refactor(ai): namespace model metadata by route ID

2026-09-07 09:26:26 +00:00 · 2026-09-07 00:48:56 -05:00
65 changed files with 739 additions and 570 deletions
@@ -81,6 +81,8 @@ export const route = Route.make({

 Route defaults are request-shaping defaults such as `headers`, `limits`, `generation`, `providerOptions`, and `http`. Endpoint host/query belongs on the route endpoint. Selected `LanguageModel` values carry only model id, provider id, and the configured route value. Model capability/catalog metadata lives outside this package; protocol support is enforced by request lowering and typed `AIError`s.

+LLM `ProviderMetadata` is indexed by the selected `LanguageModel.route.id`. Use that same route ID when emitting usage, reasoning signatures, item IDs, and hosted-tool state and when reading them for continuation. There is no independently configurable metadata namespace. Reassigning a provider does not rename the route or its metadata; changing the route ID changes both. Provider-specific APIs with different replay formats must have distinct route IDs.
+
 The four-axis decomposition is the reason DeepSeek, TogetherAI, Cerebras, Baseten, Fireworks, and DeepInfra all reuse `OpenAIChat.protocol` verbatim — each provider owns a small `Route.make(...)` composition instead of a protocol clone. Bug fixes in one protocol propagate to every consumer of that protocol in a single commit.

 When a provider supports multiple physical transports, selection remains execution policy below its semantic route. `OpenResponsesChannel.transport(...)` owns the provider-neutral Responses WebSocket concept: it prepares one final request, executes HTTP by default, strips WebSocket-disallowed fields, and passes a generic channel exchange to a per-call `WebSocketChannelExecutor` when supplied. Provider-specific Responses routes opt in with handshake and connection-age policy. `Route.streamPrepared` owns decoding and acknowledges channel completion only after successful full consumption.
@@ -534,6 +534,35 @@ Additional provider entrypoints include:
 - `@opencode-ai/ai/providers/cloudflare-ai-gateway`
 - `@opencode-ai/ai/providers/cloudflare-workers-ai`

+### Route-keyed metadata
+
+LLM messages, events, and usage retain provider-specific information under `providerMetadata[model.route.id]`. The provider ID identifies the provider; the route ID identifies its API and owns the metadata needed to replay that API's responses.
+
+```ts
+const model = GoogleVertexMessages.configure({ project, accessToken }).model("claude-sonnet-4-5")
+const response = yield * LLM.generate(LLM.request({ model, prompt: "Think through this problem." }))
+const reasoning = response.message.content.find((part) => part.type === "reasoning")
+const signature = reasoning?.providerMetadata?.[model.route.id]?.signature
+// model.route.id === "google-vertex-messages"
+```
+
+The former `providerMetadataKey` route option has been removed. Existing application-owned LLM metadata must use route keys instead of provider names, for example:
+
+| API                                                                | Metadata key                                                                                                 |
+| ------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------ |
+| OpenAI Chat / Responses                                            | `openai-chat` / `openai-responses`                                                                           |
+| Azure Chat / Responses                                             | `azure-openai-chat` / `azure-openai-responses`                                                               |
+| Anthropic Messages                                                 | `anthropic-messages`                                                                                         |
+| Google Gemini                                                      | `gemini`                                                                                                     |
+| Vertex Gemini / Chat / Responses / Messages                        | `google-vertex-gemini` / `google-vertex-chat` / `google-vertex-responses` / `google-vertex-messages`         |
+| Bedrock Converse / Mantle Chat / Mantle Responses                  | `bedrock-converse` / `bedrock-mantle-chat` / `bedrock-mantle-responses`                                      |
+| Groq / Mistral                                                     | `groq-chat` / `mistral-chat`                                                                                 |
+| xAI Chat / Responses                                               | `xai-chat` / `xai-responses`                                                                                 |
+| Baseten / Cerebras / DeepInfra / DeepSeek / Fireworks / TogetherAI | `baseten-chat` / `cerebras-chat` / `deepinfra-chat` / `deepseek-chat` / `fireworks-chat` / `togetherai-chat` |
+| Generic Chat / Responses                                           | `openai-compatible-chat` / `openai-compatible-responses`                                                     |
+
+OpenRouter and Cloudflare already use matching provider and route IDs. Generic Messages uses `anthropic-messages`. Custom routes use their own `id`.
+
 ## Provider options & HTTP overlays

 Request options in order of stability:
@@ -455,7 +455,7 @@ type AnthropicEvent = Schema.Schema.Type<typeof AnthropicEvent>
 interface ParserState {
  readonly provider: LLMRequest["model"]["provider"]
  readonly compactions: Readonly<Record<number, string | null>>
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly tools: ToolStream.State<number>
  readonly reasoningSignatures: Readonly<Record<number, string>>
  readonly usage?: Usage
@@ -563,14 +563,14 @@ const serverToolResultType = (name: string): AnthropicServerToolResultType | und

 const lowerServerToolResult = Effect.fn("AnthropicMessages.lowerServerToolResult")(function* (
  part: ToolResultPart,
-  providerMetadataKey: string,
+  routeID: string,
 ) {
  const wireType = serverToolResultType(part.name)
  if (!wireType)
    return yield* invalid(`Anthropic Messages does not know how to round-trip server tool result for ${part.name}`)
  // Prefer the provider-owned replay payload; fall back to the result value for
  // histories constructed directly from provider events.
-  const payload = part.providerMetadata?.[providerMetadataKey]?.["result"] ?? part.result.value
+  const payload = part.providerMetadata?.[routeID]?.["result"] ?? part.result.value
  return {
    type: wireType,
    tool_use_id: scrubToolCallID(part.id),
@@ -857,7 +857,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
  breakpoints: Cache.Breakpoints,
 ) {
  const messages: AnthropicMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
+  const routeID = request.model.route.id

  for (const [index, message] of request.messages.entries()) {
    if (message.role === "system") {
@@ -911,8 +911,8 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
        if (part.type === "reasoning") {
          // A signature marks visible thinking; only signature-less parts carrying
          // redactedData round-trip as opaque redacted_thinking blocks.
-          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata, providerMetadataKey)
-          const redactedData = redactedDataFromMetadata(part.providerMetadata, providerMetadataKey)
+          const signature = part.encrypted ?? signatureFromMetadata(part.providerMetadata, routeID)
+          const redactedData = redactedDataFromMetadata(part.providerMetadata, routeID)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ type: "redacted_thinking", data: redactedData })
            continue
@@ -941,7 +941,7 @@ const lowerMessages = Effect.fn("AnthropicMessages.lowerMessages")(function* (
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted) {
-          content.push(yield* lowerServerToolResult(part, providerMetadataKey))
+          content.push(yield* lowerServerToolResult(part, routeID))
          continue
        }
        return yield* invalid(
@@ -1148,7 +1148,7 @@ const mapFinishReason = (reason: string | null | undefined): FinishReason => {
 // inclusive `inputTokens` the rest of the contract expects. Extended
 // thinking tokens are included in `output_tokens`; newer responses also
 // expose that subset through `output_tokens_details.thinking_tokens`.
-const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: AnthropicUsage | undefined, routeID: string): Usage | undefined => {
  if (!usage) return undefined
  const iterations = usage.iterations?.length ? usage.iterations : [usage]
  const last = usage.iterations?.at(-1)
@@ -1175,7 +1175,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: ProviderShared.sumTokens(...iterations.map((item) => item.output_tokens_details?.thinking_tokens)),
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { [routeID]: usage },
  })
 }

@@ -1184,7 +1184,7 @@ const mapUsage = (usage: AnthropicUsage | undefined, providerMetadataKey: string
 // field prefers `right` when defined, falls back to `left`. `inputTokens` is
 // recomputed from the merged breakdown so the inclusive total stays
 // consistent with `nonCached + cacheRead + cacheWrite`.
-const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerMetadataKey: string) => {
+const mergeUsage = (left: Usage | undefined, right: Usage | undefined, routeID: string) => {
  if (!left) return right
  if (!right) return left
  const nonCachedInputTokens = right.nonCachedInputTokens ?? left.nonCachedInputTokens
@@ -1203,9 +1203,7 @@ const mergeUsage = (left: Usage | undefined, right: Usage | undefined, providerM
    reasoningTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, outputTokens, undefined),
    providerMetadata: {
-      [providerMetadataKey]:
-        mergeJsonRecords(left.providerMetadata?.[providerMetadataKey], right.providerMetadata?.[providerMetadataKey]) ??
-        {},
+      [routeID]: mergeJsonRecords(left.providerMetadata?.[routeID], right.providerMetadata?.[routeID]) ?? {},
    },
  })
 }
@@ -1223,7 +1221,7 @@ const SERVER_TOOL_RESULT_NAMES: Record<AnthropicServerToolResultType, string> =

 const isServerToolResultType = (type: string): type is AnthropicServerToolResultType => type in SERVER_TOOL_RESULT_NAMES

-const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey: string): LLMEvent | undefined => {
+const serverToolResultEvent = (block: AnthropicStreamBlock, routeID: string): LLMEvent | undefined => {
  if (!block.type || !isServerToolResultType(block.type)) return undefined
  const errorPayload =
    typeof block.content === "object" && block.content !== null && "type" in block.content
@@ -1237,7 +1235,7 @@ const serverToolResultEvent = (block: AnthropicStreamBlock, providerMetadataKey:
    providerExecuted: true,
    // The complete payload is irreducible provider replay state: subsequent
    // stateless requests must round-trip the typed result block verbatim.
-    providerMetadata: providerMetadata(providerMetadataKey, { blockType: block.type, result: block.content }),
+    providerMetadata: providerMetadata(routeID, { blockType: block.type, result: block.content }),
  })
 }

@@ -1246,8 +1244,8 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []

 const onMessageStart = (state: ParserState, event: AnthropicEvent): StepResult => {
-  const usage = mapUsage(event.message?.usage, state.providerMetadataKey)
-  return [usage ? { ...state, usage: mergeUsage(state.usage, usage, state.providerMetadataKey) } : state, NO_EVENTS]
+  const usage = mapUsage(event.message?.usage, state.routeID)
+  return [usage ? { ...state, usage: mergeUsage(state.usage, usage, state.routeID) } : state, NO_EVENTS]
 }

 const onContentBlockStart = (
@@ -1299,9 +1297,7 @@ const onContentBlockStart = (
    const events: LLMEvent[] = []
    const id = `reasoning-${event.index ?? 0}`
    const metadata =
-      block.signature === undefined
-        ? undefined
-        : providerMetadata(state.providerMetadataKey, { signature: block.signature })
+      block.signature === undefined ? undefined : providerMetadata(state.routeID, { signature: block.signature })
    const lifecycle = Lifecycle.reasoningStart(state.lifecycle, events, id, metadata)
    return [
      {
@@ -1330,14 +1326,14 @@ const onContentBlockStart = (
          state.lifecycle,
          events,
          `reasoning-${event.index ?? 0}`,
-          providerMetadata(state.providerMetadataKey, { redactedData: block.data }),
+          providerMetadata(state.routeID, { redactedData: block.data }),
        ),
      },
      events,
    ]
  }

-  const result = serverToolResultEvent(block, state.providerMetadataKey)
+  const result = serverToolResultEvent(block, state.routeID)
  if (!result) return [state, NO_EVENTS]
  const events: LLMEvent[] = []
  return [{ ...state, lifecycle: Lifecycle.stepStart(state.lifecycle, events) }, [...events, result]]
@@ -1438,7 +1434,7 @@ const onContentBlockStop = Effect.fn("AnthropicMessages.onContentBlockStop")(fun
        Lifecycle.textEnd(state.lifecycle, events, `text-${event.index}`),
        events,
        `reasoning-${event.index}`,
-        signature === undefined ? undefined : providerMetadata(state.providerMetadataKey, { signature }),
+        signature === undefined ? undefined : providerMetadata(state.routeID, { signature }),
      )
  events.push(...resultEvents)
  const reasoningSignatures = { ...state.reasoningSignatures }
@@ -1450,7 +1446,7 @@ const onMessageDelta = (
  state: ParserState,
  event: AnthropicEvent & { readonly delta?: AnthropicStreamDelta },
 ): StepResult => {
-  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.providerMetadataKey), state.providerMetadataKey)
+  const usage = mergeUsage(state.usage, mapUsage(event.usage, state.routeID), state.routeID)
  const pendingFinish = (() => {
    const stopReason = event.delta?.stop_reason
    if (stopReason === null || stopReason === undefined) return state.pendingFinish
@@ -1459,7 +1455,7 @@ const onMessageDelta = (
    const finishMetadata =
      stopSequence === null || stopSequence === undefined
        ? state.pendingFinish?.providerMetadata
-        : providerMetadata(state.providerMetadataKey, { stopSequence })
+        : providerMetadata(state.routeID, { stopSequence })
    return {
      reason: {
        normalized: mapFinishReason(stopReason),
@@ -1487,12 +1483,7 @@ const onMessageStop = Effect.fn("AnthropicMessages.onMessageStop")(function* (st
  events.push(...result.events)
  const closed = Object.entries(state.reasoningSignatures).reduce(
    (current, [index, signature]) =>
-      Lifecycle.reasoningEnd(
-        current,
-        events,
-        `reasoning-${index}`,
-        providerMetadata(state.providerMetadataKey, { signature }),
-      ),
+      Lifecycle.reasoningEnd(current, events, `reasoning-${index}`, providerMetadata(state.routeID, { signature })),
    lifecycle,
  )
  const finished = Lifecycle.finish(closed, events, {
@@ -1627,7 +1618,7 @@ export const protocol = Protocol.make({
    initial: (request) => ({
      provider: request.model.provider,
      compactions: {},
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+      routeID: request.model.route.id,
      tools: ToolStream.empty<number>(),
      reasoningSignatures: {},
      lifecycle: Lifecycle.initial(),
@@ -1670,7 +1661,6 @@ export const transport = <Body extends Pick<AnthropicMessagesBody, "messages" |
 export const route = Route.make({
  id: ADAPTER,
  provider: "anthropic",
-  providerMetadataKey: "anthropic",
  protocol,
  endpoint: Endpoint.path((input) => (input.request.model.provider === "anthropic" ? `${PATH}?beta=true` : PATH), {
    baseURL: DEFAULT_BASE_URL,
@@ -260,14 +260,14 @@ const lowerToolChoice = (toolChoice: NonNullable<LLMRequest["toolChoice"]>) =>

 const providerMetadata = (key: string, metadata: Record<string, unknown>): ProviderMetadata => ({ [key]: metadata })

-const reasoningSignature = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+const reasoningSignature = (part: ReasoningPart, routeID: string) => {
+  const metadata = part.providerMetadata?.[routeID]
  if (part.encrypted !== undefined) return part.encrypted
  if (ProviderShared.isRecord(metadata) && typeof metadata.signature === "string") return metadata.signature
 }

-const reasoningRedactedData = (part: ReasoningPart, providerMetadataKey: string) => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+const reasoningRedactedData = (part: ReasoningPart, routeID: string) => {
+  const metadata = part.providerMetadata?.[routeID]
  if (ProviderShared.isRecord(metadata) && typeof metadata.redactedData === "string") return metadata.redactedData
 }

@@ -324,7 +324,7 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
  breakpoints: BedrockCache.Breakpoints,
 ) {
  const messages: BedrockMessage[] = []
-  const providerMetadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
+  const routeID = request.model.route.id

  for (const message of request.messages) {
    if (message.role === "system") {
@@ -372,8 +372,8 @@ const lowerMessages = Effect.fn("BedrockConverse.lowerMessages")(function* (
          continue
        }
        if (part.type === "reasoning") {
-          const signature = reasoningSignature(part, providerMetadataKey)
-          const redactedData = reasoningRedactedData(part, providerMetadataKey)
+          const signature = reasoningSignature(part, routeID)
+          const redactedData = reasoningRedactedData(part, routeID)
          if (signature === undefined && redactedData !== undefined) {
            content.push({ reasoningContent: { redactedContent: redactedData } })
            continue
@@ -485,7 +485,7 @@ const mapFinishReason = (reason: string): FinishReason => {

 // AWS reports inputTokens separately from cache reads and writes.
 // Bedrock does not break reasoning out of outputTokens for current models.
-const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: BedrockUsageSchema | undefined, routeID: string): Usage | undefined => {
  if (!usage) return undefined
  const inputTokens = ProviderShared.sumTokens(
    usage.inputTokens,
@@ -499,12 +499,12 @@ const mapUsage = (usage: BedrockUsageSchema | undefined, providerMetadataKey: st
    cacheReadInputTokens: usage.cacheReadInputTokens,
    cacheWriteInputTokens: usage.cacheWriteInputTokens,
    totalTokens: ProviderShared.totalTokens(inputTokens, usage.outputTokens, usage.totalTokens),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { [routeID]: usage },
  })
 }

 interface ParserState {
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly tools: ToolStream.State<number>
  // Bedrock splits the finish into `messageStop` (carries `stopReason`) and
  // `metadata` (carries usage). Hold both in state so `onHalt` can emit exactly
@@ -587,8 +587,8 @@ const step = (state: ParserState, event: BedrockEvent) =>
      })()
      const redactedData = redactedChunks === undefined ? reasoning.data : encodeRedactedContent(redactedChunks)
      const metadata = (() => {
-        if (reasoning.signature) return providerMetadata(state.providerMetadataKey, { signature: reasoning.signature })
-        if (redactedData !== undefined) return providerMetadata(state.providerMetadataKey, { redactedData })
+        if (reasoning.signature) return providerMetadata(state.routeID, { signature: reasoning.signature })
+        if (redactedData !== undefined) return providerMetadata(state.routeID, { redactedData })
      })()
      const lifecycle = (() => {
        if (reasoning.text === undefined && metadata === undefined) return state.lifecycle
@@ -640,10 +640,10 @@ const step = (state: ParserState, event: BedrockEvent) =>
        if (resultEvents.length) return Lifecycle.stepStart(state.lifecycle, events)
        const metadata = (() => {
          const signature = state.reasoningSignatures[index]
-          if (signature) return providerMetadata(state.providerMetadataKey, { signature })
+          if (signature) return providerMetadata(state.routeID, { signature })
          const redactedContent = state.reasoningRedactedContent[index]
          if (redactedContent)
-            return providerMetadata(state.providerMetadataKey, {
+            return providerMetadata(state.routeID, {
              redactedData: encodeRedactedContent(redactedContent),
            })
        })()
@@ -697,7 +697,7 @@ const step = (state: ParserState, event: BedrockEvent) =>
    }

    if (event.metadata) {
-      const usage = mapUsage(event.metadata.usage, state.providerMetadataKey) ?? state.usage
+      const usage = mapUsage(event.metadata.usage, state.routeID) ?? state.usage
      return [
        {
          ...state,
@@ -757,7 +757,7 @@ export const protocol = Protocol.make({
  stream: {
    event: BedrockEvent,
    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+      routeID: request.model.route.id,
      tools: ToolStream.empty<number>(),
      finishReason: undefined,
      usage: undefined,
@@ -774,7 +774,6 @@ export const protocol = Protocol.make({
 export const route = Route.make({
  id: ADAPTER,
  provider: "bedrock",
-  providerMetadataKey: "bedrock",
  protocol,
  // Bedrock's URL embeds the region in the route endpoint host and the
  // validated modelId in the path. We read the validated body so the URL
@@ -232,7 +232,7 @@ type GeminiEvent = Schema.Schema.Type<typeof GeminiEvent>

 interface ParserState {
  readonly route: string
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly finishReason?: string
  readonly hasToolCalls: boolean
  readonly promptFeedback?: GeminiPromptFeedback
@@ -309,7 +309,7 @@ const lowerToolCall = (part: ToolCallPart, omitIds: boolean, metadataKey: string

