size="large"
+                          class="shrink-0"
+                          icon={<Icon name="plus" />}
+                          onClick={openNewTab}
+                          aria-label={language.t("command.session.new")}
+                        />
+                      </Tooltip>
                     </>
                   }
                 >
-                  <IconButton
-                    type="button"
-                    variant="ghost-muted"
-                    size="large"
-                    class="shrink-0"
-                    icon={<Icon name="plus" />}
-                    onClick={openNewTab}
-                    aria-label={language.t("command.session.new")}
-                  />
-                </Tooltip>
+                  {(vertical) => (
+                    <Show when={vertical().mount} keyed>
+                      {(mount) => (
+                        <Portal mount={mount}>
+                          <TitlebarTabStrip
+                            orientation="vertical"
+                            tabs={tabsStore}
+                            currentTab={currentTab()}
+                            forceTruncate={false}
+                            onOverflowChange={setTabsAreOverflowing}
+                            onNavigate={(tab, el) => {
+                              tabs.select(tab)
+                              el?.scrollIntoView({ behavior: "instant", block: "nearest" })
+                            }}
+                            onClose={(tab) => {
+                              const index = tabsStore.findIndex((item) => tabKey(item) === tabKey(tab))
+                              if (index !== -1) tabsStoreActions.closeTab(index)
+                            }}
+                            onReorder={(keys) => tabsStoreActions.reorder(keys)}
+                          />
+                          <button
+                            type="button"
+                            data-action="vertical-tabs-new-session"
+                            class="mt-1 flex h-7 w-full shrink-0 items-center gap-1.5 rounded-[6px] px-1.5 text-[13px] leading-4 text-v2-text-text-faint hover:bg-v2-background-bg-layer-02 hover:text-v2-text-text-base"
+                            onClick={openNewTab}
+                            aria-label={language.t("command.session.new")}
+                          >
+                            <Icon name="plus" />
+                            {language.t("command.session.new")}
+                          </button>
+                        </Portal>
+                      )}
+                    </Show>
+                  )}
+                </Show>
                 <div class="flex-1" />
                 <TitlebarRight state={rightState()} />
               </div>

packages/app/src/workspaces/paths.test.ts

@@ -8,6 +8,7 @@ import {
   mergeWorkspaceSessionInventory,
   sessionsForWorkspace,
   workspaceInventory,
+  workspaceSelectionDestination,
 } from "./paths"
 
 describe("isWorkspaceDirectory", () => {
@@ -41,6 +42,19 @@ describe("isWorkspaceSelection", () => {
   })
 })
 
+describe("workspaceSelectionDestination", () => {
+  test("preserves local intent for the main selection and project root", () => {
+    expect(workspaceSelectionDestination("main", "/repo")).toBe("main")
+    expect(workspaceSelectionDestination("/repo/", "/repo")).toBe("main")
+    expect(workspaceSelectionDestination("c:\\repo\\", "C:\\repo")).toBe("main")
+  })
+
+  test("preserves workspace intent without carrying project-specific paths", () => {
+    expect(workspaceSelectionDestination("create", "/repo")).toBe("create")
+    expect(workspaceSelectionDestination("/workspaces/feature", "/repo")).toBe("create")
+  })
+})
+
 test("groups and filters workspace inventory by project", () => {
   const inventory = workspaceInventory([
     {

packages/app/src/workspaces/paths.ts

@@ -106,6 +106,10 @@ export function isWorkspaceSelection(project: WorkspaceProject | undefined, sele
   return isWorkspaceDirectory(project, selection)
 }
 
+export function workspaceSelectionDestination(selection: string, projectWorktree: string) {
+  return selection === "main" || sameDirectory(selection, projectWorktree) ? "main" : "create"
+}
+
 export function workspaceDefaultSelection(
   setting: WorkspaceDefaultDestination,
   lastUsed: WorkspaceLastUsed | undefined,

packages/app/test-browser/solid-virtual.test.ts

@@ -113,6 +113,32 @@ test("reactive count updates preserve measured row sizes", () => {
   })
 })
 
+test("explicit measurement refreshes a cached row size with a custom measurer", () => {
+  const root = document.createElement("div")
+  const element = document.createElement("div")
+  element.dataset.index = "0"
+  Object.defineProperty(element, "offsetHeight", { value: 120 })
+
+  const virtualizer = new Virtualizer<HTMLDivElement, HTMLDivElement>({
+    count: 1,
+    estimateSize: () => 60,
+    initialRect: { width: 400, height: 200 },
+    getScrollElement: () => root,
+    scrollToFn: () => {},
+    observeElementRect: () => {},
+    observeElementOffset: () => {},
+    measureElement: (node) => node.offsetHeight,
+  })
+
+  virtualizer.getTotalSize()
+  virtualizer.resizeItem(0, 60)
+  virtualizer._willUpdate()
+  virtualizer.measureElement(element)
+
+  expect(virtualizer.itemSizeCache.get(0)).toBe(120)
+  expect(virtualizer.getTotalSize()).toBe(120)
+})
+
 test("initial rect projects rows before a scroll element connects", () => {
   createRoot((dispose) => {
     const virtualizer = createVirtualizer<HTMLDivElement, HTMLDivElement>({

packages/cli/src/commands/commands.ts

@@ -181,7 +181,7 @@ const Root = Spec.make(typeof OPENCODE_CLI_NAME === "string" ? OPENCODE_CLI_NAME
         Spec.make("add", {
           description: "Install a plugin and add it to the global configuration",
           params: {
-            package: Argument.string("package").pipe(Argument.withDescription("npm registry package specifier")),
+            package: Argument.string("package").pipe(Argument.withDescription("npm registry or Git package specifier")),
           },
         }),
         Spec.make("remove", {