 const lowerMessages = Effect.fn("Gemini.lowerMessages")(function* (request: LLMRequest) {
  const contents: GeminiContent[] = []
-  const metadataKey = request.model.route.providerMetadataKey ?? String(request.model.provider)
+  const metadataKey = request.model.route.id
  const omitCallIds = omitsFunctionCallIds(request.model.id)
  const legacyToolMedia = routesLegacyToolMedia(request.model.id)
  let pendingMedia: GeminiInlineDataPart[] | undefined
@@ -583,7 +583,7 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      state.reasoningId,
      state.reasoningSignature === undefined
        ? undefined
-        : providerMetadata(state.providerMetadataKey, { thoughtSignature: state.reasoningSignature }),
+        : providerMetadata(state.routeID, { thoughtSignature: state.reasoningSignature }),
    )
  if (state.textId !== undefined)
    lifecycle = Lifecycle.textEnd(
@@ -592,7 +592,7 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
      state.textId,
      state.textSignature === undefined
        ? undefined
-        : providerMetadata(state.providerMetadataKey, { thoughtSignature: state.textSignature }),
+        : providerMetadata(state.routeID, { thoughtSignature: state.textSignature }),
    )
  Lifecycle.finish(lifecycle, events, {
    reason: {
@@ -604,7 +604,7 @@ const finish = (state: ParserState): ReadonlyArray<LLMEvent> => {
    providerMetadata:
      state.promptFeedback === undefined
        ? undefined
-        : providerMetadata(state.providerMetadataKey, { promptFeedback: state.promptFeedback }),
+        : providerMetadata(state.routeID, { promptFeedback: state.promptFeedback }),
  })
  return events
 }
@@ -634,9 +634,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
  const nextState = {
    ...state,
    promptFeedback: event.promptFeedback ?? state.promptFeedback,
-    usage: event.usageMetadata
-      ? (mapUsage(event.usageMetadata, state.providerMetadataKey) ?? state.usage)
-      : state.usage,
+    usage: event.usageMetadata ? (mapUsage(event.usageMetadata, state.routeID) ?? state.usage) : state.usage,
  }
  const candidate = event.candidates?.[0]
  if (candidate?.finishReason && mapFinishReason(candidate.finishReason, state.hasToolCalls) === "error")
@@ -692,9 +690,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
            lifecycle,
            events,
            textId,
-            textSignature
-              ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature })
-              : undefined,
+            textSignature ? providerMetadata(state.routeID, { thoughtSignature: textSignature }) : undefined,
          )
          textId = undefined
          textSignature = undefined
@@ -708,7 +704,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          events,
          reasoningId,
          part.text,
-          signature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: signature }) : undefined,
+          signature ? providerMetadata(state.routeID, { thoughtSignature: signature }) : undefined,
        )
        continue
      }
@@ -717,9 +713,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          lifecycle,
          events,
          reasoningId,
-          reasoningSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-            : undefined,
+          reasoningSignature ? providerMetadata(state.routeID, { thoughtSignature: reasoningSignature }) : undefined,
        )
        reasoningId = undefined
        reasoningSignature = undefined
@@ -733,7 +727,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
        events,
        textId,
        part.text,
-        textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
+        textSignature ? providerMetadata(state.routeID, { thoughtSignature: textSignature }) : undefined,
      )
      textSignature = undefined
      continue
@@ -754,9 +748,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          lifecycle,
          events,
          reasoningId,
-          reasoningSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: reasoningSignature })
-            : undefined,
+          reasoningSignature ? providerMetadata(state.routeID, { thoughtSignature: reasoningSignature }) : undefined,
        )
        reasoningId = undefined
        reasoningSignature = undefined
@@ -766,7 +758,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          lifecycle,
          events,
          textId,
-          textSignature ? providerMetadata(state.providerMetadataKey, { thoughtSignature: textSignature }) : undefined,
+          textSignature ? providerMetadata(state.routeID, { thoughtSignature: textSignature }) : undefined,
        )
        textId = undefined
        textSignature = undefined
@@ -778,7 +770,7 @@ const step = (state: ParserState, event: GeminiEvent) => {
          name: part.functionCall.name,
          input,
          providerMetadata: part.thoughtSignature
-            ? providerMetadata(state.providerMetadataKey, { thoughtSignature: part.thoughtSignature })
+            ? providerMetadata(state.routeID, { thoughtSignature: part.thoughtSignature })
            : undefined,
        }),
      )
@@ -821,7 +813,7 @@ export const protocol = Protocol.make({
    event: Protocol.jsonEvent(GeminiEvent),
    initial: (request) => ({
      route: `${request.model.provider}/${request.model.route.id}`,
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+      routeID: request.model.route.id,
      hasToolCalls: false,
      lifecycle: Lifecycle.initial(),
      nextReasoningId: 0,
@@ -835,7 +827,6 @@ export const protocol = Protocol.make({
 export const route = Route.make({
  id: ADAPTER,
  provider: "google",
-  providerMetadataKey: "google",
  protocol,
  // Gemini's path embeds the model id and pins SSE framing at the URL level.
  endpoint: Endpoint.path(({ request }) => `/models/${request.model.id}:streamGenerateContent?alt=sse`, {
@@ -287,9 +287,10 @@ const lowerAssistant = Effect.fn("MistralChat.lowerAssistant")(function* (
  message: LLMRequest["messages"][number],
  normalizeID: (id: string) => string,
  prefix: boolean,
+  routeID: string,
 ) {
  const structured = message.content.some(
-    (part) => part.type === "reasoning" && isMistralThinkingContent(part.providerMetadata?.mistral?.thinking),
+    (part) => part.type === "reasoning" && isMistralThinkingContent(part.providerMetadata?.[routeID]?.thinking),
  )
  const content: Array<Schema.Schema.Type<typeof MistralTextContent> | MistralThinkingContent> = []
  const text: string[] = []
@@ -301,7 +302,7 @@ const lowerAssistant = Effect.fn("MistralChat.lowerAssistant")(function* (
      continue
    }
    if (part.type === "reasoning") {
-      const native = part.providerMetadata?.mistral?.thinking
+      const native = part.providerMetadata?.[routeID]?.thinking
      if (structured && isMistralThinkingContent(native)) content.push(native)
      else if (structured) content.push({ type: "text", text: part.text })
      else text.push(part.text)
@@ -378,13 +379,22 @@ const lowerMessages = Effect.fn("MistralChat.lowerMessages")(function* (request:
    if (message.role === "assistant") {
      const hasToolCalls = message.content.some((part) => part.type === "tool-call")
      const hasNativeThinking = message.content.some(
-        (part) => part.type === "reasoning" && isMistralThinkingContent(part.providerMetadata?.mistral?.thinking),
+        (part) =>
+          part.type === "reasoning" &&
+          isMistralThinkingContent(part.providerMetadata?.[request.model.route.id]?.thinking),
      )
      const text = message.content
        .flatMap((part) => (part.type === "text" || part.type === "reasoning" ? [part.text] : []))
        .join("")
      if (!hasToolCalls && !hasNativeThinking && text.trim() === "") continue
-      messages.push(yield* lowerAssistant(message, normalizeID, !hasToolCalls && message === request.messages.at(-1)))
+      messages.push(
+        yield* lowerAssistant(
+          message,
+          normalizeID,
+          !hasToolCalls && message === request.messages.at(-1),
+          request.model.route.id,
+        ),
+      )
      continue
    }
    messages.push(...(yield* lowerToolResults(message, normalizeID)))
@@ -455,6 +465,7 @@ interface ActiveContent {
 }

 export interface ParserState {
+  readonly routeID: string
  readonly tools: ToolStream.State<ToolKey>
  readonly pendingTools: Partial<Record<ToolKey, PendingTool>>
  readonly toolIDs: ReadonlyMap<string, string>
@@ -469,7 +480,7 @@ export interface ParserState {
  readonly finishReason?: FinishReasonDetails
 }

-const mapUsage = (usage: MistralEvent["usage"]): Usage | undefined => {
+const mapUsage = (usage: MistralEvent["usage"], routeID: string): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const reported =
@@ -485,7 +496,7 @@ const mapUsage = (usage: MistralEvent["usage"]): Usage | undefined => {
    nonCachedInputTokens: ProviderShared.subtractTokens(input, cached),
    cacheReadInputTokens: cached,
    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { mistral: usage },
+    providerMetadata: { [routeID]: usage },
  })
 }

@@ -517,7 +528,7 @@ const thinkingUnits = (value: unknown): ReadonlyArray<MistralThinkingUnit> => {
 const thinkingText = (thinking: ReadonlyArray<MistralThinkingUnit>) =>
  thinking.flatMap((unit) => (typeof unit.text === "string" ? [unit.text] : [])).join("")

-const thinkingMetadata = (thinking: MistralThinkingContent) => ({ mistral: { thinking } })
+const thinkingMetadata = (thinking: MistralThinkingContent, routeID: string) => ({ [routeID]: { thinking } })

 const closeActive = (state: ParserState, events: LLMEvent[]) => {
  if (!state.active) return state
@@ -528,7 +539,7 @@ const closeActive = (state: ParserState, events: LLMEvent[]) => {
          state.lifecycle,
          events,
          state.active.id,
-          thinkingMetadata(state.active.thinking ?? { type: "thinking", thinking: [] }),
+          thinkingMetadata(state.active.thinking ?? { type: "thinking", thinking: [] }, state.routeID),
          thinkingText(state.active.thinking?.thinking ?? []),
        )
  return { ...state, lifecycle, active: undefined }
@@ -561,8 +572,14 @@ const appendThinking = (state: ParserState, events: LLMEvent[], part: MistralOut
    ...current,
    lifecycle:
      text.length > 0
-        ? Lifecycle.reasoningDelta(current.lifecycle, events, active.id, text, thinkingMetadata(thinking))
-        : Lifecycle.reasoningStart(current.lifecycle, events, active.id, thinkingMetadata(thinking)),
+        ? Lifecycle.reasoningDelta(
+            current.lifecycle,
+            events,
+            active.id,
+            text,
+            thinkingMetadata(thinking, state.routeID),
+          )
+        : Lifecycle.reasoningStart(current.lifecycle, events, active.id, thinkingMetadata(thinking, state.routeID)),
    active: { ...active, thinking },
    nextContent: current.active ? current.nextContent : current.nextContent + 1,
  }
@@ -683,7 +700,7 @@ const step = Effect.fn("MistralChat.step")(function* (state: ParserState, event:
    })
  }
  const events: LLMEvent[] = []
-  const usage = mapUsage(event.usage) ?? state.usage
+  const usage = mapUsage(event.usage, state.routeID) ?? state.usage
  if (state.finishReason) {
    if (hasLateContent(event))
      return yield* ProviderShared.eventError(
@@ -761,7 +778,8 @@ export const protocol = Protocol.make({
  body: { schema: MistralBody, from: fromRequest },
  stream: {
    event: MistralStreamEvent,
-    initial: (): ParserState => ({
+    initial: (request): ParserState => ({
+      routeID: request.model.route.id,
      tools: ToolStream.empty<ToolKey>(),
      pendingTools: {},
      toolIDs: new Map(),
@@ -783,7 +801,6 @@ export const httpTransport = HttpTransport.sseJson.with<MistralBody>().with({ fr
 export const route = Route.make({
  id: ADAPTER,
  provider: "mistral",
-  providerMetadataKey: "mistral",
  protocol,
  endpoint: Endpoint.path(PATH, { baseURL: DEFAULT_BASE_URL }),
  auth: Auth.none,
@@ -423,7 +423,7 @@ export interface ParserState {
  readonly completedCompactions: ReadonlySet<string>
  readonly id: string
  readonly name: string
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly tools: ToolStream.State<string>
  readonly hasFunctionCall: boolean
  readonly lifecycle: Lifecycle.State
@@ -476,15 +476,15 @@ export const lowerToolChoice = (protocolName: string, toolChoice: NonNullable<LL

 // 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]
+const itemID = (providerMetadata: ProviderMetadata | undefined, routeID: string) => {
+  const metadata = providerMetadata?.[routeID]
  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 lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenResponsesInputItem => {
-  const id = itemID(part.providerMetadata, providerMetadataKey)
+const lowerToolCall = (part: ToolCallPart, routeID: string): OpenResponsesInputItem => {
+  const id = itemID(part.providerMetadata, routeID)
  return {
    type: "function_call",
    ...(id === undefined ? {} : { id }),
@@ -495,10 +495,10 @@ const lowerToolCall = (part: ToolCallPart, providerMetadataKey: string): OpenRes
  }
 }

-const lowerReasoning = (part: ReasoningPart, providerMetadataKey: string): OpenResponsesReasoningInput | undefined => {
-  const metadata = part.providerMetadata?.[providerMetadataKey]
+const lowerReasoning = (part: ReasoningPart, routeID: string): OpenResponsesReasoningInput | undefined => {
+  const metadata = part.providerMetadata?.[routeID]
  if (!ProviderShared.isRecord(metadata)) return undefined
-  const id = itemID(part.providerMetadata, providerMetadataKey)
+  const id = itemID(part.providerMetadata, routeID)
  const encryptedContent =
    typeof metadata.reasoningEncryptedContent === "string" || metadata.reasoningEncryptedContent === null
      ? metadata.reasoningEncryptedContent
@@ -521,7 +521,7 @@ const lowerMedia = Effect.fn("OpenResponses.lowerMedia")(function* (
  const providerMedia = adapter.lowerMedia?.({ part, media, request })
  if (providerMedia) return providerMedia
  const detail = yield* ProviderShared.validateWith(Schema.decodeUnknownEffect(OpenResponsesInputImage.fields.detail))(
-    part.providerMetadata?.[metadataKey(request.model)]?.detail,
+    part.providerMetadata?.[request.model.route.id]?.detail,
  )
  const url =
    typeof part.data === "string" && (part.data.startsWith("https://") || part.data.startsWith("http://"))
@@ -608,12 +608,12 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (
  adapter: ProviderAdapter,
 ) {
  const input: LoweredInputItem[] = []
-  const providerMetadataKey = metadataKey(request.model)
+  const routeID = request.model.route.id

  for (const message of request.messages) {
    const metadata = yield* ProviderShared.validateWith(
      Schema.decodeUnknownEffect(Schema.UndefinedOr(MessageMetadata)),
-    )(message.providerMetadata?.[providerMetadataKey])
+    )(message.providerMetadata?.[routeID])
    if (message.role === "system") {
      input.push({
        role: "developer",
@@ -638,8 +638,8 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (
        const groups = content.reduce<
          Array<{ id: string | undefined; phase: MessagePhase | null | undefined; parts: TextPart[] }>
        >((groups, part) => {
-          const partMetadata = part.providerMetadata?.[providerMetadataKey]
-          const id = itemID(part.providerMetadata, providerMetadataKey) ?? metadata?.itemId
+          const partMetadata = part.providerMetadata?.[routeID]
+          const id = itemID(part.providerMetadata, routeID) ?? metadata?.itemId
          const partPhase = messagePhase(partMetadata?.phase)
          const phase = partPhase === undefined ? metadata?.phase : partPhase
          const group = groups.at(-1)
@@ -675,7 +675,7 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (
        }
        if (part.type === "reasoning") {
          flushText()
-          const reasoning = lowerReasoning(part, providerMetadataKey)
+          const reasoning = lowerReasoning(part, routeID)
          if (!reasoning) continue
          const existing = reasoning.id === undefined ? undefined : reasoningItems[reasoning.id]
          if (existing) {
@@ -691,12 +691,12 @@ const lowerMessages = Effect.fn("OpenResponses.lowerMessages")(function* (
        if (part.type === "tool-call") {
          flushText()
          if (part.providerExecuted === true) continue
-          input.push(lowerToolCall(part, providerMetadataKey))
+          input.push(lowerToolCall(part, routeID))
          continue
        }
        if (part.type === "tool-result" && part.providerExecuted === true) {
          flushText()
-          const id = itemID(part.providerMetadata, providerMetadataKey)
+          const id = itemID(part.providerMetadata, routeID)
          const hosted =
            part.result.type !== "json"
              ? undefined
@@ -844,7 +844,7 @@ export const fromRequest = Effect.fn("OpenResponses.fromRequest")(function* (req
 // cached-read and cache-write subsets, and `output_tokens` (inclusive total)
 // with a `reasoning_tokens` subset. Pass the totals through and derive the
 // non-cached breakdown.
-export const mapUsage = (usage: OpenResponsesUsage | null | undefined, providerMetadataKey: string) => {
+export const mapUsage = (usage: OpenResponsesUsage | null | undefined, routeID: string) => {
  if (!usage) return undefined
  const cached = usage.input_tokens_details?.cached_tokens
  const cacheWrite = usage.input_tokens_details?.cache_write_tokens
@@ -858,7 +858,7 @@ export const mapUsage = (usage: OpenResponsesUsage | null | undefined, providerM
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(usage.input_tokens, usage.output_tokens, usage.total_tokens),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { [routeID]: usage },
  })
 }

@@ -874,10 +874,8 @@ const mapFinishReason = (event: Event, hasFunctionCall: boolean): FinishReason =
  return hasFunctionCall ? "tool-calls" : "unknown"
 }

-export const metadataKey = (model: LLMRequest["model"]) => model.route.providerMetadataKey ?? "openresponses"
-
 export const providerMetadata = (state: ParserState, metadata: Record<string, unknown>): ProviderMetadata => ({
-  [state.providerMetadataKey]: metadata,
+  [state.routeID]: metadata,
 })

 export type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
@@ -1349,7 +1347,7 @@ const onResponseFinish = Effect.fn("OpenResponses.onResponseFinish")(function* (
      normalized: mapFinishReason(event, current.hasFunctionCall),
      raw: event.response?.incomplete_details?.reason,
    },
-    usage: mapUsage(event.response?.usage, current.providerMetadataKey),
+    usage: mapUsage(event.response?.usage, current.routeID),
    providerMetadata:
      event.response?.id || event.response?.service_tier
        ? providerMetadata(current, {
@@ -1486,7 +1484,7 @@ export const initial = (request: LLMRequest, adapter: ProviderAdapter = BASE_ADA
  completedCompactions: new Set<string>(),
  id: adapter.id,
  name: adapter.name,
-  providerMetadataKey: metadataKey(request.model),
+  routeID: request.model.route.id,
  hasFunctionCall: false,
  tools: ToolStream.empty<string>(),
  lifecycle: Lifecycle.initial(),
@@ -257,7 +257,7 @@ interface PendingToolDelta {
 }

 export interface ParserState {
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly tools: ToolStream.State<number>
  readonly pendingTools: Partial<Record<number, PendingToolDelta>>
  readonly toolCallEvents: ReadonlyArray<LLMEvent>
@@ -329,18 +329,17 @@ const lowerMedia = Effect.fn("OpenAIChat.lowerMedia")(function* (part: MediaPart
 const openAICompatibleReasoningContent = (native: unknown) =>
  isRecord(native) && typeof native.reasoning_content === "string" ? native.reasoning_content : undefined

-const reasoningField = (part: ReasoningPart, providerMetadataKey: string) => {
-  const field = part.providerMetadata?.[providerMetadataKey]?.reasoningField
+const reasoningField = (part: ReasoningPart, routeID: string) => {
+  const field = part.providerMetadata?.[routeID]?.reasoningField
  return typeof field === "string" ? field : undefined
 }

-const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown, providerMetadataKey: string) => {
+const reasoningDetails = (parts: ReadonlyArray<ReasoningPart>, native: unknown, routeID: string) => {
  const observed = parts.flatMap((part) => {
-    const details = part.providerMetadata?.[providerMetadataKey]?.reasoningDetails
+    const details = part.providerMetadata?.[routeID]?.reasoningDetails
    return Array.isArray(details) ? details : []
  })
-  if (parts.some((part) => Array.isArray(part.providerMetadata?.[providerMetadataKey]?.reasoningDetails)))
-    return observed
+  if (parts.some((part) => Array.isArray(part.providerMetadata?.[routeID]?.reasoningDetails))) return observed
  if (isRecord(native) && Array.isArray(native.reasoning_details)) return native.reasoning_details
 }

@@ -372,7 +371,7 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
  message: OpenAIChatRequestMessage,
  configuredField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions & { readonly routeID: string },
 ) {
  const content: TextPart[] = []
  const reasoning: ReasoningPart[] = []
@@ -394,13 +393,13 @@ const lowerAssistantMessage = Effect.fn("OpenAIChat.lowerAssistantMessage")(func
    }
  }
  const text = reasoning.map((part) => part.text).join("")
-  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.providerMetadataKey)
+  const details = reasoningDetails(reasoning, message.native?.openaiCompatible, options.routeID)
  const observedField = reasoning
-    .map((part) => reasoningField(part, options.providerMetadataKey))
+    .map((part) => reasoningField(part, options.routeID))
    .find((value) => value !== undefined)
  const nativeReasoning = openAICompatibleReasoningContent(message.native?.openaiCompatible)
  const fullyStructured = reasoning.every((part) =>
-    Array.isArray(part.providerMetadata?.[options.providerMetadataKey]?.reasoningDetails),
+    Array.isArray(part.providerMetadata?.[options.routeID]?.reasoningDetails),
  )
  const field = (() => {
    if (configuredField !== undefined && (requireReasoning || reasoning.length > 0 || nativeReasoning !== undefined))
@@ -469,7 +468,7 @@ const lowerMessage = Effect.fn("OpenAIChat.lowerMessage")(function* (
  message: OpenAIChatRequestMessage,
  reasoningField: string | undefined,
  requireReasoning: boolean,
-  options: LoweringOptions & { readonly providerMetadataKey: string },
+  options: LoweringOptions & { readonly routeID: string },
 ) {
  if (message.role === "user") return [yield* lowerUserMessage(message, options)]
  if (message.role === "assistant")
@@ -505,7 +504,7 @@ const lowerMessages = Effect.fn("OpenAIChat.lowerMessages")(function* (request:
  const mistral = ["mistral", "devstral", "codestral", "pixtral", "mixtral"].some((family) => modelID.includes(family))
  const lowering = {
    ...options,
-    providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+    routeID: request.model.route.id,
    toolCallID: (id: string) => {
      if (mistral)
        return id
@@ -833,7 +832,7 @@ const mapFinishReason = Effect.fn("OpenAIChat.mapFinishReason")(function* (event
 // Providers differ on cache-hit location: OpenAI uses
 // `prompt_tokens_details.cached_tokens`, DeepSeek uses
 // `prompt_cache_hit_tokens`, and Zai uses top-level `cached_tokens`.
-const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string): Usage | undefined => {
+const mapUsage = (usage: OpenAIChatEvent["usage"], routeID: string): Usage | undefined => {
  if (!usage) return undefined
  const input = usage.prompt_tokens ?? undefined
  const output = usage.completion_tokens ?? undefined
@@ -852,7 +851,7 @@ const mapUsage = (usage: OpenAIChatEvent["usage"], providerMetadataKey: string):
    cacheWriteInputTokens: cacheWrite,
    reasoningTokens: reasoning,
    totalTokens: ProviderShared.totalTokens(input, output, usage.total_tokens ?? undefined),
-    providerMetadata: { [providerMetadataKey]: usage },
+    providerMetadata: { [routeID]: usage },
  })
 }

@@ -927,11 +926,11 @@ const conflictingDetailValue = (previous: unknown, current: unknown) =>
  previous !== undefined && previous !== null && current !== undefined && current !== null && previous !== current

 const reasoningMetadata = (
-  providerMetadataKey: string,
+  routeID: string,
  field: ParserState["reasoningField"],
  details?: ReadonlyArray<unknown>,
 ) => ({
-  [providerMetadataKey]: {
+  [routeID]: {
    ...(field ? { reasoningField: field } : {}),
    ...(details ? { reasoningDetails: details } : {}),
  },
@@ -955,8 +954,8 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    // `choice.usage` instead of the top-level `usage` field.
    const choiceUsage = (choice as unknown as { usage?: OpenAIChatEvent["usage"] })?.usage
    const usage =
-      mapUsage(event.usage, state.providerMetadataKey) ??
-      (choiceUsage ? mapUsage(choiceUsage, state.providerMetadataKey) : undefined) ??
+      mapUsage(event.usage, state.routeID) ??
+      (choiceUsage ? mapUsage(choiceUsage, state.routeID) : undefined) ??
      state.usage
    const rawFinishReason = choice?.finish_reason
    const finishReason = rawFinishReason
@@ -995,7 +994,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>
    const detailDelta = Array.isArray(delta?.reasoning_details) ? delta.reasoning_details : undefined
    if (detailDelta !== undefined) appendReasoningDetails(state.reasoningDetails, detailDelta)
    const reasoningDetailsObserved = state.reasoningDetailsObserved || detailDelta !== undefined
-    const deltaMetadata = reasoningMetadata(state.providerMetadataKey, reasoningField)
+    const deltaMetadata = reasoningMetadata(state.routeID, reasoningField)
    const text = detailDelta?.length ? (detailText(detailDelta) ?? reasoning?.text) : reasoning?.text
    if (text !== undefined) lifecycle = Lifecycle.reasoningDelta(lifecycle, events, "reasoning-0", text, deltaMetadata)
    else if (
@@ -1084,7 +1083,7 @@ const step = (state: ParserState, event: OpenAIChatEvent) =>

    return [
      {
-        providerMetadataKey: state.providerMetadataKey,
+        routeID: state.routeID,
        tools: finished?.tools ?? tools,
        pendingTools,
        toolCallEvents: finished?.events ?? state.toolCallEvents,
@@ -1128,7 +1127,7 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
  // Snapshot details at publish time so the emitted event never observes later
  // mutation of the accumulated `reasoningDetails` array.
  const metadata = reasoningMetadata(
-    state.providerMetadataKey,
+    state.routeID,
    state.reasoningField,
    state.reasoningDetailsObserved ? [...state.reasoningDetails] : undefined,
  )
@@ -1138,7 +1137,7 @@ const finishEvents = Effect.fn("OpenAIChat.finishEvents")(function* (state: Pars
          state.lifecycle,
          events,
          "reasoning-0",
-          reasoningMetadata(state.providerMetadataKey, state.reasoningField),
+          reasoningMetadata(state.routeID, state.reasoningField),
        )
      : state.lifecycle
  const ended = Lifecycle.reasoningEnd(started, events, "reasoning-0", metadata)
@@ -1166,7 +1165,7 @@ export const protocol = Protocol.make({
  stream: {
    event: OpenAIChatStreamEvent,
    initial: (request) => ({
-      providerMetadataKey: request.model.route.providerMetadataKey ?? String(request.model.provider),
+      routeID: request.model.route.id,
      tools: ToolStream.empty<number>(),
      pendingTools: {},
      toolCallEvents: [],
@@ -1190,7 +1189,6 @@ export const httpTransport = HttpTransport.sseJson.with<OpenAIChatBody>().with({
 export const route = Route.make({
  id: ADAPTER,
  provider: "openai",
-  providerMetadataKey: "openai",
  protocol,
  endpoint: Endpoint.path(PATH, { baseURL: DEFAULT_BASE_URL }),
  auth: Auth.none,
@@ -15,7 +15,6 @@ export type OpenAICompatibleChatLanguageModelInput = RouteRoutedLanguageModelInp
 */
 export const route = Route.make({
  id: ADAPTER,
-  providerMetadataKey: "openai",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions"),
  framing: OpenAIChat.framing,
@@ -13,7 +13,6 @@ export type OpenAICompatibleResponsesLanguageModelInput = RouteRoutedLanguageMod
 */
 export const route = Route.make({
  id: ADAPTER,
-  providerMetadataKey: "openresponses",
  protocol: OpenResponses.protocol,
  endpoint: Endpoint.path(OpenResponses.PATH),
  transport: OpenResponses.httpTransport,
@@ -313,7 +313,6 @@ export const route = Route.make({
  compact: { endpoint: ResponsesCompaction.make(adapter), trigger: ResponsesCheckpoint.make(checkpointBody) },
  id: ADAPTER,
  provider: "openai",
-  providerMetadataKey: "openai",
  protocol,
  endpoint,
  auth,
@@ -93,7 +93,7 @@ export const make = <Body>(body: RouteBody<Body>): TriggerCompactOperation =>
          result = new CompactionCheckpointResponse({
            checkpoint,
            responseID,
-            usage: OpenResponses.mapUsage(event.response?.usage, OpenResponses.metadataKey(request.model)),
+            usage: OpenResponses.mapUsage(event.response?.usage, request.model.route.id),
          })
          return [next, [LLMEvent.finish({ reason: { normalized: "stop" } })]] as const
        }),
@@ -102,7 +102,6 @@ export const make = <Body>(body: RouteBody<Body>): TriggerCompactOperation =>
    const route = Route.make({
      id: source.id,
      provider: source.provider,
-      providerMetadataKey: source.providerMetadataKey,
      protocol,
      endpoint: source.endpoint,
      auth: source.auth,
@@ -125,7 +125,7 @@ export const make = (adapter: OpenResponses.ProviderAdapter): CompactOperation =
      return yield* invalid("Compaction response did not contain a checkpoint")
    return new CompactionResponse({
      replacement: result.output.map((item) => toMessage(item, request.model)),
-      usage: OpenResponses.mapUsage(result.usage, OpenResponses.metadataKey(request.model)),
+      usage: OpenResponses.mapUsage(result.usage, request.model.route.id),
    })
  })

@@ -135,7 +135,7 @@ function toMessage(item: (typeof Response.Type.output)[number], model: LLMReques
      CompactionPart.make({ provider: model.provider, id: item.id ?? undefined, encrypted: item.encrypted_content }),
    )

-  const key = OpenResponses.metadataKey(model)
+  const key = model.route.id
  if (item.type === "reasoning") {
    const summary = item.summary.length ? item.summary : [{ text: "" }]
    return Message.assistant(
@@ -36,7 +36,6 @@ export interface Settings extends ProviderPackage.Settings {
 const responsesRoute = Route.make({
  id: "bedrock-mantle-responses",
  provider: id,
-  providerMetadataKey: "mantle",
  protocol: OpenAIResponses.protocol,
  endpoint: OpenAIResponses.route.endpoint,
  auth: OpenAIResponses.route.auth,
@@ -47,7 +46,6 @@ const responsesRoute = Route.make({
 const chatRoute = OpenAIChat.route.with({
  id: "bedrock-mantle-chat",
  provider: id,
-  providerMetadataKey: "mantle",
 })

 export const routes = [responsesRoute, chatRoute]
@@ -45,7 +45,6 @@ const configuredRoute = (input: Config) => {
  return BedrockConverse.route.with({
    ...rest,
    provider: id,
-    providerMetadataKey: "bedrock",
    endpoint: { baseURL: baseURL ?? bedrockBaseURL(resolvedRegion) },
    auth: BedrockAuth.resolveAuth({ apiKey, credentials, profile }, resolvedRegion, { mode: auth }),
  })
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "baseten-chat",
  provider: id,
-  providerMetadataKey: "baseten",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "cerebras-chat",
  provider: id,
-  providerMetadataKey: "cerebras",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -53,7 +53,6 @@ const auth = (input: LanguageModelOptions) => {
 export const route = Route.make({
  id: "cloudflare-ai-gateway",
  provider: id,
-  providerMetadataKey: "cloudflare-ai-gateway",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions"),
  framing: OpenAIChat.framing,
@@ -35,7 +35,6 @@ export const baseURL = (input: WorkersAIURL) => {
 export const route = Route.make({
  id: "cloudflare-workers-ai",
  provider: id,
-  providerMetadataKey: "cloudflare-workers-ai",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions"),
  framing: OpenAIChat.framing,
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "deepinfra-chat",
  provider: id,
-  providerMetadataKey: "deepinfra",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "deepseek-chat",
  provider: id,
-  providerMetadataKey: "deepseek",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "fireworks-chat",
  provider: id,
-  providerMetadataKey: "fireworks",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -28,7 +28,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = Route.make({
  id: "google-vertex-chat",
  provider: id,
-  providerMetadataKey: "vertex",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions"),
  framing: OpenAIChat.framing,
@@ -37,7 +37,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = Route.make({
  id: "google-vertex-messages",
  provider: id,
-  providerMetadataKey: "anthropic",
  protocol: Protocol.make({
    id: AnthropicMessages.protocol.id,
    body: {
@@ -28,7 +28,6 @@ export interface Settings extends ProviderPackage.Settings {
 const route = Route.make({
  id: "google-vertex-responses",
  provider: id,
-  providerMetadataKey: "vertex",
  protocol: OpenResponses.protocol,
  endpoint: Endpoint.path(OpenResponses.PATH),
  transport: OpenResponses.httpTransport,
@@ -68,7 +68,6 @@ const protocol = {
 const route = Route.make({
  id: "google-vertex-gemini",
  provider: id,
-  providerMetadataKey: "vertex",
  protocol,
  endpoint: Endpoint.path(({ request }) => {
    const model = String(request.model.id)
@@ -71,7 +71,6 @@ export const protocol = Protocol.make({
 export const route = Route.make({
  id: "groq-chat",
  provider: id,
-  providerMetadataKey: "openai",
  protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -162,7 +162,6 @@ const bodyOptions = (input: unknown) => {
 export const route = Route.make({
  id: ADAPTER,
  provider: id,
-  providerMetadataKey: "openrouter",
  protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -24,7 +24,6 @@ export interface Settings extends ProviderPackage.Settings {
 export const route = Route.make({
  id: "togetherai-chat",
  provider: id,
-  providerMetadataKey: "togetherai",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -32,13 +32,12 @@ const RESPONSES_WEBSOCKET_ROTATE_AFTER_MS = 24 * 60 * 1000

 const responsesRoute = Route.make({
  compact: { endpoint: XAIResponses.compact },
-  id: "openai-responses",
+  id: "xai-responses",
  provider: id,
-  providerMetadataKey: "xai",
  protocol: XAIResponses.protocol,
  endpoint: Endpoint.path("/responses", { baseURL }),
  transport: OpenResponsesChannel.transport({
-    id: "openai-responses",
+    id: "xai-responses",
    name: "xAI Responses",
    rotateAfterMs: RESPONSES_WEBSOCKET_ROTATE_AFTER_MS,
  }),
@@ -46,9 +45,8 @@ const responsesRoute = Route.make({
 })

 const chatRoute = Route.make({
-  id: "openai-compatible-chat",
+  id: "xai-chat",
  provider: id,
-  providerMetadataKey: "xai",
  protocol: OpenAIChat.protocol,
  endpoint: Endpoint.path("/chat/completions", { baseURL }),
  framing: OpenAIChat.framing,
@@ -42,10 +42,9 @@ export interface Route<
  Compact extends CompactionOperations | undefined = CompactionOperations | undefined,
 > {
  readonly compact: Compact
+  /** Route identity and namespace for metadata emitted and replayed by this route. */
  readonly id: string
  readonly provider?: ProviderID
-  /** ProviderMetadata namespace emitted and consumed by this route. */
-  readonly providerMetadataKey?: string
  readonly protocol: ProtocolID
  readonly endpoint: Endpoint.Definition<Body>
  readonly auth: Auth.Definition
@@ -113,7 +112,6 @@ export interface RoutePatch<Body, Prepared> extends RouteDefaultsInput {
  readonly compact?: CompactionOperations
  readonly id?: string
  readonly provider?: string | ProviderID
-  readonly providerMetadataKey?: string
  readonly auth?: Auth.Definition
  readonly transport?: Transport<Body, Prepared, unknown>
  readonly endpoint?: EndpointPatch<Body>
@@ -276,12 +274,10 @@ const resolveRequestOptions = (request: LLMRequest) => {

 export interface MakeInput<Body, Frame, Event, State> {
  readonly compact?: CompactionOperations
-  /** Route id used in diagnostics and prepared request metadata. */
+  /** Route identity, also used as the namespace for emitted and replayed ProviderMetadata. */
  readonly id: string
  /** Provider identity for route-owned model construction. */
  readonly provider?: string | ProviderID
-  /** ProviderMetadata namespace emitted and consumed by this route. */
-  readonly providerMetadataKey?: string
  /** Semantic API contract — owns body construction, body schema, and parsing. */
  readonly protocol: Protocol<Body, Frame, Event, State>
  /** Where the request is sent. */
@@ -298,12 +294,10 @@ export interface MakeInput<Body, Frame, Event, State> {

 export interface MakeTransportInput<Body, Prepared, Frame, Event, State> {
  readonly compact?: CompactionOperations
-  /** Route id used in diagnostics and prepared request metadata. */
+  /** Route identity, also used as the namespace for emitted and replayed ProviderMetadata. */
  readonly id: string
  /** Provider identity for route-owned model construction. */
  readonly provider?: string | ProviderID
-  /** ProviderMetadata namespace emitted and consumed by this route. */
-  readonly providerMetadataKey?: string
  /** Semantic API contract — owns body construction, body schema, and parsing. */
  readonly protocol: Protocol<Body, Frame, Event, State>
  /** Where the request is sent. */
@@ -376,7 +370,6 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
      compact: routeInput.compact,
      id: routeInput.id,
      provider: routeInput.provider === undefined ? undefined : ProviderID.make(routeInput.provider),
-      providerMetadataKey: routeInput.providerMetadataKey,
      protocol: protocol.id,
      endpoint: routeInput.endpoint,
      auth: routeInput.auth ?? Auth.none,
@@ -385,17 +378,12 @@ function makeFromTransport<Body, Prepared, Frame, Event, State>(
      defaults: routeInput.defaults ?? {},
      body: protocol.body,
      with: (patch: RoutePatch<Body, Prepared>) => {
-        const { compact, id, provider, providerMetadataKey, auth, transport, endpoint, ...defaults } = patch
+        const { compact, id, provider, auth, transport, endpoint, ...defaults } = patch
        return build({
          ...routeInput,
          compact: "compact" in patch ? compact : routeInput.compact,
          id: id ?? routeInput.id,
          provider: provider ?? routeInput.provider,
-          providerMetadataKey:
-            providerMetadataKey ??
-            (provider !== undefined && String(provider) !== String(routeInput.provider)
-              ? String(provider)
-              : routeInput.providerMetadataKey),
          auth: auth ?? routeInput.auth,
          endpoint: endpoint ? Endpoint.merge(routeInput.endpoint, endpoint) : routeInput.endpoint,
          transport: (transport as Transport<Body, Prepared, Frame> | undefined) ?? routeInput.transport,
@@ -536,7 +524,6 @@ export function make<Body, Prepared, Frame, Event, State>(
    compact: input.compact,
    id: input.id,
    provider: input.provider,
-    providerMetadataKey: input.providerMetadataKey,
    protocol,
    endpoint: input.endpoint,
    auth: input.auth,
@@ -15,6 +15,7 @@ import { ProviderID } from "./ids.js"
 export const MessageRole = Schema.Literals(["system", "user", "assistant", "tool"])
 export type MessageRole = Schema.Schema.Type<typeof MessageRole>

+/** Provider-specific data; LLM routes emit and replay it under their route ID. */
 export const ProviderMetadata = Schema.Record(Schema.String, Schema.Record(Schema.String, Schema.Unknown)).annotate({
  identifier: "LLM.ProviderMetadata",
 })
@@ -66,14 +66,14 @@ export function continuationRequest(input: {
    firstAssistant.push({
      type: "reasoning",
      text: "I inspected the previous turn.",
-      providerMetadata: { anthropic: { signature: "sig_continuation_1" } },
+      providerMetadata: { [input.model.route.id]: { signature: "sig_continuation_1" } },
    })
  if (features.has("encrypted-reasoning"))
    firstAssistant.push({
      type: "reasoning",
      text: "I inspected the previous turn.",
      providerMetadata: {
-        openai: {
+        [input.model.route.id]: {
          itemId: "rs_continuation_1",
          reasoningEncryptedContent: "encrypted-continuation-state",
        },
@@ -69,8 +69,8 @@ describe("public exports", () => {
    expect(XAI.model).toBeFunction()
    expect(XAI.provider.responses).toBe(XAI.responses)
    expect(XAI.provider.chat).toBe(XAI.chat)
-    expect(XAI.configure({ apiKey: "fixture" }).responses("grok-4.3").route.id).toBe("openai-responses")
-    expect(XAI.configure({ apiKey: "fixture" }).chat("grok-4.3").route.id).toBe("openai-compatible-chat")
+    expect(XAI.configure({ apiKey: "fixture" }).responses("grok-4.3").route.id).toBe("xai-responses")
+    expect(XAI.configure({ apiKey: "fixture" }).chat("grok-4.3").route.id).toBe("xai-chat")
  })

  test("protocol barrels expose supported low-level routes", () => {
@@ -102,7 +102,7 @@ describe("Anthropic Messages route", () => {
            Message.assistant([{ type: "reasoning", text: "" }]),
            Message.assistant([
              { type: "text", text: "" },
-              { type: "reasoning", text: "", providerMetadata: { anthropic: { signature: "sig_1" } } },
+              { type: "reasoning", text: "", providerMetadata: { "anthropic-messages": { signature: "sig_1" } } },
            ]),
          ],
          cache: "none",
@@ -753,7 +753,11 @@ describe("Anthropic Messages route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "reasoning", text: "thinking", providerMetadata: { anthropic: { signature: "sig_1" } } },
+              {
+                type: "reasoning",
+                text: "thinking",
+                providerMetadata: { "anthropic-messages": { signature: "sig_1" } },
+              },
            ]),
          ],
        }),
@@ -831,8 +835,12 @@ describe("Anthropic Messages route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "reasoning", text: "", providerMetadata: { anthropic: { redactedData: "opaque_1" } } },
-              { type: "reasoning", text: "visible", providerMetadata: { anthropic: { signature: "sig_1" } } },
+              { type: "reasoning", text: "", providerMetadata: { "anthropic-messages": { redactedData: "opaque_1" } } },
+              {
+                type: "reasoning",
+                text: "visible",
+                providerMetadata: { "anthropic-messages": { signature: "sig_1" } },
+              },
            ]),
          ],
        }),
@@ -912,20 +920,28 @@ describe("Anthropic Messages route", () => {
      )

      expect(response.message.content).toMatchObject([
-        { type: "reasoning", text: "Thinking.", providerMetadata: { "custom-anthropic": { signature: "custom_sig" } } },
-        { type: "reasoning", text: "", providerMetadata: { "custom-anthropic": { redactedData: "custom_redacted" } } },
+        {
+          type: "reasoning",
+          text: "Thinking.",
+          providerMetadata: { "custom-anthropic-messages": { signature: "custom_sig" } },
+        },
+        {
+          type: "reasoning",
+          text: "",
+          providerMetadata: { "custom-anthropic-messages": { redactedData: "custom_redacted" } },
+        },
        { type: "tool-call", id: "custom_tool", providerExecuted: true },
        {
          type: "tool-result",
          providerExecuted: true,
-          providerMetadata: { "custom-anthropic": { blockType: "web_search_tool_result", result } },
+          providerMetadata: { "custom-anthropic-messages": { blockType: "web_search_tool_result", result } },
        },
      ])
      expect(response.usage?.providerMetadata).toEqual({
-        "custom-anthropic": { input_tokens: 5, custom_start: true, output_tokens: 2, custom_terminal: true },
+        "custom-anthropic-messages": { input_tokens: 5, custom_start: true, output_tokens: 2, custom_terminal: true },
      })
      expect(response.events.at(-1)).toMatchObject({
-        providerMetadata: { "custom-anthropic": { stopSequence: "custom_stop" } },
+        providerMetadata: { "custom-anthropic-messages": { stopSequence: "custom_stop" } },
      })

      const prepared = yield* compileRequest(
@@ -974,13 +990,13 @@ describe("Anthropic Messages route", () => {
      const reasoningEnds = response.events.filter((event) => event.type === "reasoning-end")
      expect(reasoningEnds).toHaveLength(1)
      expect(reasoningEnds[0]).toMatchObject({
-        providerMetadata: { "custom-anthropic": { signature: "sig_1" } },
+        providerMetadata: { "custom-anthropic-messages": { signature: "sig_1" } },
      })
      expect(response.message.content).toEqual([
        {
          type: "reasoning",
          text: "Reasoning.",
-          providerMetadata: { "custom-anthropic": { signature: "sig_1" } },
+          providerMetadata: { "custom-anthropic-messages": { signature: "sig_1" } },
        },
      ])

@@ -1024,18 +1040,18 @@ describe("Anthropic Messages route", () => {
        totalTokens: 8,
      })
      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { anthropic: { signature: "sig_1" } },
+        providerMetadata: { "anthropic-messages": { signature: "sig_1" } },
      })
      expect(response.events.filter((event) => event.type === "reasoning-end")).toHaveLength(1)
      expect(response.events.find((event) => event.type === "reasoning-delta" && event.text === "")).toBeUndefined()
      expect(response.message.content).toEqual([
        { type: "text", text: "Hello!" },
-        { type: "reasoning", text: "thinking", providerMetadata: { anthropic: { signature: "sig_1" } } },
+        { type: "reasoning", text: "thinking", providerMetadata: { "anthropic-messages": { signature: "sig_1" } } },
      ])
      expect(response.events.at(-1)).toMatchObject({
        type: "finish",
        reason: { normalized: "stop", raw: "end_turn" },
-        providerMetadata: { anthropic: { stopSequence: "\n\nHuman:" } },
+        providerMetadata: { "anthropic-messages": { stopSequence: "\n\nHuman:" } },
      })
    }),
  )
@@ -1064,13 +1080,13 @@ describe("Anthropic Messages route", () => {
      expect(response.events.find((event) => event.type === "step-finish")).toMatchObject({
        reason: { normalized: "stop", raw: "end_turn" },
        usage: { inputTokens: 5, outputTokens: 10, totalTokens: 15 },
-        providerMetadata: { anthropic: { stopSequence: "X" } },
+        providerMetadata: { "anthropic-messages": { stopSequence: "X" } },
      })
      expect(response.events.at(-1)).toMatchObject({
        type: "finish",
        reason: { normalized: "stop", raw: "end_turn" },
        usage: { inputTokens: 5, outputTokens: 10, totalTokens: 15 },
-        providerMetadata: { anthropic: { stopSequence: "X" } },
+        providerMetadata: { "anthropic-messages": { stopSequence: "X" } },
      })
    }),
  )
@@ -1296,7 +1312,7 @@ describe("Anthropic Messages route", () => {
        reasoningTokens: 3,
        totalTokens: 15,
        providerMetadata: {
-          anthropic: {
+          "anthropic-messages": {
            input_tokens: null,
            cache_read_input_tokens: 2,
            service_tier: "standard",
@@ -1337,7 +1353,7 @@ describe("Anthropic Messages route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { anthropic: { signature: "sig_1" } } },
+        { type: "reasoning", text: "", providerMetadata: { "anthropic-messages": { signature: "sig_1" } } },
      ])

      const prepared = yield* compileRequest(LLM.request({ model, messages: [response.message], cache: "none" }))
@@ -1368,10 +1384,10 @@ describe("Anthropic Messages route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { anthropic: { signature: "sig_1" } } },
+        { type: "reasoning", text: "", providerMetadata: { "anthropic-messages": { signature: "sig_1" } } },
      ])
      expect(response.events.find((event) => event.type === "reasoning-end")).toMatchObject({
-        providerMetadata: { anthropic: { signature: "sig_1" } },
+        providerMetadata: { "anthropic-messages": { signature: "sig_1" } },
      })
    }),
  )
@@ -1435,10 +1451,10 @@ describe("Anthropic Messages route", () => {
      const response = yield* LLMClient.generate(request).pipe(Effect.provide(fixedResponse(body)))

      expect(response.events.find((event) => event.type === "reasoning-start")).toMatchObject({
-        providerMetadata: { anthropic: { redactedData: "opaque_1" } },
+        providerMetadata: { "anthropic-messages": { redactedData: "opaque_1" } },
      })
      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { anthropic: { redactedData: "opaque_1" } } },
+        { type: "reasoning", text: "", providerMetadata: { "anthropic-messages": { redactedData: "opaque_1" } } },
        { type: "text", text: "Hello" },
      ])
    }),
@@ -1582,7 +1598,7 @@ describe("Anthropic Messages route", () => {
        cacheReadInputTokens: undefined,
        cacheWriteInputTokens: undefined,
        totalTokens: 6,
-        providerMetadata: { anthropic: { input_tokens: 5, output_tokens: 1 } },
+        providerMetadata: { "anthropic-messages": { input_tokens: 5, output_tokens: 1 } },
      })

      expect(response.toolCalls).toEqual([
@@ -1891,7 +1907,7 @@ describe("Anthropic Messages route", () => {
        // The complete payload rides in provider metadata as irreducible replay
        // state for later stateless requests.
        providerMetadata: {
-          anthropic: {
+          "anthropic-messages": {
            blockType: "web_search_tool_result",
            result: [{ type: "web_search_result", url: "https://example.com", title: "Example" }],
          },
@@ -837,7 +837,7 @@ describe("Bedrock Converse route", () => {
                  {
                    type: "reasoning",
                    text: "Partial thought",
-                    providerMetadata: signature === undefined ? undefined : { bedrock: { signature } },
+                    providerMetadata: signature === undefined ? undefined : { "bedrock-converse": { signature } },
                    cache: new CacheHint({ type: "ephemeral" }),
                  },
                ]),
@@ -865,7 +865,7 @@ describe("Bedrock Converse route", () => {
            Message.user("Think"),
            Message.assistant([
              { type: "reasoning", text: "", cache },
-              { type: "reasoning", text: "  ", providerMetadata: { bedrock: { signature: "" } }, cache },
+              { type: "reasoning", text: "  ", providerMetadata: { "bedrock-converse": { signature: "" } }, cache },
            ]),
            Message.user([{ type: "text", text: "Continue", cache }]),
          ],
@@ -893,11 +893,16 @@ describe("Bedrock Converse route", () => {
              {
                type: "reasoning",
                text: "Signed thought",
-                providerMetadata: { bedrock: { signature: "sig_1" } },
+                providerMetadata: { "bedrock-converse": { signature: "sig_1" } },
                cache,
              },
              { type: "reasoning", text: "", encrypted: "sig_2", cache },
-              { type: "reasoning", text: "", providerMetadata: { bedrock: { redactedData: "cmVkYWN0ZWQ=" } }, cache },
+              {
+                type: "reasoning",
+                text: "",
+                providerMetadata: { "bedrock-converse": { redactedData: "cmVkYWN0ZWQ=" } },
+                cache,
+              },
              { type: "text", text: "Checking" },
              ToolCallPart.make({ id: "call_1", name: "lookup", input: {} }),
            ]),
@@ -937,7 +942,7 @@ describe("Bedrock Converse route", () => {
      expect(reasoning).toEqual({
        type: "reasoning-end",
        id: "reasoning-0",
-        providerMetadata: { bedrock: { signature: "sig_1" } },
+        providerMetadata: { "bedrock-converse": { signature: "sig_1" } },
      })

      const prepared = yield* compileRequest(
@@ -960,7 +965,7 @@ describe("Bedrock Converse route", () => {
    }),
  )

-  it.effect("round-trips reassigned provider reasoning and usage metadata in its own namespace", () =>
+  it.effect("preserves route-keyed reasoning and usage metadata when the provider is reassigned", () =>
    Effect.gen(function* () {
      const compatible = model.route.with({ provider: "custom-bedrock" }).model({ id: model.id })
      const redactedData = "cmVkYWN0ZWQtdGhpbmtpbmc="
@@ -988,12 +993,12 @@ describe("Bedrock Converse route", () => {
        {
          type: "reasoning",
          text: "Let me think.",
-          providerMetadata: { "custom-bedrock": { signature: "custom_sig" } },
+          providerMetadata: { "bedrock-converse": { signature: "custom_sig" } },
        },
-        { type: "reasoning", text: "", providerMetadata: { "custom-bedrock": { redactedData } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { redactedData } } },
      ])
      expect(response.usage?.providerMetadata).toEqual({
-        "custom-bedrock": { inputTokens: 5, outputTokens: 2, totalTokens: 7 },
+        "bedrock-converse": { inputTokens: 5, outputTokens: 2, totalTokens: 7 },
      })

      const prepared = yield* compileRequest(
@@ -1030,13 +1035,13 @@ describe("Bedrock Converse route", () => {
        type: "reasoning-delta",
        id: "reasoning-0",
        text: "",
-        providerMetadata: { bedrock: { signature: "sig_1" } },
+        providerMetadata: { "bedrock-converse": { signature: "sig_1" } },
      })
      expect(response.message.content).toEqual([
        {
          type: "reasoning",
          text: "Let me think.",
-          providerMetadata: { bedrock: { signature: "sig_1" } },
+          providerMetadata: { "bedrock-converse": { signature: "sig_1" } },
        },
      ])

@@ -1061,7 +1066,7 @@ describe("Bedrock Converse route", () => {
      const response = yield* LLMClient.generate(baseRequest).pipe(Effect.provide(fixedBytes(body)))

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { bedrock: { signature: "sig_1" } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { signature: "sig_1" } } },
      ])
    }),
  )
@@ -1081,10 +1086,10 @@ describe("Bedrock Converse route", () => {
        type: "reasoning-delta",
        id: "reasoning-0",
        text: "",
-        providerMetadata: { bedrock: { redactedData } },
+        providerMetadata: { "bedrock-converse": { redactedData } },
      })
      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { bedrock: { redactedData } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { redactedData } } },
      ])
    }),
  )
@@ -1124,12 +1129,12 @@ describe("Bedrock Converse route", () => {
        type: "reasoning-delta",
        id: "reasoning-0",
        text: "",
-        providerMetadata: { bedrock: { redactedData } },
+        providerMetadata: { "bedrock-converse": { redactedData } },
      })
      expect(response.events.find((event) => event.type === "reasoning-end")).toEqual({
        type: "reasoning-end",
        id: "reasoning-0",
-        providerMetadata: { bedrock: { redactedData } },
+        providerMetadata: { "bedrock-converse": { redactedData } },
      })
      const prepared = yield* compileRequest(
        LLM.request({
@@ -1181,8 +1186,8 @@ describe("Bedrock Converse route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { bedrock: { redactedData: "AQI=" } } },
-        { type: "reasoning", text: "", providerMetadata: { bedrock: { redactedData: "AwQ=" } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { redactedData: "AQI=" } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { redactedData: "AwQ=" } } },
      ])
    }),
  )
@@ -1203,7 +1208,7 @@ describe("Bedrock Converse route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { bedrock: { redactedData: "AQID" } } },
+        { type: "reasoning", text: "", providerMetadata: { "bedrock-converse": { redactedData: "AQID" } } },
      ])
    }),
  )
@@ -41,8 +41,8 @@ describe("Amazon Bedrock Mantle provider", () => {
        protocol: "openai-responses",
        body: { model: "openai.gpt-oss-120b", store: false },
      })
-      expect(provider.model("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
-      expect(provider.chat("openai.gpt-oss-120b").route.providerMetadataKey).toBe("mantle")
+      expect(provider.model("openai.gpt-oss-120b").route.id).toBe("bedrock-mantle-responses")
+      expect(provider.chat("openai.gpt-oss-120b").route.id).toBe("bedrock-mantle-chat")
    }),
  )

@@ -160,7 +160,7 @@ describe("Amazon Bedrock Mantle provider", () => {
      )

      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        mantle: { itemId: "msg_95d4d0af4350432a", reasoningEncryptedContent: "mantle-state" },
+        "bedrock-mantle-responses": { itemId: "msg_95d4d0af4350432a", reasoningEncryptedContent: "mantle-state" },
      })
      expect(prepared.body.input).toEqual([
        {
@@ -126,7 +126,7 @@ for (const model of [
            name: item.type,
            result: { type: "json", value: item },
            providerExecuted: true,
-            providerMetadata: { [model.route.providerMetadataKey ?? model.provider]: { itemId: item.id } },
+            providerMetadata: { [model.route.id]: { itemId: item.id } },
          }),
        ],
      })
@@ -196,11 +196,10 @@ const retainedItems = [

 for (const model of [
  OpenAI.configure({ apiKey: "test" }).responses("gpt-5.3-codex"),
-  ...[undefined, "custom"].map((providerMetadataKey) =>
+  ...["default-metadata", "custom"].map((id) =>
    Route.make({
-      id: providerMetadataKey ?? "default-metadata",
+      id,
      provider: "openai",
-      providerMetadataKey,
      protocol: OpenAIResponses.protocol,
      compact: OpenAIResponses.route.compact,
      endpoint: OpenAIResponses.route.endpoint,
@@ -493,7 +493,7 @@ describe("Gemini route", () => {
                id: "call_image",
                name: "read",
                input: { path: "pixel.png" },
-                providerMetadata: { google: { thoughtSignature: "sig_1" } },
+                providerMetadata: { gemini: { thoughtSignature: "sig_1" } },
              }),
            ]),
            Message.tool({
@@ -872,7 +872,7 @@ describe("Gemini route", () => {
        reasoningTokens: 1,
        totalTokens: 7,
        providerMetadata: {
-          google: {
+          gemini: {
            promptTokenCount: 5,
            candidatesTokenCount: 2,
            totalTokenCount: 7,
@@ -973,22 +973,22 @@ describe("Gemini route", () => {
        {
          type: "reasoning",
          text: "A",
-          providerMetadata: { google: { thoughtSignature: "reasoning_sig_a" } },
+          providerMetadata: { gemini: { thoughtSignature: "reasoning_sig_a" } },
        },
        {
          type: "text",
          text: "X",
-          providerMetadata: { google: { thoughtSignature: "text_sig_x" } },
+          providerMetadata: { gemini: { thoughtSignature: "text_sig_x" } },
        },
        {
          type: "reasoning",
          text: "B",
-          providerMetadata: { google: { thoughtSignature: "reasoning_sig_b" } },
+          providerMetadata: { gemini: { thoughtSignature: "reasoning_sig_b" } },
        },
        {
          type: "text",
          text: "Y",
-          providerMetadata: { google: { thoughtSignature: "text_sig_y" } },
+          providerMetadata: { gemini: { thoughtSignature: "text_sig_y" } },
        },
      ])
    }),
@@ -1079,11 +1079,11 @@ describe("Gemini route", () => {
      expect(reasoningEnd).toEqual({
        type: "reasoning-end",
        id: "reasoning-0",
-        providerMetadata: { google: { thoughtSignature: "thought_sig" } },
+        providerMetadata: { gemini: { thoughtSignature: "thought_sig" } },
      })
      expect(toolCall).toMatchObject({
        id: "provider_call",
-        providerMetadata: { google: { thoughtSignature: "tool_sig" } },
+        providerMetadata: { gemini: { thoughtSignature: "tool_sig" } },
      })
      expect(response.events.findIndex((event) => event.type === "reasoning-end")).toBeLessThan(
        response.events.findIndex((event) => event.type === "tool-call"),
@@ -1153,7 +1153,7 @@ describe("Gemini route", () => {
      expect(delta).toMatchObject({
        id: "text-0",
        text: "All done.",
-        providerMetadata: { google: { thoughtSignature: "text_sig" } },
+        providerMetadata: { gemini: { thoughtSignature: "text_sig" } },
      })

      const prepared = yield* compileRequest(
@@ -1190,7 +1190,7 @@ describe("Gemini route", () => {
      expect(delta).toMatchObject({ id: "text-0", text: "Working.", providerMetadata: undefined })
      expect(end).toMatchObject({
        id: "text-0",
-        providerMetadata: { google: { thoughtSignature: "tail_sig" } },
+        providerMetadata: { gemini: { thoughtSignature: "tail_sig" } },
      })
    }),
  )
@@ -1244,7 +1244,7 @@ describe("Gemini route", () => {
                id: "tool_0",
                name: "lookup",
                input: { query: "weather" },
-                providerMetadata: { google: { thoughtSignature: "parallel_signature" } },
+                providerMetadata: { gemini: { thoughtSignature: "parallel_signature" } },
              }),
              ToolCallPart.make({ id: "tool_1", name: "lookup", input: { query: "news" } }),
              ToolCallPart.make({ id: "tool_2", name: "lookup", input: { query: "sports" } }),
@@ -1333,7 +1333,7 @@ describe("Gemini route", () => {
        cacheReadInputTokens: undefined,
        reasoningTokens: undefined,
        totalTokens: 6,
-        providerMetadata: { google: { promptTokenCount: 5, candidatesTokenCount: 1 } },
+        providerMetadata: { gemini: { promptTokenCount: 5, candidatesTokenCount: 1 } },
      })

      expect(response.toolCalls[0].id).toMatch(/^tool_[0-9a-zA-Z]+$/)
@@ -1629,7 +1629,7 @@ describe("Gemini route", () => {
        type: "finish",
        reason: { normalized: "content-filter", raw: "FUTURE_SAFETY_REASON" },
        providerMetadata: {
-          google: {
+          gemini: {
            promptFeedback: {
              blockReason: "FUTURE_SAFETY_REASON",
              blockReasonMessage: "Prompt blocked",
@@ -11,6 +11,49 @@ import { deltaChunk, finishChunk } from "../lib/openai-chunks.js"
 import { sseEvents } from "../lib/sse.js"

 describe("Google Vertex providers", () => {
+  it.effect("round-trips Claude signatures under the Vertex Messages route ID", () =>
+    Effect.gen(function* () {
+      const model = GoogleVertexMessages.configure({ project: "project", accessToken: "token" }).model(
+        "claude-sonnet-4-5",
+      )
+      const response = yield* LLMClient.generate(LLM.request({ model, prompt: "Think." })).pipe(
+        Effect.provide(
+          fixedResponse(
+            sseEvents(
+              { type: "message_start", message: { usage: { input_tokens: 5 } } },
+              { type: "content_block_start", index: 0, content_block: { type: "thinking", thinking: "Thinking." } },
+              {
+                type: "content_block_delta",
+                index: 0,
+                delta: { type: "signature_delta", signature: "vertex-signature" },
+              },
+              { type: "content_block_stop", index: 0 },
+              { type: "message_delta", delta: { stop_reason: "end_turn" }, usage: { output_tokens: 2 } },
+              { type: "message_stop" },
+            ),
+          ),
+        ),
+      )
+      expect(response.message.content).toEqual([
+        {
+          type: "reasoning",
+          text: "Thinking.",
+          providerMetadata: { "google-vertex-messages": { signature: "vertex-signature" } },
+        },
+      ])
+      expect(response.usage?.providerMetadata).toEqual({
+        "google-vertex-messages": { input_tokens: 5, output_tokens: 2 },
+      })
+      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
+      expect(replay.body.messages).toEqual([
+        {
+          role: "assistant",
+          content: [{ type: "thinking", thinking: "Thinking.", signature: "vertex-signature" }],
+        },
+      ])
+    }),
+  )
+
  it.effect("sends Gemini requests to the global Vertex endpoint", () =>
    Effect.gen(function* () {
      const response = yield* LLMClient.generate(
@@ -89,7 +132,7 @@ describe("Google Vertex providers", () => {
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
-                providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
+                providerMetadata: { "google-vertex-gemini": { functionCallId: "provider_call_1" } },
              }),
            ]),
            Message.tool({
@@ -97,7 +140,7 @@ describe("Google Vertex providers", () => {
              name: "lookup",
              result: "sunny",
              resultType: "text",
-              providerMetadata: { vertex: { functionCallId: "provider_call_1" } },
+              providerMetadata: { "google-vertex-gemini": { functionCallId: "provider_call_1" } },
            }),
          ],
        }),
@@ -158,17 +201,17 @@ describe("Google Vertex providers", () => {
      const text = response.events.find((event) => event.type === "text-delta")
      const toolCall = response.toolCalls[0]

-      expect(reasoning?.providerMetadata).toEqual({ vertex: { thoughtSignature: "reasoning_sig" } })
-      expect(text?.providerMetadata).toEqual({ vertex: { thoughtSignature: "text_sig" } })
+      expect(reasoning?.providerMetadata).toEqual({ "google-vertex-gemini": { thoughtSignature: "reasoning_sig" } })
+      expect(text?.providerMetadata).toEqual({ "google-vertex-gemini": { thoughtSignature: "text_sig" } })
      expect(toolCall).toMatchObject({
        id: "provider_call_1",
-        providerMetadata: { vertex: { thoughtSignature: "tool_sig" } },
+        providerMetadata: { "google-vertex-gemini": { thoughtSignature: "tool_sig" } },
      })
      expect(response.usage?.providerMetadata).toEqual({
-        vertex: { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
+        "google-vertex-gemini": { promptTokenCount: 5, candidatesTokenCount: 2, thoughtsTokenCount: 1 },
      })
      expect(response.events.at(-1)?.providerMetadata).toEqual({
-        vertex: { promptFeedback: { blockReasonMessage: "Reviewed" } },
+        "google-vertex-gemini": { promptFeedback: { blockReasonMessage: "Reviewed" } },
      })

      const prepared = yield* compileRequest(
@@ -18,8 +18,7 @@ for (const model of [
          type: "media",
          mediaType: "image/png",
          data: "https://example.com/image.png",
-          providerMetadata:
-            detail === undefined ? undefined : { [model.route.providerMetadataKey ?? model.provider]: { detail } },
+          providerMetadata: detail === undefined ? undefined : { [model.route.id]: { detail } },
        })),
      )
      const codec = Schema.fromJsonString(Message)
@@ -50,7 +49,7 @@ it.effect("rejects malformed image detail instead of silently discarding it", ()
            type: "media",
            mediaType: "image/png",
            data: "https://example.com/image.png",
-            providerMetadata: { openai: { detail: 42 } },
+            providerMetadata: { "openai-responses": { detail: 42 } },
          }),
        ],
      }),
@@ -26,7 +26,6 @@ describe("Mistral Chat", () => {
    expect(Mistral.route).toMatchObject({
      id: "mistral-chat",
      provider: "mistral",
-      providerMetadataKey: "mistral",
      protocol: "mistral-chat",
    })
    expect(Mistral.route.endpoint).toMatchObject({
@@ -316,7 +315,7 @@ describe("Mistral Chat", () => {
          type: "reasoning",
          text: "Consider",
          providerMetadata: {
-            mistral: {
+            "mistral-chat": {
              thinking: {
                type: "thinking",
                thinking: [{ type: "text", text: "Consider" }],
@@ -360,7 +359,7 @@ describe("Mistral Chat", () => {
          type: "reasoning",
          text: "",
          providerMetadata: {
-            mistral: { thinking: { type: "thinking", thinking: [], marker: "opaque" } },
+            "mistral-chat": { thinking: { type: "thinking", thinking: [], marker: "opaque" } },
          },
        },
      ])
@@ -34,45 +34,48 @@ import { dynamicResponse } from "../lib/http.js"
 import { sseEvents } from "../lib/sse.js"

 describe("native OpenAI-compatible providers", () => {
-  test("assigns provider-owned metadata namespaces across native routes", () => {
+  test("assigns API-specific route identities across native providers", () => {
    const vertex = { project: "project", accessToken: "token" }
    const providers = [
-      [OpenAI.configure({ apiKey: "test" }).chat("model"), "openai"],
-      [OpenAI.configure({ apiKey: "test" }).responses("model"), "openai"],
-      [Azure.configure({ resourceName: "resource", apiKey: "test" }).chat("model"), "azure"],
-      [Azure.configure({ resourceName: "resource", apiKey: "test" }).responses("model"), "azure"],
-      [AmazonBedrock.configure({ apiKey: "test" }).model("model"), "bedrock"],
-      [AmazonBedrockMantle.configure({ apiKey: "test" }).chat("model"), "mantle"],
-      [AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"), "mantle"],
-      [Google.configure({ apiKey: "test" }).model("model"), "google"],
-      [GoogleVertex.configure(vertex).model("model"), "vertex"],
-      [GoogleVertexChat.configure(vertex).model("model"), "vertex"],
-      [GoogleVertexResponses.configure(vertex).model("model"), "vertex"],
-      [GoogleVertexMessages.configure(vertex).model("model"), "anthropic"],
-      [Anthropic.configure({ apiKey: "test" }).model("model"), "anthropic"],
+      [OpenAI.configure({ apiKey: "test" }).chat("model"), "openai-chat"],
+      [OpenAI.configure({ apiKey: "test" }).responses("model"), "openai-responses"],
+      [Azure.configure({ resourceName: "resource", apiKey: "test" }).chat("model"), "azure-openai-chat"],
+      [Azure.configure({ resourceName: "resource", apiKey: "test" }).responses("model"), "azure-openai-responses"],
+      [AmazonBedrock.configure({ apiKey: "test" }).model("model"), "bedrock-converse"],
+      [AmazonBedrockMantle.configure({ apiKey: "test" }).chat("model"), "bedrock-mantle-chat"],
+      [AmazonBedrockMantle.configure({ apiKey: "test" }).responses("model"), "bedrock-mantle-responses"],
+      [Google.configure({ apiKey: "test" }).model("model"), "gemini"],
+      [GoogleVertex.configure(vertex).model("model"), "google-vertex-gemini"],
+      [GoogleVertexChat.configure(vertex).model("model"), "google-vertex-chat"],
+      [GoogleVertexResponses.configure(vertex).model("model"), "google-vertex-responses"],
+      [GoogleVertexMessages.configure(vertex).model("model"), "google-vertex-messages"],
+      [Anthropic.configure({ apiKey: "test" }).model("model"), "anthropic-messages"],
      [
        AnthropicCompatible.configure({ baseURL: "https://example.test/v1", provider: "minimax" }).model("model"),
-        "minimax",
+        "anthropic-messages",
+      ],
+      [
+        OpenAICompatible.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"),
+        "openai-compatible-chat",
      ],
-      [OpenAICompatible.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"), "custom"],
      [
        OpenAICompatibleResponses.configure({ baseURL: "https://example.test/v1", provider: "custom" }).model("model"),
-        "custom",
+        "openai-compatible-responses",
      ],
-      [Cerebras.configure({ apiKey: "test" }).model("model"), "cerebras"],
-      [Baseten.configure({ apiKey: "test" }).model("model"), "baseten"],
-      [DeepSeek.configure({ apiKey: "test" }).model("model"), "deepseek"],
-      [Fireworks.configure({ apiKey: "test" }).model("model"), "fireworks"],
-      [DeepInfra.configure({ apiKey: "test" }).model("model"), "deepinfra"],
-      [TogetherAI.configure({ apiKey: "test" }).model("model"), "togetherai"],
+      [Cerebras.configure({ apiKey: "test" }).model("model"), "cerebras-chat"],
+      [Baseten.configure({ apiKey: "test" }).model("model"), "baseten-chat"],
+      [DeepSeek.configure({ apiKey: "test" }).model("model"), "deepseek-chat"],
+      [Fireworks.configure({ apiKey: "test" }).model("model"), "fireworks-chat"],
+      [DeepInfra.configure({ apiKey: "test" }).model("model"), "deepinfra-chat"],
+      [TogetherAI.configure({ apiKey: "test" }).model("model"), "togetherai-chat"],
      [CloudflareAIGateway.configure({ accountId: "account" }).model("model"), "cloudflare-ai-gateway"],
      [CloudflareWorkersAI.configure({ accountId: "account" }).model("model"), "cloudflare-workers-ai"],
      [OpenRouter.configure({ apiKey: "test" }).model("model"), "openrouter"],
-      [XAI.configure({ apiKey: "test" }).chat("model"), "xai"],
-      [XAI.configure({ apiKey: "test" }).responses("model"), "xai"],
+      [XAI.configure({ apiKey: "test" }).chat("model"), "xai-chat"],
+      [XAI.configure({ apiKey: "test" }).responses("model"), "xai-responses"],
    ] as const

-    for (const [model, key] of providers) expect(model.route.providerMetadataKey).toBe(key)
+    for (const [model, id] of providers) expect(model.route.id).toBe(id)
  })

  test("preserves native Together AI and Cerebras provider and route identities", () => {
@@ -37,7 +37,7 @@ describe("Open Responses completed item text", () => {
            type: "text-end",
            id: "msg_1",
            text,
-            providerMetadata: { "openai-compatible": { itemId: "msg_1", phase: "final_answer" } },
+            providerMetadata: { "openai-compatible-responses": { itemId: "msg_1", phase: "final_answer" } },
          },
        ])
      }),
@@ -138,7 +138,7 @@ describe("Open Responses completed item reasoning", () => {
        expect(response.reasoning).toBe(fixture.text)
        expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
        expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-          "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: "encrypted" },
+          "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted" },
        })
      }),
    )
@@ -97,27 +97,37 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "reasoning-start",
          id: "rs_1:0",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:0", text: "First" },
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { "openai-compatible": { itemId: "rs_1" } } },
+        {
+          type: "reasoning-end",
+          id: "rs_1:0",
+          providerMetadata: { "openai-compatible-responses": { itemId: "rs_1" } },
+        },
        {
          type: "reasoning-start",
          id: "rs_1:1",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:1", text: "Second" },
-        { type: "reasoning-end", id: "rs_1:1", providerMetadata: { "openai-compatible": { itemId: "rs_1" } } },
+        {
+          type: "reasoning-end",
+          id: "rs_1:1",
+          providerMetadata: { "openai-compatible-responses": { itemId: "rs_1" } },
+        },
        {
          type: "reasoning-start",
          id: "rs_1:2",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:2", text: "Third" },
        {
          type: "reasoning-end",
          id: "rs_1:2",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: {
+            "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+          },
        },
      ])
    }),
@@ -141,13 +151,17 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "reasoning-start",
          id: "rs_1",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: {
+            "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+          },
        },
        {
          type: "reasoning-end",
          id: "rs_1",
          text: "Not streamed",
-          providerMetadata: { "openai-compatible": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: {
+            "openai-compatible-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+          },
        },
      ])
    }),
@@ -176,32 +190,36 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "text-start",
          id: "msg_1",
-          providerMetadata: { "openai-compatible": { itemId: "msg_1", phase: "commentary" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_1", phase: "commentary" } },
        },
        { type: "text-delta", id: "msg_1", text: "Checking" },
        {
          type: "text-end",
          id: "msg_1",
-          providerMetadata: { "openai-compatible": { itemId: "msg_1", phase: "commentary" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_1", phase: "commentary" } },
        },
        {
          type: "text-start",
          id: "msg_2",
-          providerMetadata: { "openai-compatible": { itemId: "msg_2", phase: null } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_2", phase: null } },
        },
        { type: "text-delta", id: "msg_2", text: "Cannot help." },
        {
          type: "text-end",
          id: "msg_2",
-          providerMetadata: { "openai-compatible": { itemId: "msg_2", phase: "final_answer" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_2", phase: "final_answer" } },
        },
        {
          type: "text-start",
          id: "msg_3",
-          providerMetadata: { "openai-compatible": { itemId: "msg_3", phase: null } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_3", phase: null } },
        },
        { type: "text-delta", id: "msg_3", text: "Done-only refusal." },
-        { type: "text-end", id: "msg_3", providerMetadata: { "openai-compatible": { itemId: "msg_3", phase: null } } },
+        {
+          type: "text-end",
+          id: "msg_3",
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_3", phase: null } },
+        },
      ])
    }),
  )
@@ -232,24 +250,24 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "text-start",
          id: "msg_text",
-          providerMetadata: { "openai-compatible": { itemId: "msg_text" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_text" } },
        },
        {
          type: "text-end",
          id: "msg_text",
          text: "Done-only text.",
-          providerMetadata: { "openai-compatible": { itemId: "msg_text" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_text" } },
        },
        {
          type: "text-start",
          id: "msg_refusal",
-          providerMetadata: { "openai-compatible": { itemId: "msg_refusal" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_refusal" } },
        },
        {
          type: "text-end",
          id: "msg_refusal",
          text: "Done-only refusal.",
-          providerMetadata: { "openai-compatible": { itemId: "msg_refusal" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_refusal" } },
        },
      ])
    }),
@@ -305,7 +323,7 @@ describe("Open Responses basic-item lifecycles", () => {
          response: { id: "resp_1", output: [{ type: "reasoning", id: "msg_879a68b589198b4c" }, terminal] },
        },
      )
-      const providerMetadata = { "openai-compatible": { itemId: "fc_828bee50dee1d029" } }
+      const providerMetadata = { "openai-compatible-responses": { itemId: "fc_828bee50dee1d029" } }
      expect(events.filter((event) => event.type.startsWith("tool-"))).toEqual([
        { type: "tool-input-start", id: "call_bc1eb4b42e70ee53", name: "get_weather", providerMetadata },
        {
@@ -338,7 +356,7 @@ describe("Open Responses basic-item lifecycles", () => {
      )
      const call = events.find(LLMEvent.is.toolCall)
      expect(call).toMatchObject({ id: "call_1", name: "lookup", input: { query: "weather" } })
-      expect(call?.providerMetadata?.["openai-compatible"]).toMatchObject({
+      expect(call?.providerMetadata?.["openai-compatible-responses"]).toMatchObject({
        itemId: expect.stringMatching(/^fc_[0-9a-f]{32}$/),
      })
    }),
@@ -354,7 +372,7 @@ describe("Open Responses basic-item lifecycles", () => {
        arguments: '{"query":"weather"}',
      }
      const events = yield* collect({ type: "response.output_item.done", item }, completed)
-      const providerMetadata = { "openai-compatible": { itemId: "fc_1" } }
+      const providerMetadata = { "openai-compatible-responses": { itemId: "fc_1" } }
      expect(events.filter((event) => event.type.startsWith("tool-"))).toEqual([
        { type: "tool-input-start", id: "call_1", name: "lookup", namespace: undefined, providerMetadata },
        { type: "tool-input-end", id: "call_1", name: "lookup", namespace: undefined, providerMetadata },
@@ -371,7 +389,7 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "finish",
          reason: { normalized: "tool-calls", raw: undefined },
-          providerMetadata: { "openai-compatible": { responseId: "resp_1", serviceTier: undefined } },
+          providerMetadata: { "openai-compatible-responses": { responseId: "resp_1", serviceTier: undefined } },
        },
      ])
    }),
@@ -405,7 +423,7 @@ describe("Open Responses basic-item lifecycles", () => {
          type: "tool-input-end",
          id: "call_1",
          name: "lookup",
-          providerMetadata: { "openai-compatible": { itemId: "fc_1" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "fc_1" } },
        },
        {
          type: "tool-call",
@@ -413,7 +431,7 @@ describe("Open Responses basic-item lifecycles", () => {
          name: "lookup",
          input: { query: "final" },
          providerExecuted: undefined,
-          providerMetadata: { "openai-compatible": { itemId: "fc_1" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "fc_1" } },
        },
        { type: "reasoning-end", id: "rs_1:0" },
      ])
@@ -438,7 +456,7 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "reasoning-end",
          id: ":0",
-          providerMetadata: { "openai-compatible": { itemId: "", reasoningEncryptedContent: "state" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "", reasoningEncryptedContent: "state" } },
        },
      ])
    }),
@@ -474,7 +492,7 @@ describe("Open Responses basic-item lifecycles", () => {
        completed,
      )
      // Generic terminal closure does not repeat the message's phase metadata.
-      const providerMetadata = { "openai-compatible": { itemId: "fc_1" } }
+      const providerMetadata = { "openai-compatible-responses": { itemId: "fc_1" } }
      expect(events.slice(4, -2)).toEqual([
        { type: "tool-input-end", id: "call_1", name: "lookup", providerMetadata },
        { type: "tool-call", id: "call_1", name: "lookup", input: {}, providerMetadata },
@@ -518,7 +536,7 @@ describe("Open Responses basic-item lifecycles", () => {
        {
          type: "finish",
          reason: { normalized: "length", raw: "max_output_tokens" },
-          providerMetadata: { "openai-compatible": { responseId: "resp_1", serviceTier: undefined } },
+          providerMetadata: { "openai-compatible-responses": { responseId: "resp_1", serviceTier: undefined } },
        },
      ])
    }),
@@ -69,7 +69,7 @@ for (const item of cases) {
          expect(response.reasoning.length).toBeGreaterThan(0)
          expect(response.events.some(LLMEvent.is.reasoningDelta)).toBe(true)
          const metadata = response.message.content.find((part) => part.type === "reasoning")?.providerMetadata?.[
-            item.model.route.providerMetadataKey ?? String(item.model.provider)
+            item.model.route.id
          ]
          expect(metadata?.reasoningField).toBe(item.structured ? "reasoning" : "reasoning_content")
          expect(Array.isArray(metadata?.reasoningDetails)).toBe(item.structured)
@@ -128,11 +128,7 @@ for (const item of cases) {
          ).toMatch(/^Paris is sunny\.?$/)
          const details = events
            .filter(LLMEvent.is.reasoningEnd)
-            .map(
-              (event) =>
-                event.providerMetadata?.[item.model.route.providerMetadataKey ?? String(item.model.provider)]
-                  ?.reasoningDetails,
-            )
+            .map((event) => event.providerMetadata?.[item.model.route.id]?.reasoningDetails)
            .find(Array.isArray)
          expect(Array.isArray(details)).toBe(item.structured)
          if (!item.structured || !Array.isArray(details)) return
@@ -154,7 +154,7 @@ describe("OpenAI Chat route", () => {
              {
                type: "reasoning",
                text: "thinking",
-                providerMetadata: { openai: { reasoningField: "reasoning" } },
+                providerMetadata: { "openai-chat": { reasoningField: "reasoning" } },
              },
              { type: "text", text: "Hello" },
            ]),
@@ -180,7 +180,7 @@ describe("OpenAI Chat route", () => {
              {
                type: "reasoning",
                text: "thinking",
-                providerMetadata: { openai: { reasoningField: "reasoning_text" } },
+                providerMetadata: { "openai-chat": { reasoningField: "reasoning_text" } },
              },
              { type: "text", text: "Hello" },
            ]),
@@ -775,7 +775,7 @@ describe("OpenAI Chat route", () => {
        reasoningTokens: 0,
        totalTokens: 7,
        providerMetadata: {
-          openai: {
+          "openai-chat": {
            prompt_tokens: 5,
            completion_tokens: 2,
            total_tokens: 7,
@@ -868,7 +868,7 @@ describe("OpenAI Chat route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { openai: { reasoningDetails: [] } } },
+        { type: "reasoning", text: "", providerMetadata: { "openai-chat": { reasoningDetails: [] } } },
        {
          type: "text",
          text: "I can't help with that.",
@@ -917,7 +917,7 @@ describe("OpenAI Chat route", () => {
        expect(response.reasoning).toBe("thinking")
        expect(response.text).toBe("Hello")
        expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-          openai: { reasoningField: field },
+          "openai-chat": { reasoningField: field },
        })

        const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -926,10 +926,10 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("uses the configured provider metadata namespace for reasoning and usage", () =>
+  it.effect("uses the route ID for reasoning and usage metadata", () =>
    Effect.gen(function* () {
      const selected = LanguageModel.update(model, {
-        route: { ...model.route, providerMetadataKey: "vendor" },
+        route: model.route.with({ id: "vendor" }),
      })
      const details = [{ type: "reasoning.text", text: "thinking", signature: "signed" }]
      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
@@ -959,12 +959,9 @@ describe("OpenAI Chat route", () => {
    }),
  )

-  it.effect("falls back to the selected provider for the metadata namespace", () =>
+  it.effect("preserves the route metadata namespace when the selected provider changes", () =>
    Effect.gen(function* () {
-      const compatible = model.route.with({ provider: "deepseek" }).model({ id: "deepseek-chat" })
-      const selected = LanguageModel.update(compatible, {
-        route: { ...compatible.route, providerMetadataKey: undefined },
-      })
+      const selected = model.route.with({ provider: "deepseek" }).model({ id: "deepseek-chat" })
      const response = yield* LLMClient.generate(LLMRequest.update(request, { model: selected })).pipe(
        Effect.provide(
          fixedResponse(
@@ -979,10 +976,10 @@ describe("OpenAI Chat route", () => {
      )

      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        deepseek: { reasoningField: "reasoning_content" },
+        "openai-chat": { reasoningField: "reasoning_content" },
      })
      expect(response.usage?.providerMetadata).toEqual({
-        deepseek: { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
+        "openai-chat": { prompt_tokens: 5, completion_tokens: 2, total_tokens: 7 },
      })

      const replay = yield* compileRequest(LLM.request({ model: selected, messages: [response.message] }))
@@ -1007,7 +1004,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "vendor_reasoning" },
+        "openai-chat": { reasoningField: "vendor_reasoning" },
      })

      const replay = yield* compileRequest(LLM.request({ model: custom, messages: [response.message] }))
@@ -1050,7 +1047,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: details },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: details },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -1093,7 +1090,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningDetails: details },
+        "openai-chat": { reasoningDetails: details },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -1118,7 +1115,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: details },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: details },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -1145,7 +1142,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: details },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: details },
      })
    }),
  )
@@ -1171,7 +1168,7 @@ describe("OpenAI Chat route", () => {
      expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
      expect(response.message.content.filter((part) => part.type === "reasoning")).toHaveLength(1)
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: details },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: details },
      })
    }),
  )
@@ -1197,7 +1194,7 @@ describe("OpenAI Chat route", () => {
      expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
      expect(response.message.content.filter((part) => part.type === "reasoning")).toHaveLength(1)
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: details },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: details },
      })
    }),
  )
@@ -1218,7 +1215,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("")
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningDetails: [] },
+        "openai-chat": { reasoningDetails: [] },
      })

      const replay = yield* compileRequest(LLM.request({ model, messages: [response.message] }))
@@ -1268,12 +1265,14 @@ describe("OpenAI Chat route", () => {
      expect(response.reasoning).toBe("thinking")
      expect(response.message.content.filter((part) => part.type === "reasoning")).toHaveLength(1)
      expect(response.message.content.find((part) => part.type === "reasoning")?.providerMetadata).toEqual({
-        openai: { reasoningField: "reasoning", reasoningDetails: merged },
+        "openai-chat": { reasoningField: "reasoning", reasoningDetails: merged },
      })
      expect(response.events.filter(LLMEvent.is.reasoningStart)).toHaveLength(1)
      expect(response.events.filter(LLMEvent.is.reasoningDelta)).toHaveLength(1)
      expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
-      expect(publishedEndMetadata).toEqual([{ openai: { reasoningField: "reasoning", reasoningDetails: merged } }])
+      expect(publishedEndMetadata).toEqual([
+        { "openai-chat": { reasoningField: "reasoning", reasoningDetails: merged } },
+      ])
      expect(response.events.findIndex(LLMEvent.is.reasoningStart)).toBeLessThan(
        response.events.findIndex(LLMEvent.is.textStart),
      )
@@ -1304,7 +1303,7 @@ describe("OpenAI Chat route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "", providerMetadata: { openai: { reasoningDetails: details } } },
+        { type: "reasoning", text: "", providerMetadata: { "openai-chat": { reasoningDetails: details } } },
      ])
      expect(response.events.filter(LLMEvent.is.reasoningStart)).toHaveLength(1)
      expect(response.events.filter(LLMEvent.is.reasoningEnd)).toHaveLength(1)
@@ -1330,7 +1329,7 @@ describe("OpenAI Chat route", () => {

      expect(response.reasoning).toBe("summary")
      expect(response.message.content).toEqual([
-        { type: "reasoning", text: "summary", providerMetadata: { openai: { reasoningDetails: details } } },
+        { type: "reasoning", text: "summary", providerMetadata: { "openai-chat": { reasoningDetails: details } } },
      ])
    }),
  )
@@ -1347,12 +1346,12 @@ describe("OpenAI Chat route", () => {
              {
                type: "reasoning",
                text: "first",
-                providerMetadata: { openai: { reasoningDetails: [first] } },
+                providerMetadata: { "openai-chat": { reasoningDetails: [first] } },
              },
              {
                type: "reasoning",
                text: "second",
-                providerMetadata: { openai: { reasoningField: "reasoning", reasoningDetails: [second] } },
+                providerMetadata: { "openai-chat": { reasoningField: "reasoning", reasoningDetails: [second] } },
              },
            ]),
          ],
@@ -1376,7 +1375,7 @@ describe("OpenAI Chat route", () => {
              {
                type: "reasoning",
                text: "A",
-                providerMetadata: { openai: { reasoningDetails: [detail] } },
+                providerMetadata: { "openai-chat": { reasoningDetails: [detail] } },
              },
              { type: "reasoning", text: "B" },
            ]),
@@ -405,7 +405,7 @@ describe("OpenAI-compatible Chat route", () => {
        outputTokens: undefined,
        totalTokens: undefined,
        providerMetadata: {
-          deepseek: {
+          "openai-compatible-chat": {
            prompt_tokens: null,
            completion_tokens: null,
            total_tokens: null,
@@ -276,19 +276,31 @@ describe("Open Responses-compatible route", () => {
          model,
          messages: [
            Message.assistant([
-              { type: "text", text: "Kept.", providerMetadata: { "openai-compatible": { itemId: "history_1" } } },
+              {
+                type: "text",
+                text: "Kept.",
+                providerMetadata: { "openai-compatible-responses": { itemId: "history_1" } },
+              },
              {
                type: "text",
                text: "Long.",
-                providerMetadata: { "openai-compatible": { itemId: `history_${"a".repeat(64)}` } },
+                providerMetadata: { "openai-compatible-responses": { itemId: `history_${"a".repeat(64)}` } },
              },
              {
                type: "text",
                text: "Opaque.",
-                providerMetadata: { "openai-compatible": { itemId: "provider_value/with+symbols" } },
+                providerMetadata: { "openai-compatible-responses": { itemId: "provider_value/with+symbols" } },
+              },
+              {
+                type: "text",
+                text: "No suffix.",
+                providerMetadata: { "openai-compatible-responses": { itemId: "msg_" } },
+              },
+              {
+                type: "text",
+                text: "No prefix.",
+                providerMetadata: { "openai-compatible-responses": { itemId: "_item" } },
              },
-              { type: "text", text: "No suffix.", providerMetadata: { "openai-compatible": { itemId: "msg_" } } },
-              { type: "text", text: "No prefix.", providerMetadata: { "openai-compatible": { itemId: "_item" } } },
            ]),
          ],
        }),
@@ -348,7 +360,7 @@ describe("Open Responses-compatible route", () => {
              name: item.type,
              result: { type: "json", value: item },
              providerExecuted: true,
-              providerMetadata: { example: { itemId: item.id } },
+              providerMetadata: { "openai-compatible-responses": { itemId: item.id } },
            }),
          ),
        }),
@@ -383,7 +395,7 @@ describe("Open Responses-compatible route", () => {
      )

      expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
+        { type: "text", text: "Indexed", providerMetadata: { "openai-compatible-responses": { itemId: "msg_1" } } },
      ])
    }),
  )
@@ -449,7 +461,7 @@ describe("Open Responses-compatible route", () => {
                ),
              )

-              const metadata = { "openai-compatible": { itemId: routing.id } }
+              const metadata = { "openai-compatible-responses": { itemId: routing.id } }
              if (fixture.item.type === "function_call") {
                expect(response.toolCalls).toEqual([
                  expect.objectContaining({
@@ -467,7 +479,10 @@ describe("Open Responses-compatible route", () => {
                    type: "reasoning",
                    text: "Preserved",
                    providerMetadata: {
-                      "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: "encrypted-state" },
+                      "openai-compatible-responses": {
+                        itemId: routing.id,
+                        reasoningEncryptedContent: "encrypted-state",
+                      },
                    },
                  },
                ])
@@ -519,13 +534,13 @@ describe("Open Responses-compatible route", () => {
            {
              type: "reasoning",
              text: "First.",
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
+              providerMetadata: { "openai-compatible-responses": { itemId: routing.id } },
            },
            {
              type: "reasoning",
              text: "Second.",
              providerMetadata: {
-                "openai-compatible": { itemId: routing.id, reasoningEncryptedContent: null },
+                "openai-compatible-responses": { itemId: routing.id, reasoningEncryptedContent: null },
              },
            },
          ])
@@ -534,7 +549,7 @@ describe("Open Responses-compatible route", () => {
              type: "reasoning-end",
              id: `${routing.id}:0`,
              text: undefined,
-              providerMetadata: { "openai-compatible": { itemId: routing.id } },
+              providerMetadata: { "openai-compatible-responses": { itemId: routing.id } },
            },
            { type: "reasoning-end", id: `${routing.id}:1` },
          ])
@@ -565,7 +580,7 @@ describe("Open Responses-compatible route", () => {
            id: "call_1",
            name: "lookup",
            input: { query: "complete" },
-            providerMetadata: { "openai-compatible": { itemId: "" } },
+            providerMetadata: { "openai-compatible-responses": { itemId: "" } },
          }),
        ])
      }),
@@ -592,7 +607,11 @@ describe("Open Responses-compatible route", () => {
        )

        expect(response.message.content).toEqual([
-          { type: "text", text: "Before after", providerMetadata: { "openai-compatible": { itemId: "msg_1" } } },
+          {
+            type: "text",
+            text: "Before after",
+            providerMetadata: { "openai-compatible-responses": { itemId: "msg_1" } },
+          },
        ])
        expect(response.events.map((event) => event.type)).toEqual([
          "step-start",
@@ -705,7 +724,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
        input: { query: "complete" },
-        providerMetadata: { example: { itemId: "item_1" } },
+        providerMetadata: { "openai-compatible-responses": { itemId: "item_1" } },
      })
    }),
  )
@@ -787,7 +806,7 @@ describe("Open Responses-compatible route", () => {
            Message.assistant({
              type: "text",
              text: "Unclassified.",
-              providerMetadata: { "openai-compatible": { phase: null } },
+              providerMetadata: { "openai-compatible-responses": { phase: null } },
            }),
          ],
        }),
@@ -846,7 +865,7 @@ describe("Open Responses-compatible route", () => {
        {
          type: "text",
          text: "I can't help with that.",
-          providerMetadata: { example: { itemId: "msg_refusal" } },
+          providerMetadata: { "openai-compatible-responses": { itemId: "msg_refusal" } },
        },
      ])

@@ -935,7 +954,7 @@ describe("Open Responses-compatible route", () => {

      expect(response.toolCalls).toEqual([])
      expect(response.events.find(LLMEvent.is.finish)).toMatchObject({
-        providerMetadata: { example: { responseId: "resp_1" } },
+        providerMetadata: { "openai-compatible-responses": { responseId: "resp_1" } },
      })
    }),
  )
@@ -1938,7 +1938,7 @@ describe("OpenAI Responses route", () => {
        reasoningTokens: 0,
        totalTokens: 7,
        providerMetadata: {
-          openai: {
+          "openai-responses": {
            input_tokens: 5,
            output_tokens: 2,
            total_tokens: 7,
@@ -1951,7 +1951,7 @@ describe("OpenAI Responses route", () => {
      expect(response.text).toBe("Hello!")
      expect(response.events).toEqual([
        { type: "step-start", index: 0 },
-        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
+        { type: "text-start", id: "msg_1", providerMetadata: { "openai-responses": { itemId: "msg_1" } } },
        { type: "text-delta", id: "msg_1", text: "Hello" },
        { type: "text-delta", id: "msg_1", text: "!" },
        { type: "text-end", id: "msg_1" },
@@ -1959,13 +1959,13 @@ describe("OpenAI Responses route", () => {
          type: "step-finish",
          index: 0,
          reason: { normalized: "stop", raw: undefined },
-          providerMetadata: { openai: { responseId: "resp_1", serviceTier: "default" } },
+          providerMetadata: { "openai-responses": { responseId: "resp_1", serviceTier: "default" } },
          usage,
        },
        {
          type: "finish",
          reason: { normalized: "stop", raw: undefined },
-          providerMetadata: { openai: { responseId: "resp_1", serviceTier: "default" } },
+          providerMetadata: { "openai-responses": { responseId: "resp_1", serviceTier: "default" } },
          usage,
        },
      ])
@@ -2040,7 +2040,7 @@ describe("OpenAI Responses route", () => {
        {
          type: "text",
          text: "I can't help with that.",
-          providerMetadata: { openai: { itemId: "msg_refusal", phase: "final_answer" } },
+          providerMetadata: { "openai-responses": { itemId: "msg_refusal", phase: "final_answer" } },
        },
      ])

@@ -2112,17 +2112,17 @@ describe("OpenAI Responses route", () => {
        {
          type: "text",
          text: "Checking.",
-          providerMetadata: { openai: { itemId: "msg_commentary", phase: "commentary" } },
+          providerMetadata: { "openai-responses": { itemId: "msg_commentary", phase: "commentary" } },
        },
        {
          type: "text",
          text: "Finished.",
-          providerMetadata: { openai: { itemId: "msg_final", phase: "final_answer" } },
+          providerMetadata: { "openai-responses": { itemId: "msg_final", phase: "final_answer" } },
        },
        {
          type: "text",
          text: "Unclassified.",
-          providerMetadata: { openai: { itemId: "msg_null", phase: null } },
+          providerMetadata: { "openai-responses": { itemId: "msg_null", phase: null } },
        },
      ])

@@ -2170,7 +2170,7 @@ describe("OpenAI Responses route", () => {

      expect(response.text).toBe("Indexed")
      expect(response.message.content).toEqual([
-        { type: "text", text: "Indexed", providerMetadata: { openai: { itemId: "msg_1" } } },
+        { type: "text", text: "Indexed", providerMetadata: { "openai-responses": { itemId: "msg_1" } } },
      ])
    }),
  )
@@ -2331,7 +2331,7 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning",
          text: "Thinking",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "state" } },
        },
      ])
    }),
@@ -2379,7 +2379,7 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning",
          text: "Raw",
-          providerMetadata: { openai: { itemId: "", reasoningEncryptedContent: "state" } },
+          providerMetadata: { "openai-responses": { itemId: "", reasoningEncryptedContent: "state" } },
        },
      ])
    }),
@@ -2588,12 +2588,12 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.events.filter((event) => event.type.startsWith("text-"))).toEqual([
-        { type: "text-start", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
+        { type: "text-start", id: "msg_1", providerMetadata: { "openai-responses": { itemId: "msg_1" } } },
        { type: "text-delta", id: "msg_1", text: "First", providerMetadata: undefined },
-        { type: "text-end", id: "msg_1", providerMetadata: { openai: { itemId: "msg_1" } } },
-        { type: "text-start", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
+        { type: "text-end", id: "msg_1", providerMetadata: { "openai-responses": { itemId: "msg_1" } } },
+        { type: "text-start", id: "msg_2", providerMetadata: { "openai-responses": { itemId: "msg_2" } } },
        { type: "text-delta", id: "msg_2", text: "Second", providerMetadata: undefined },
-        { type: "text-end", id: "msg_2", providerMetadata: { openai: { itemId: "msg_2" } } },
+        { type: "text-end", id: "msg_2", providerMetadata: { "openai-responses": { itemId: "msg_2" } } },
      ])
    }),
  )
@@ -2629,9 +2629,9 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning",
          text: "thinking",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
-        { type: "text", text: "Hello", providerMetadata: { openai: { itemId: "msg_1" } } },
+        { type: "text", text: "Hello", providerMetadata: { "openai-responses": { itemId: "msg_1" } } },
      ])
    }),
  )
@@ -2663,7 +2663,7 @@ describe("OpenAI Responses route", () => {
        expect.objectContaining({
          type: "reasoning-end",
          id: "rs_1:0",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        }),
      )
    }),
@@ -2715,7 +2715,7 @@ describe("OpenAI Responses route", () => {
      expect(response.message.content).toContainEqual({
        type: "reasoning",
        text: "Checked the diff.",
-        providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+        providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
      })
    }),
  )
@@ -2829,20 +2829,20 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning-start",
          id: "rs_1:0",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:0", text: "First" },
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
+        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { "openai-responses": { itemId: "rs_1" } } },
        {
          type: "reasoning-start",
          id: "rs_1:1",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:1", text: "Second" },
        {
          type: "reasoning-end",
          id: "rs_1:1",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        },
        { type: "step-finish", index: 0, reason: { normalized: "stop", raw: undefined } },
        { type: "finish", reason: { normalized: "stop", raw: undefined } },
@@ -2887,31 +2887,31 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning-start",
          id: "rs_1:0",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:0", text: "First", providerMetadata: undefined },
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
+        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { "openai-responses": { itemId: "rs_1" } } },
        {
          type: "reasoning-start",
          id: "rs_1:1",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:1", text: "Second", providerMetadata: undefined },
        {
          type: "reasoning-end",
          id: "rs_1:1",
-          providerMetadata: { openai: { itemId: "rs_1" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1" } },
        },
        {
          type: "reasoning-start",
          id: "rs_1:2",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:2", text: "Third", providerMetadata: undefined },
        {
          type: "reasoning-end",
          id: "rs_1:2",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        },
      ])
    }),
@@ -3093,14 +3093,14 @@ describe("OpenAI Responses route", () => {
        {
          type: "reasoning-start",
          id: "rs_1:0",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: null } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: null } },
        },
        { type: "reasoning-delta", id: "rs_1:0", text: "Checked the diff.", providerMetadata: undefined },
        {
          type: "reasoning-end",
          id: "rs_1:0",
          text: "Checked the diff.",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        },
      ])

@@ -3178,11 +3178,11 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.events.filter((event) => event.type === "reasoning-end")).toEqual([
-        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { openai: { itemId: "rs_1" } } },
+        { type: "reasoning-end", id: "rs_1:0", providerMetadata: { "openai-responses": { itemId: "rs_1" } } },
        {
          type: "reasoning-end",
          id: "rs_1:1",
-          providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+          providerMetadata: { "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
        },
      ])
    }),
@@ -3201,7 +3201,7 @@ describe("OpenAI Responses route", () => {
                type: "reasoning",
                text: "Checked the previous diff.",
                providerMetadata: {
-                  openai: {
+                  "openai-responses": {
                    itemId: "rs_1",
                    reasoningEncryptedContent: "encrypted-state",
                  },
@@ -3262,7 +3262,7 @@ describe("OpenAI Responses route", () => {
                type: "reasoning",
                text: "Checked order.",
                providerMetadata: {
-                  openai: {
+                  "openai-responses": {
                    itemId: "rs_1",
                    reasoningEncryptedContent: "encrypted-state",
                  },
@@ -3298,7 +3298,9 @@ describe("OpenAI Responses route", () => {
              {
                type: "reasoning",
                text: "Checked the previous diff.",
-                providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+                providerMetadata: {
+                  "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+                },
              },
            ]),
          ],
@@ -3330,7 +3332,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                input: { query: "effect 4" },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              }),
              {
                type: "tool-result",
@@ -3338,7 +3340,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                result: { type: "json", value: item },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              },
            ]),
            Message.user("Continue."),
@@ -3368,7 +3370,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                input: { query: "effect 4" },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              }),
              {
                type: "tool-result",
@@ -3376,7 +3378,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                result: { type: "json", value: item },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              },
            ]),
            Message.user("Continue."),
@@ -3410,7 +3412,7 @@ describe("OpenAI Responses route", () => {
              name: item.type,
              result: { type: "json", value: item },
              providerExecuted: true,
-              providerMetadata: { openai: { itemId: item.id } },
+              providerMetadata: { "openai-responses": { itemId: item.id } },
            }),
          ),
        }),
@@ -3437,7 +3439,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                input: { query: "effect 4" },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              }),
              {
                type: "tool-result",
@@ -3445,7 +3447,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                result: { type: "json", value: item },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              },
            ]),
            Message.user("Continue."),
@@ -3477,7 +3479,7 @@ describe("OpenAI Responses route", () => {
                name: "web_search",
                result: { type: "json", value: { type: "web_search_call", id: "ws_other", status: "completed" } },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ws_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ws_1" } },
              },
            ]),
          ],
@@ -3503,39 +3505,41 @@ describe("OpenAI Responses route", () => {
              {
                type: "text",
                text: "Hello",
-                providerMetadata: { openai: { itemId: "history_1" } },
+                providerMetadata: { "openai-responses": { itemId: "history_1" } },
              },
              {
                type: "text",
                text: "World",
-                providerMetadata: { openai: { itemId: `message_${"a".repeat(64)}` } },
+                providerMetadata: { "openai-responses": { itemId: `message_${"a".repeat(64)}` } },
              },
              {
                type: "reasoning",
                text: "Checked the diff.",
-                providerMetadata: { openai: { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" } },
+                providerMetadata: {
+                  "openai-responses": { itemId: "thinking_1", reasoningEncryptedContent: "encrypted-state" },
+                },
              },
              {
                type: "reasoning",
                text: "Missing suffix.",
-                providerMetadata: { openai: { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
+                providerMetadata: { "openai-responses": { itemId: "rs_", reasoningEncryptedContent: "another-state" } },
              },
              {
                type: "reasoning",
                text: "No prefix separator.",
-                providerMetadata: { openai: { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
+                providerMetadata: { "openai-responses": { itemId: "550e8400-e29b-41d4-a716-446655440000" } },
              },
              ToolCallPart.make({
                id: "call_1",
                name: "lookup",
                input: { query: "weather" },
-                providerMetadata: { openai: { itemId: "toolu_01A" } },
+                providerMetadata: { "openai-responses": { itemId: "toolu_01A" } },
              }),
              ToolCallPart.make({
                id: "call_2",
                name: "lookup",
                input: { query: "news" },
-                providerMetadata: { openai: { itemId: "fc_" } },
+                providerMetadata: { "openai-responses": { itemId: "fc_" } },
              }),
            ]),
          ],
@@ -3597,7 +3601,7 @@ describe("OpenAI Responses route", () => {
            name: "web_search",
            input: { query: "effect 4" },
            providerExecuted: true,
-            providerMetadata: { openai: { itemId } },
+            providerMetadata: { "openai-responses": { itemId } },
          }),
          {
            type: "tool-result" as const,
@@ -3605,7 +3609,7 @@ describe("OpenAI Responses route", () => {
            name: "web_search",
            result: { type: "json" as const, value: item },
            providerExecuted: true as const,
-            providerMetadata: { openai: { itemId } },
+            providerMetadata: { "openai-responses": { itemId } },
          },
        ]
      }
@@ -3641,7 +3645,7 @@ describe("OpenAI Responses route", () => {
                name: "image_generation",
                input: {},
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ig_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ig_1" } },
              }),
              ToolResultPart.make({
                id: "ig_1",
@@ -3651,7 +3655,7 @@ describe("OpenAI Responses route", () => {
                  value: [{ type: "file", uri: "data:image/png;base64,AQID", mime: "image/png" }],
                },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ig_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ig_1" } },
              }),
            ]),
            Message.user("Make it blue."),
@@ -3682,7 +3686,7 @@ describe("OpenAI Responses route", () => {
                name: "image_generation",
                input: {},
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ig_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ig_1" } },
              }),
              ToolResultPart.make({
                id: "ig_1",
@@ -3692,7 +3696,7 @@ describe("OpenAI Responses route", () => {
                  value: [{ type: "file", uri: "data:image/png;base64,AQID", mime: "image/png" }],
                },
                providerExecuted: true,
-                providerMetadata: { openai: { itemId: "ig_1" } },
+                providerMetadata: { "openai-responses": { itemId: "ig_1" } },
              }),
            ]),
          ],
@@ -3717,12 +3721,14 @@ describe("OpenAI Responses route", () => {
              {
                type: "reasoning",
                text: "First",
-                providerMetadata: { openai: { itemId: "rs_1" } },
+                providerMetadata: { "openai-responses": { itemId: "rs_1" } },
              },
              {
                type: "reasoning",
                text: "Second",
-                providerMetadata: { openai: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+                providerMetadata: {
+                  "openai-responses": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" },
+                },
              },
            ]),
          ],
@@ -3757,7 +3763,7 @@ describe("OpenAI Responses route", () => {
                type: "reasoning",
                text: "Checked the previous diff.",
                providerMetadata: {
-                  openai: {
+                  "openai-responses": {
                    itemId: "rs_1",
                    reasoningEncryptedContent: null,
                  },
@@ -3821,7 +3827,7 @@ describe("OpenAI Responses route", () => {
        cacheReadInputTokens: undefined,
        reasoningTokens: undefined,
        totalTokens: 6,
-        providerMetadata: { openai: { input_tokens: 5, output_tokens: 1 } },
+        providerMetadata: { "openai-responses": { input_tokens: 5, output_tokens: 1 } },
      })

      expect(response.events).toEqual([
@@ -3830,7 +3836,7 @@ describe("OpenAI Responses route", () => {
          type: "tool-input-start",
          id: "call_1",
          name: "lookup",
-          providerMetadata: { openai: { itemId: "fc_item_1" } },
+          providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
        },
        {
          type: "tool-input-delta",
@@ -3850,7 +3856,7 @@ describe("OpenAI Responses route", () => {
          type: "tool-input-end",
          id: "call_1",
          name: "lookup",
-          providerMetadata: { openai: { itemId: "fc_item_1" } },
+          providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
        },
        {
          type: "tool-call",
@@ -3858,7 +3864,7 @@ describe("OpenAI Responses route", () => {
          name: "lookup",
          input: { query: "weather" },
          providerExecuted: undefined,
-          providerMetadata: { openai: { itemId: "fc_item_1" } },
+          providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
        },
        {
          type: "step-finish",
@@ -3927,7 +3933,7 @@ describe("OpenAI Responses route", () => {
          name: "lookup",
          input: { query: "weather" },
          providerExecuted: undefined,
-          providerMetadata: { openai: { itemId: "fc_item_1" } },
+          providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
        },
      ])
    }),
@@ -4104,7 +4110,7 @@ describe("OpenAI Responses route", () => {
      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
        id: "call_1",
        input: { query: "weather" },
-        providerMetadata: { openai: { itemId: "fc_item_1" } },
+        providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
      })
      expect(response.events.filter(LLMEvent.is.toolInputEnd)).toHaveLength(1)
      expect(response.events.filter(LLMEvent.is.toolCall)).toHaveLength(1)
@@ -4228,7 +4234,7 @@ describe("OpenAI Responses route", () => {
            type: "tool-input-end",
            id: "call_1",
            name: "lookup",
-            providerMetadata: { openai: { itemId: "fc_item_1" } },
+            providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
          },
          {
            type: "tool-call",
@@ -4236,7 +4242,7 @@ describe("OpenAI Responses route", () => {
            name: "lookup",
            input: {},
            providerExecuted: undefined,
-            providerMetadata: { openai: { itemId: "fc_item_1" } },
+            providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
          },
        ],
      )
@@ -4330,7 +4336,7 @@ describe("OpenAI Responses route", () => {

      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
        id: "call_1",
-        providerMetadata: { openai: { itemId: "fc_item_1" } },
+        providerMetadata: { "openai-responses": { itemId: "fc_item_1" } },
      })
    }),
  )
@@ -4360,7 +4366,7 @@ describe("OpenAI Responses route", () => {
          name: "web_search",
          input: { type: "search", query: "effect 4" },
          providerExecuted: true,
-          providerMetadata: { openai: { itemId: "ws_1" } },
+          providerMetadata: { "openai-responses": { itemId: "ws_1" } },
        },
        {
          type: "tool-result",
@@ -4368,7 +4374,7 @@ describe("OpenAI Responses route", () => {
          name: "web_search",
          result: { type: "json", value: item },
          providerExecuted: true,
-          providerMetadata: { openai: { itemId: "ws_1" } },
+          providerMetadata: { "openai-responses": { itemId: "ws_1" } },
        },
      ])
    }),
@@ -4401,7 +4407,7 @@ describe("OpenAI Responses route", () => {
          name: "computer_use",
          input: { type: "click", x: 100, y: 200 },
          providerExecuted: true,
-          providerMetadata: { openai: { itemId: "computer_1" } },
+          providerMetadata: { "openai-responses": { itemId: "computer_1" } },
        },
        {
          type: "tool-result",
@@ -4409,7 +4415,7 @@ describe("OpenAI Responses route", () => {
          name: "computer_use",
          result: { type: "json", value: item },
          providerExecuted: true,
-          providerMetadata: { openai: { itemId: "computer_1" } },
+          providerMetadata: { "openai-responses": { itemId: "computer_1" } },
        },
      ])
    }),
@@ -4437,7 +4443,7 @@ describe("OpenAI Responses route", () => {
      )

      expect(response.events.find(LLMEvent.is.toolCall)).toMatchObject({
-        providerMetadata: { openai: { itemId: "ig_1" } },
+        providerMetadata: { "openai-responses": { itemId: "ig_1" } },
      })
      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
        id: "ig_1",
@@ -4447,7 +4453,7 @@ describe("OpenAI Responses route", () => {
          type: "content",
          value: [{ type: "file", uri: "data:image/png;base64,AQID", mime: "image/png" }],
        },
-        providerMetadata: { openai: { itemId: "ig_1" } },
+        providerMetadata: { "openai-responses": { itemId: "ig_1" } },
      })

      const prepared = yield* Effect.forEach([false, true], (store) =>
@@ -4481,7 +4487,7 @@ describe("OpenAI Responses route", () => {

      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
        result: { type: "error", value: item.error },
-        providerMetadata: { openai: { itemId: "ws_failed" } },
+        providerMetadata: { "openai-responses": { itemId: "ws_failed" } },
      })

      const prepared = yield* compileRequest(
@@ -4541,7 +4547,7 @@ describe("OpenAI Responses route", () => {
        name: "code_interpreter",
        input: { code: "print(1+1)", container_id: "cnt_xyz" },
        providerExecuted: true,
-        providerMetadata: { openai: { itemId: "ci_1" } },
+        providerMetadata: { "openai-responses": { itemId: "ci_1" } },
      })
      const toolResult = response.events.find((event) => event.type === "tool-result")
      expect(toolResult).toEqual({
@@ -4550,7 +4556,7 @@ describe("OpenAI Responses route", () => {
        name: "code_interpreter",
        result: { type: "json", value: item },
        providerExecuted: true,
-        providerMetadata: { openai: { itemId: "ci_1" } },
+        providerMetadata: { "openai-responses": { itemId: "ci_1" } },
      })
    }),
  )
@@ -65,7 +65,7 @@ describe("xAI Responses route", () => {
      expect(response.message.content.find((part) => part.type === "reasoning")).toMatchObject({
        type: "reasoning",
        text: "Considering.",
-        providerMetadata: { xai: { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
+        providerMetadata: { "xai-responses": { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
      })
    }),
  )
@@ -101,7 +101,7 @@ describe("xAI Responses route", () => {

      expect(response.reasoning).toBe("Considering.")
      expect(response.message.content.find((part) => part.type === "reasoning")).toMatchObject({
-        providerMetadata: { xai: { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
+        providerMetadata: { "xai-responses": { itemId: "reasoning_1", reasoningEncryptedContent: "opaque" } },
      })
    }),
  )
@@ -120,7 +120,7 @@ describe("xAI Responses route", () => {
                name: "x_search",
                result: { type: "json", value: item },
                providerExecuted: true,
-                providerMetadata: { xai: { itemId: "x_search_1" } },
+                providerMetadata: { "xai-responses": { itemId: "x_search_1" } },
              },
            ]),
          ],
@@ -148,7 +148,7 @@ describe("xAI Responses route", () => {
              name: item.type,
              result: { type: "json", value: item },
              providerExecuted: true,
-              providerMetadata: { xai: { itemId: item.id } },
+              providerMetadata: { "xai-responses": { itemId: item.id } },
            }),
          ),
        }),
@@ -181,11 +181,11 @@ describe("xAI Responses route", () => {
        name: "x_search",
        input: { query: "news" },
        providerExecuted: true,
-        providerMetadata: { xai: { itemId: "x_search_1" } },
+        providerMetadata: { "xai-responses": { itemId: "x_search_1" } },
      })
      expect(response.events.find(LLMEvent.is.toolResult)).toMatchObject({
        result: { type: "json", value: item },
-        providerMetadata: { xai: { itemId: "x_search_1" } },
+        providerMetadata: { "xai-responses": { itemId: "x_search_1" } },
      })
    }),
  )
@@ -253,10 +253,10 @@ const assistantMessageFromResponse = (response: LLMResponse, step: AssistantStep
  if (step.reasoning === "openai-encrypted") {
    const reasoning = response.events.find(
      (event): event is Extract<LLMEvent, { readonly type: "reasoning-end" }> =>
-        LLMEvent.is.reasoningEnd(event) && typeof event.providerMetadata?.openai?.itemId === "string",
+        LLMEvent.is.reasoningEnd(event) && typeof event.providerMetadata?.["openai-responses"]?.itemId === "string",
    )
    if (!reasoning) throw new Error("OpenAI Responses did not return reasoning metadata")
-    expect(reasoning.providerMetadata?.openai?.reasoningEncryptedContent).toEqual(expect.any(String))
+    expect(reasoning.providerMetadata?.["openai-responses"]?.reasoningEncryptedContent).toEqual(expect.any(String))
    content.push({ type: "reasoning", text: response.reasoning, providerMetadata: reasoning.providerMetadata })
  }

@@ -47,14 +47,29 @@ describe("Route.with", () => {
    })
  })

-  test("assigns metadata ownership to a replacement provider and preserves explicit overrides", () => {
-    const route = OpenAIChat.route.with({ provider: "azure" })
-    const overridden = route.with({ providerMetadataKey: "custom-azure" }).with({ headers: { "x-test": "value" } })
-
-    expect(route.providerMetadataKey).toBe("azure")
-    expect(overridden.providerMetadataKey).toBe("custom-azure")
-    expect(overridden.defaults).not.toHaveProperty("providerMetadataKey")
-  })
+  it.effect("namespaces emitted metadata by route ID independently of provider identity", () =>
+    Effect.gen(function* () {
+      const base = OpenAIChat.route.with({ provider: "azure", auth: Auth.none })
+      for (const route of [base, base.with({ id: "custom-azure-chat" })]) {
+        const response = yield* LLMClient.generate(
+          LLM.request({ model: route.model({ id: "test" }), prompt: "Hello" }),
+        ).pipe(
+          Effect.provide(
+            fixedResponse(
+              sseEvents({
+                choices: [{ delta: { content: "Hello" }, finish_reason: "stop" }],
+                usage: { prompt_tokens: 2, completion_tokens: 1, total_tokens: 3 },
+              }),
+            ),
+          ),
+        )
+        expect(response.usage?.providerMetadata).toEqual({
+          [route.id]: { prompt_tokens: 2, completion_tokens: 1, total_tokens: 3 },
+        })
+        expect(route).not.toHaveProperty("providerMetadataKey")
+      }
+    }),
+  )
 })

 describe("Route diagnostics", () => {
@@ -309,7 +309,6 @@ function modelFromLanguage(info: Info, language: LanguageModelV3) {
    compact: undefined,
    id: `ai-sdk:${packageName}`,
    provider: ProviderID.make(providerID),
-    providerMetadataKey: optionKey,
    protocol: "ai-sdk",
    endpoint: Endpoint.path("/", { baseURL: "https://ai-sdk.local" }),
    auth: Auth.none,
@@ -342,7 +341,16 @@ function modelFromLanguage(info: Info, language: LanguageModelV3) {
    model: (input) =>
      LanguageModel.make({ ...input, provider: "provider" in input ? input.provider : providerID, route }),
    prepareTransport: (body) => Effect.succeed(body),
-    streamPrepared: (prepared) => streamLanguage(language, prepared as LanguageModelV3CallOptions),
+    streamPrepared: (prepared) =>
+      streamLanguage(language, prepared as LanguageModelV3CallOptions).pipe(
+        Stream.map((event) => {
+          if (!("providerMetadata" in event)) return event
+          // Preserve the same inner state that Session persists, while keeping SDK
+          // option namespaces private to this bridge.
+          const state = event.providerMetadata?.[optionKey]
+          return { ...event, providerMetadata: state === undefined ? undefined : { [route.id]: state } }
+        }),
+      ),
  }
  return LanguageModel.make({
    id: info.modelID ?? info.id,
@@ -418,7 +426,7 @@ function callOptions(
 ): LanguageModelV3CallOptions {
  const flattened = ProviderShared.flattenToolRequest(request)
  return {
-    prompt: prompt(flattened.request),
+    prompt: prompt(flattened.request, optionKey),
    maxOutputTokens: request.generation?.maxTokens,
    temperature: request.generation?.temperature,
    stopSequences: request.generation?.stop === undefined ? undefined : [...request.generation.stop],
@@ -434,15 +442,21 @@ function callOptions(
  }
 }

-function prompt(request: LLMRequest): LanguageModelV3Prompt {
+type MetadataOptions = (input: ProviderMetadata | undefined) => SharedV3ProviderOptions | undefined
+
+function prompt(request: LLMRequest, optionKey: string): LanguageModelV3Prompt {
+  const metadata: MetadataOptions = (input) => {
+    const state = input?.[request.model.route.id]
+    return state === undefined ? undefined : { [optionKey]: jsonObject(state) }
+  }
  const system = request.system
    .map((part) => part.text)
    .filter(Boolean)
    .join("\n\n")
  const pending: UserContent = []
  const messages = request.messages.flatMap((input, index) => {
-    if (input.role !== "tool") return message(input)
-    const lowered = toolMessage(input)
+    if (input.role !== "tool") return message(input, metadata)
+    const lowered = toolMessage(input, metadata)
    pending.push(...lowered.media)
    if (request.messages[index + 1]?.role === "tool" || pending.length === 0) return lowered.messages
    const media = [...pending]
@@ -459,7 +473,7 @@ function prompt(request: LLMRequest): LanguageModelV3Prompt {
  return [{ role: "system", content: system }, ...messages]
 }

-function message(input: LLMRequest["messages"][number]): LanguageModelV3Message[] {
+function message(input: LLMRequest["messages"][number], metadata: MetadataOptions): LanguageModelV3Message[] {
  switch (input.role) {
    case "system":
      // The initial privileged prompt lives in `request.system` and is prepended above. A system message here is a
@@ -479,29 +493,32 @@ function message(input: LLMRequest["messages"][number]): LanguageModelV3Message[
    case "user":
      return [{ role: "user", content: input.content.flatMap(userPart) }]
    case "assistant":
-      return [{ role: "assistant", content: input.content.flatMap(assistantPart) }]
+      return [{ role: "assistant", content: input.content.flatMap((part) => assistantPart(part, metadata)) }]
    case "tool":
-      return toolMessage(input).messages
+      return toolMessage(input, metadata).messages
  }
 }

-function toolMessage(input: LLMRequest["messages"][number]) {
+function toolMessage(input: LLMRequest["messages"][number], metadata: MetadataOptions) {
  const media: UserContent = []
  const content = input.content.flatMap((part) => {
-    if (part.type !== "tool-result" || part.result.type !== "content") return toolResultPart(part)
+    if (part.type !== "tool-result" || part.result.type !== "content") return toolResultPart(part, metadata)
    const value = part.result.value.filter((item) => {
      if (item.type !== "file") return true
      if (!item.mime.startsWith("image/") && item.mime !== "application/pdf") return true
      media.push({ type: "file", mediaType: item.mime, data: fileData(item.uri), filename: item.name })
      return false
    })
-    return toolResultPart({
-      ...part,
-      result:
-        value.length === 0
-          ? { type: "text", value: "Media attached in the following user message." }
-          : { ...part.result, value },
-    })
+    return toolResultPart(
+      {
+        ...part,
+        result:
+          value.length === 0
+            ? { type: "text", value: "Media attached in the following user message." }
+            : { ...part.result, value },
+      },
+      metadata,
+    )
  })
  return {
    messages: content.length ? ([{ role: "tool", content }] satisfies LanguageModelV3Message[]) : [],
@@ -520,7 +537,7 @@ function userPart(part: ContentPart): UserContent {
  return []
 }

-function assistantPart(part: ContentPart): AssistantContent {
+function assistantPart(part: ContentPart, metadata: MetadataOptions): AssistantContent {
  switch (part.type) {
    case "compaction":
      throw ProviderShared.unsupportedOperation({
@@ -529,11 +546,11 @@ function assistantPart(part: ContentPart): AssistantContent {
        message: "AI SDK routes cannot replay native provider compaction state",
      })
    case "text":
-      return [{ type: "text", text: part.text, providerOptions: metadataProviderOptions(part.providerMetadata) }]
+      return [{ type: "text", text: part.text, providerOptions: metadata(part.providerMetadata) }]
    case "media":
      return [{ type: "file", mediaType: part.mediaType, data: fileData(part.data), filename: part.filename }]
    case "reasoning":
-      return [{ type: "reasoning", text: part.text, providerOptions: metadataProviderOptions(part.providerMetadata) }]
+      return [{ type: "reasoning", text: part.text, providerOptions: metadata(part.providerMetadata) }]
    case "tool-call":
      return [
        {
@@ -542,11 +559,11 @@ function assistantPart(part: ContentPart): AssistantContent {
          toolName: part.name,
          input: part.input,
          providerExecuted: part.providerExecuted,
-          providerOptions: metadataProviderOptions(part.providerMetadata),
+          providerOptions: metadata(part.providerMetadata),
        },
      ]
    case "tool-result":
-      return toolResultPart(part)
+      return toolResultPart(part, metadata)
  }
 }

@@ -559,7 +576,7 @@ function fileData(data: Extract<ContentPart, { type: "media" }>["data"]) {
  return url.protocol === "http:" || url.protocol === "https:" ? url : data
 }

-function toolResultPart(part: ContentPart): ToolResultContent[] {
+function toolResultPart(part: ContentPart, metadata: MetadataOptions): ToolResultContent[] {
  if (part.type !== "tool-result") return []
  return [
    {
@@ -567,7 +584,7 @@ function toolResultPart(part: ContentPart): ToolResultContent[] {
      toolCallId: part.id,
      toolName: part.name,
      output: toolOutput(part.result),
-      providerOptions: metadataProviderOptions(part.providerMetadata),
+      providerOptions: metadata(part.providerMetadata),
    },
  ]
 }
@@ -624,11 +641,6 @@ function requestProviderOptions(
  return { [optionKey]: options }
 }

-function metadataProviderOptions(input: ProviderMetadata | undefined): SharedV3ProviderOptions | undefined {
-  if (!input) return undefined
-  return Object.fromEntries(Object.entries(input).map(([key, value]) => [key, jsonObject(value)]))
-}
-
 function streamLanguage(language: LanguageModelV3, options: LanguageModelV3CallOptions) {
  const state = { step: 0, toolNames: {} as Record<string, string> }
  return Stream.concat(
@@ -147,7 +147,7 @@ export const estimateTokens = (input: RequiredInput) => {
  const last = input.messages[index]
  // Keep the anchor's local tool results: they are not covered by its provider usage.
  const added = SessionModelRequest.unsupportedParts(
-    toLLMMessages(input.messages.slice(Math.max(0, index)), input.resolved.ref),
+    toLLMMessages(input.messages.slice(Math.max(0, index)), input.resolved.ref, input.resolved.model.route.id),
    input.resolved.capabilities,
  )
    .filter((message) => message.role !== "assistant" || message.id !== last?.id)
@@ -450,8 +450,7 @@ export const layer = Layer.effect(
              })
            }
            if (LLMEvent.is.stepFinish(event)) {
-              providerState =
-                event.providerMetadata?.[context.model.model.route.providerMetadataKey ?? context.model.model.provider]
+              providerState = event.providerMetadata?.[context.model.model.route.id]
              const step = SessionUsage.record(event.usage, context.model.cost)
              usage = usage ? SessionUsage.add(usage, step) : step
            }
@@ -92,9 +92,7 @@ export const baseTranscript = (input: {
  readonly initial: string
  readonly messages: ReadonlyArray<SessionMessage.Info>
 }) => {
-  const providerMetadataKey = input.model.model.route.providerMetadataKey ?? input.model.model.provider
  return {
-    providerMetadataKey,
    system: [
      input.agent.system
        ? input.agent.system
@@ -103,7 +101,7 @@ export const baseTranscript = (input: {
    ]
      .filter((part) => part.length > 0)
      .map(SystemPart.make),
-    messages: toLLMMessages(input.messages, input.model.ref, providerMetadataKey),
+    messages: toLLMMessages(input.messages, input.model.ref, input.model.model.route.id),
  }
 }

@@ -18,7 +18,7 @@ type Input = {
  readonly sessionID: SessionSchema.ID
  readonly agent: Agent.ID
  readonly model: Model.Ref
-  readonly providerMetadataKey: string
+  readonly routeID: string
  readonly snapshot?: Snapshot.ID
  readonly assistantMessageID: SessionMessage.ID
 }
@@ -115,7 +115,7 @@ export const createLLMEventPublisher = (bus: Pick<Bus.Interface, "publish">, inp
      assistantMessageID: yield* startAssistant(),
    })
  })
-  const providerState = (metadata: ProviderMetadata | undefined) => metadata?.[input.providerMetadataKey]
+  const providerState = (metadata: ProviderMetadata | undefined) => metadata?.[input.routeID]
  const fragments = (
    name: string,
    ended: (id: string, value: string, ordinal: number, state?: Record<string, unknown>) => Effect.Effect<void>,
@@ -75,7 +75,7 @@ export const make = Effect.gen(function* () {
      assistantMessageID: input.assistantMessageID,
      agent: input.agent,
      model: input.model.ref,
-      providerMetadataKey: input.model.model.route.providerMetadataKey ?? input.model.model.provider,
+      routeID: input.model.model.route.id,
      snapshot: startSnapshot,
    })
    const toolRuns: Array<{
@@ -95,10 +95,8 @@ const userAttachmentContent = (files: readonly FileAttachment[]) => {

 const decodeToolInput = Schema.decodeUnknownOption(Schema.fromJsonString(Schema.Unknown))

-const providerMetadata = (
-  provider: string,
-  state: Record<string, unknown> | undefined,
-): ProviderMetadata | undefined => (state === undefined ? undefined : { [provider]: state })
+const providerMetadata = (routeID: string, state: Record<string, unknown> | undefined): ProviderMetadata | undefined =>
+  state === undefined ? undefined : { [routeID]: state }

 const toolInput = (tool: SessionMessage.AssistantTool) =>
  tool.state.status === "streaming"
@@ -142,7 +140,7 @@ const toolResult = (tool: SessionMessage.AssistantTool, providerMetadata: Provid
  }
 }

-const assistant = (message: SessionMessage.Assistant, model: Model.Ref, providerMetadataKey: string) => {
+const assistant = (message: SessionMessage.Assistant, model: Model.Ref, routeID: string) => {
  const sameProvider = String(message.model.providerID) === String(model.providerID)
  const sameModel = sameProvider && String(message.model.id) === String(model.id)
  const reuseProviderMetadata = sameModel && message.error === undefined
@@ -154,7 +152,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
          text: item.text,
          // Text can carry provider-bound state (e.g. Gemini thought signatures),
          // which is only replayable against the model that produced it.
-          providerMetadata: reuseProviderMetadata ? providerMetadata(providerMetadataKey, item.state) : undefined,
+          providerMetadata: reuseProviderMetadata ? providerMetadata(routeID, item.state) : undefined,
        },
      ]
    // Let the destination adapter handle readable reasoning after a model/provider switch.
@@ -164,7 +162,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
            {
              type: "reasoning",
              text: item.text,
-              providerMetadata: providerMetadata(providerMetadataKey, item.state),
+              providerMetadata: providerMetadata(routeID, item.state),
            },
          ]
        : item.text.length > 0
@@ -176,10 +174,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
    const reuseToolProviderMetadata =
      reuseProviderMetadata ||
      (sameModel && item.executed === true && (item.state.status === "completed" || item.state.status === "error"))
-    const call = toolCall(
-      item,
-      reuseToolProviderMetadata ? providerMetadata(providerMetadataKey, item.providerState) : undefined,
-    )
+    const call = toolCall(item, reuseToolProviderMetadata ? providerMetadata(routeID, item.providerState) : undefined)
    if (item.executed !== true) return [call]
    // Hosted tools (e.g. google_search) run inside the provider, so their
    // result payload (`providerResultState`) is provider-format data rather
@@ -190,9 +185,9 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
    const result = toolResult(
      item,
      reuseToolProviderMetadata
-        ? providerMetadata(providerMetadataKey, item.providerResultState ?? item.providerState)
+        ? providerMetadata(routeID, item.providerResultState ?? item.providerState)
        : sameProvider && item.providerResultState !== undefined
-          ? providerMetadata(providerMetadataKey, item.providerResultState)
+          ? providerMetadata(routeID, item.providerResultState)
          : undefined,
    )
    return result ? [call, result] : [call]
@@ -207,9 +202,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
    .map((item) =>
      toolResult(
        item,
-        reuseProviderMetadata
-          ? providerMetadata(providerMetadataKey, item.providerResultState ?? item.providerState)
-          : undefined,
+        reuseProviderMetadata ? providerMetadata(routeID, item.providerResultState ?? item.providerState) : undefined,
      ),
    )
    .filter((message) => message !== undefined)
@@ -221,7 +214,7 @@ const assistant = (message: SessionMessage.Assistant, model: Model.Ref, provider
  ]
 }

-function toLLMMessage(message: SessionMessage.Info, model: Model.Ref, providerMetadataKey: string): Message[] {
+function toLLMMessage(message: SessionMessage.Info, model: Model.Ref, routeID: string): Message[] {
  switch (message.type) {
    case "agent-switched":
    case "model-switched":
@@ -271,7 +264,7 @@ function toLLMMessage(message: SessionMessage.Info, model: Model.Ref, providerMe
        }),
      ]
    case "assistant":
-      return assistant(message, model, providerMetadataKey)
+      return assistant(message, model, routeID)
    case "compaction":
      if (message.status !== "completed") return []
      return [
@@ -296,8 +289,5 @@ ${message.recent}
 }

 /** Translate projected Session history into canonical @opencode-ai/ai context. */
-export const toLLMMessages = (
-  messages: readonly SessionMessage.Info[],
-  model: Model.Ref,
-  providerMetadataKey: string = model.providerID,
-) => messages.flatMap((message) => toLLMMessage(message, model, providerMetadataKey))
+export const toLLMMessages = (messages: readonly SessionMessage.Info[], model: Model.Ref, routeID: string) =>
+  messages.flatMap((message) => toLLMMessage(message, model, routeID))
@@ -145,7 +145,7 @@ it.effect("uses canonical names and metadata without merging connection cache pa

    const resolved = yield* aisdk.model(input)
    expect(resolved).toMatchObject({ id: "api-model", provider: "openai" })
-    expect(resolved.route).toMatchObject({ provider: "openai", providerMetadataKey: "openai" })
+    expect(resolved.route).toMatchObject({ provider: "openai", id: "ai-sdk:@ai-sdk/openai-compatible" })
    expect(resolved.route.model({ id: "another-model" })).toMatchObject({ provider: "openai" })
    const prepared = yield* compileRequest(LLM.request({ model: resolved, prompt: "Hello" }))
    expect(prepared.body.providerOptions).toEqual({ openai: { reasoningEffort: "high" } })
@@ -385,22 +385,26 @@ it.effect("projects replay metadata onto AI SDK prompt parts", () =>
    })

    const resolved = yield* aisdk.model(model("@ai-sdk/anthropic"))
-    expect(resolved.route.providerMetadataKey).toBe("anthropic")
+    expect(resolved.route.id).toBe("ai-sdk:@ai-sdk/anthropic")
    const prepared = yield* compileRequest(
      LLM.request({
        model: resolved,
        messages: [
          Message.assistant([
-            { type: "text", text: "Answer", providerMetadata: { anthropic: { cacheControl: { type: "ephemeral" } } } },
+            {
+              type: "text",
+              text: "Answer",
+              providerMetadata: { [resolved.route.id]: { cacheControl: { type: "ephemeral" } } },
+            },
            { type: "text", text: " without metadata" },
-            { type: "reasoning", text: "Think", providerMetadata: { anthropic: { signature: "signed" } } },
+            { type: "reasoning", text: "Think", providerMetadata: { [resolved.route.id]: { signature: "signed" } } },
            {
              type: "tool-call",
              id: "hosted",
              name: "web_search",
              input: { query: "Effect" },
              providerExecuted: true,
-              providerMetadata: { anthropic: { blockType: "server_tool_use" } },
+              providerMetadata: { [resolved.route.id]: { blockType: "server_tool_use" } },
            },
          ]),
        ],
@@ -440,6 +444,41 @@ it.effect("projects replay metadata onto AI SDK prompt parts", () =>
  }),
 )

+it.effect("round-trips SDK metadata through the route namespace", () =>
+  Effect.gen(function* () {
+    const aisdk = yield* AISDK.Service
+    yield* aisdk.hook.sdk((event) => {
+      event.sdk = {
+        languageModel: () =>
+          streamModel([
+            { type: "reasoning-start", id: "thinking" },
+            { type: "reasoning-delta", id: "thinking", delta: "Think" },
+            { type: "reasoning-end", id: "thinking", providerMetadata: { anthropic: { signature: "signed" } } },
+            { type: "finish", finishReason: { unified: "stop", raw: "end_turn" }, usage },
+          ]),
+      }
+    })
+    const resolved = yield* aisdk.model(model("@ai-sdk/anthropic"))
+    const response = yield* LLMClient.generate(LLM.request({ model: resolved, prompt: "Think" })).pipe(
+      Effect.provide(client),
+    )
+    expect(response.message.content).toEqual([
+      {
+        type: "reasoning",
+        text: "Think",
+        providerMetadata: { "ai-sdk:@ai-sdk/anthropic": { signature: "signed" } },
+      },
+    ])
+    const replay = yield* compileRequest(LLM.request({ model: resolved, messages: [response.message] }))
+    expect(replay.body.prompt).toEqual([
+      {
+        role: "assistant",
+        content: [{ type: "reasoning", text: "Think", providerOptions: { anthropic: { signature: "signed" } } }],
+      },
+    ])
+  }),
+)
+
 it.effect("normalizes file data across AI SDK prompt parts", () =>
  Effect.gen(function* () {
    const aisdk = yield* AISDK.Service
@@ -253,7 +253,6 @@ describe("ModelResolver", () => {
      expect(resolved).toMatchObject({ id: "api-test-model", provider: "test-provider" })
      expect(resolved.route).toMatchObject({
        id: "openai-responses",
-        providerMetadataKey: "openai",
        endpoint: { baseURL: "https://openai.example/v1" },
        defaults: {
          headers: { "x-test": "header" },
@@ -469,7 +468,6 @@ describe("ModelResolver", () => {
      expect(headers.authorization).toBe("Bearer settings-secret")
      expect(resolved.route.id).toBe("openai-compatible-chat")
      expect(String(resolved.provider)).toBe("deepseek")
-      expect(resolved.route.providerMetadataKey).toBe("deepseek")
      expect(resolved.compatibility?.reasoningField).toBe("vendor_reasoning")
      expect(resolved.compatibility?.requireReasoning).toBe(true)
      expect(resolved.compatibility?.maxTokensField).toBe("max_completion_tokens")
@@ -640,7 +638,6 @@ describe("ModelResolver", () => {

      expect(resolved.route).toMatchObject({
        id: "anthropic-messages",
-        providerMetadataKey: "anthropic",
        endpoint: { baseURL: "https://anthropic.example/v1" },
      })
    }),
@@ -1262,7 +1259,7 @@ describe("ModelResolver", () => {
        reasoningEffort: "high",
      })
      expect(String(mistral.provider)).toBe("test-provider")
-      expect(xai.route.id).toBe("openai-responses")
+      expect(xai.route.id).toBe("xai-responses")
      expect(xai.route.defaults.providerOptions).toEqual({
        reasoningEffort: "high",
        store: false,
@@ -314,7 +314,7 @@ describe("GithubCopilotPlugin", () => {
        loadAISDK: (model) => aisdk.model(model),
      })
      expect(resolved.route.id).toBe("ai-sdk:@ai-sdk/github-copilot")
-      expect(resolved.route.providerMetadataKey).toBe("copilot")
+      expect(resolved.route.id).toBe("ai-sdk:@ai-sdk/github-copilot")
    }),
  )

@@ -317,7 +317,7 @@ describe("SessionInstructions", () => {
      metadata: { instruction: { paths: ["/repo/sub/AGENTS.md"] } },
      time: { created },
    })
-    const messages = toLLMMessages([synthetic], model)
+    const messages = toLLMMessages([synthetic], model, "test-route")
    expect(messages).toHaveLength(1)
    expect(messages[0]!.role).toBe("user")
    expect(messages[0]!.content).toEqual([{ type: "text", text: "Instructions from: /repo/sub/AGENTS.md\ncontent" }])
@@ -38,7 +38,7 @@ describe("toLLMMessages", () => {
      time: { created },
    })

-    expect(toLLMMessages([shell], model)).toEqual([])
+    expect(toLLMMessages([shell], model, "test-route")).toEqual([])
    const completed = SessionMessage.Shell.make({
      ...shell,
      status: "exited",
@@ -46,8 +46,8 @@ describe("toLLMMessages", () => {
      output: { output: "/project", cursor: 8, size: 8, truncated: false },
      time: { created, completed: created },
    })
-    expect(toLLMMessages([completed], model)).toEqual([])
-    expect(toLLMMessages([completed, notification], model)).toEqual([
+    expect(toLLMMessages([completed], model, "test-route")).toEqual([])
+    expect(toLLMMessages([completed, notification], model, "test-route")).toEqual([
      Message.make({ id: notification.id, role: "user", content: notification.text }),
    ])
  })
@@ -77,6 +77,7 @@ describe("toLLMMessages", () => {
        ]),
      ],
      model,
+      "test-route",
    )

    expect(messages.map((message) => message.id)).toEqual([id("text"), id("reasoning")])
@@ -153,6 +154,7 @@ describe("toLLMMessages", () => {
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages.map((message) => message.role)).toEqual(["user", "system", "user", "user", "user", "user"])
@@ -220,6 +222,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages).toHaveLength(1)
@@ -259,6 +262,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages).toEqual([
@@ -293,6 +297,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]).toMatchObject({
@@ -320,6 +325,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toMatchObject([
@@ -350,6 +356,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages).toHaveLength(1)
@@ -393,6 +400,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages).toHaveLength(1)
@@ -423,6 +431,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages).toHaveLength(1)
@@ -452,6 +461,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -482,6 +492,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -517,6 +528,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -555,6 +567,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -599,6 +612,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -655,6 +669,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content.filter((part) => part.type === "media")).toHaveLength(4)
@@ -744,12 +759,13 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages.map((message) => message.role)).toEqual(["assistant", "tool"])
    expect(messages[0]?.content).toEqual([
      { type: "text", text: "Checking" },
-      { type: "reasoning", text: "Think", providerMetadata: { provider: { signature: "sig_1" } } },
+      { type: "reasoning", text: "Think", providerMetadata: { "test-route": { signature: "sig_1" } } },
      { type: "tool-call", id: "pending", name: "read", input: { path: "README.md" } },
      { type: "tool-call", id: "running", name: "read", input: { path: "README.md" } },
      {
@@ -764,14 +780,14 @@ Recent work
        name: "web_search",
        input: { query: "Effect" },
        providerExecuted: true,
-        providerMetadata: { provider: { continuation: "hosted-call" } },
+        providerMetadata: { "test-route": { continuation: "hosted-call" } },
      },
      {
        type: "tool-result",
        id: "hosted",
        name: "web_search",
        providerExecuted: true,
-        providerMetadata: { provider: { continuation: "hosted-result" } },
+        providerMetadata: { "test-route": { continuation: "hosted-result" } },
        result: { type: "text", value: "Found it" },
      },
      {
@@ -780,14 +796,14 @@ Recent work
        name: "write",
        input: { path: "README.md" },
        providerExecuted: true,
-        providerMetadata: { provider: { continuation: "failed" } },
+        providerMetadata: { "test-route": { continuation: "failed" } },
      },
      {
        type: "tool-result",
        id: "hosted-failed",
        name: "write",
        providerExecuted: true,
-        providerMetadata: { provider: { continuation: "failed" } },
+        providerMetadata: { "test-route": { continuation: "failed" } },
        result: {
          type: "error",
          value: { error: { type: "unknown", message: "Denied" }, content: [] },
@@ -829,13 +845,14 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
      {
        type: "reasoning",
        text: "Think",
-        providerMetadata: { provider: { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
+        providerMetadata: { "test-route": { itemId: "rs_1", reasoningEncryptedContent: "encrypted-state" } },
      },
    ])
  })
@@ -860,11 +877,11 @@ Recent work
        }),
      ],
      opencode,
-      "anthropic",
+      "anthropic-messages",
    )

    expect(messages[0]?.content).toEqual([
-      { type: "reasoning", text: "Think", providerMetadata: { anthropic: { signature: "signed" } } },
+      { type: "reasoning", text: "Think", providerMetadata: { "anthropic-messages": { signature: "signed" } } },
    ])
  })

@@ -917,6 +934,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -927,7 +945,7 @@ Recent work
        name: "web_search",
        input: { query: "Effect" },
        providerExecuted: true,
-        providerMetadata: { provider: { itemId: "call_completed" } },
+        providerMetadata: { "test-route": { itemId: "call_completed" } },
      },
      {
        type: "tool-result",
@@ -937,7 +955,7 @@ Recent work
        providerExecuted: true,
        cache: undefined,
        metadata: undefined,
-        providerMetadata: { provider: { itemId: "result_completed" } },
+        providerMetadata: { "test-route": { itemId: "result_completed" } },
      },
      {
        type: "tool-call",
@@ -945,7 +963,7 @@ Recent work
        name: "web_search",
        input: { query: "Effect" },
        providerExecuted: true,
-        providerMetadata: { provider: { itemId: "call_failed" } },
+        providerMetadata: { "test-route": { itemId: "call_failed" } },
      },
      {
        type: "tool-result",
@@ -961,7 +979,7 @@ Recent work
        providerExecuted: true,
        cache: undefined,
        metadata: undefined,
-        providerMetadata: { provider: { itemId: "result_failed" } },
+        providerMetadata: { "test-route": { itemId: "result_failed" } },
      },
    ])
  })
@@ -1013,6 +1031,7 @@ Recent work
        }),
      ],
      model,
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -1035,7 +1054,7 @@ Recent work
        metadata: undefined,
        // Hosted result payloads are provider-format state and must survive a
        // model switch within the same provider for replay to stay valid.
-        providerMetadata: { provider: { itemId: "hosted-old-model" } },
+        providerMetadata: { "test-route": { itemId: "hosted-old-model" } },
      },
      {
        type: "tool-call",
@@ -1079,13 +1098,14 @@ Recent work
        }),
      ],
      Model.Ref.make({ id: Model.ID.make("fast"), providerID: Provider.ID.make("provider") }),
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
      {
        type: "reasoning",
        text: "Visible thought",
-        providerMetadata: { provider: { reasoningEncryptedContent: "encrypted" } },
+        providerMetadata: { "test-route": { reasoningEncryptedContent: "encrypted" } },
      },
    ])
  })
@@ -1110,6 +1130,7 @@ Recent work
        }),
      ],
      Model.Ref.make({ id: Model.ID.make("new"), providerID: Provider.ID.make("provider") }),
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
@@ -1140,13 +1161,14 @@ Recent work
        }),
      ],
      Model.Ref.make({ id: Model.ID.make("same"), providerID: Provider.ID.make("provider") }),
+      "test-route",
    )

    expect(messages[0]?.content).toEqual([
      {
        type: "text",
        text: "Checking.",
-        providerMetadata: { provider: { phase: "commentary" } },
+        providerMetadata: { "test-route": { phase: "commentary" } },
      },
    ])
  })
@@ -29,7 +29,7 @@ const sessionID = Session.ID.make("ses_tool_event_test")
 const base64 = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAAB"

 const capture = (
-  providerMetadataKey = "anthropic",
+  routeID = "anthropic-messages",
  options?: { readonly interruptProgress?: boolean; readonly beforeTextDelta?: Effect.Effect<void> },
 ) => {
  const published: Array<{ readonly type: string; readonly data: unknown }> = []
@@ -59,7 +59,7 @@ const capture = (
        id: Model.ID.make("model"),
        providerID: Provider.ID.opencode,
      },
-      providerMetadataKey,
+      routeID,
      assistantMessageID: SessionMessage.ID.create(),
    }),
  }
@@ -114,7 +114,7 @@ test("provider-executed success derives content and retains provider result stat
      LLMEvent.toolResult({
        ...hostedResult,
        providerExecuted: true,
-        providerMetadata: { anthropic: { result: { type: "content", value: [] } } },
+        providerMetadata: { "anthropic-messages": { result: { type: "content", value: [] } } },
      }),
    ),
  )
@@ -162,7 +162,7 @@ testEffect(
        assistantMessageID,
        agent: Agent.defaultID,
        model: { id: Model.ID.make("test-model"), providerID: Provider.ID.opencode },
-        providerMetadataKey: "openai",
+        routeID: "openai-responses",
      },
    )
    yield* publisher.publish(LLMEvent.toolCall({ ...call, providerExecuted: true }))
@@ -213,7 +213,7 @@ testEffect(
 )

 test("interrupted progress metadata remains in the terminal failure snapshot", async () => {
-  const { published, publisher } = capture("anthropic", { interruptProgress: true })
+  const { published, publisher } = capture("anthropic-messages", { interruptProgress: true })
  await Effect.runPromise(publisher.publish(call))
  const exit = await Effect.runPromiseExit(publisher.progress(call.id, { phase: "visible" }))
  expect(Exit.isFailure(exit) && Cause.hasInterruptsOnly(exit.cause)).toBe(true)
@@ -225,7 +225,7 @@ test("interrupted progress metadata remains in the terminal failure snapshot", a
 })

 test("interrupted subagent failures expose their existing child session to the model", async () => {
-  const { published, publisher } = capture("anthropic", { interruptProgress: true })
+  const { published, publisher } = capture("anthropic-messages", { interruptProgress: true })
  const subagent = LLMEvent.toolCall({
    id: "call-subagent",
    name: "subagent",
@@ -267,7 +267,7 @@ test("local failure metadata completes the progress snapshot", async () => {
 })

 test("failure snapshot retains canonical progress above the default byte limit", async () => {
-  const { published, publisher } = capture("anthropic", { interruptProgress: true })
+  const { published, publisher } = capture("anthropic-messages", { interruptProgress: true })
  await Effect.runPromise(publisher.publish(call))
  const detail = "x".repeat(60 * 1024)
  await Effect.runPromiseExit(publisher.progress(call.id, { detail }))
@@ -292,7 +292,7 @@ test("provider metadata is flattened using the route key", async () => {
  const { published, publisher } = capture()
  await Effect.runPromise(
    publisher.publish(
-      LLMEvent.reasoningStart({ id: "reasoning", providerMetadata: { anthropic: { signature: "signed" } } }),
+      LLMEvent.reasoningStart({ id: "reasoning", providerMetadata: { "anthropic-messages": { signature: "signed" } } }),
    ),
  )

@@ -305,7 +305,10 @@ test("reasoning state from start, empty delta, and end is merged", async () => {
  const { published, publisher } = capture()
  await Effect.runPromise(
    publisher.publish(
-      LLMEvent.reasoningStart({ id: "reasoning", providerMetadata: { anthropic: { blockType: "thinking" } } }),
+      LLMEvent.reasoningStart({
+        id: "reasoning",
+        providerMetadata: { "anthropic-messages": { blockType: "thinking" } },
+      }),
    ),
  )
  await Effect.runPromise(
@@ -313,13 +316,16 @@ test("reasoning state from start, empty delta, and end is merged", async () => {
      LLMEvent.reasoningDelta({
        id: "reasoning",
        text: "",
-        providerMetadata: { anthropic: { signature: "signed" }, gateway: { traceID: "trace" } },
+        providerMetadata: { "anthropic-messages": { signature: "signed" }, gateway: { traceID: "trace" } },
      }),
    ),
  )
  await Effect.runPromise(
    publisher.publish(
-      LLMEvent.reasoningEnd({ id: "reasoning", providerMetadata: { anthropic: { stopReason: "tool_use" } } }),
+      LLMEvent.reasoningEnd({
+        id: "reasoning",
+        providerMetadata: { "anthropic-messages": { stopReason: "tool_use" } },
+      }),
    ),
  )

@@ -389,7 +395,7 @@ it.effect("retains new chunks and orders text-end behind an in-flight timer publ
  Effect.gen(function* () {
    const entered = yield* Deferred.make<void>()
    const release = yield* Deferred.make<void>()
-    const { published, publisher } = capture("anthropic", {
+    const { published, publisher } = capture("anthropic-messages", {
      beforeTextDelta: Deferred.succeed(entered, undefined).pipe(Effect.andThen(Deferred.await(release))),
    })
    yield* publisher.publish(LLMEvent.textStart({ id: "text" }))
@@ -481,7 +487,7 @@ test("tool input deltas are accumulated without being published", async () => {
 })

 test("provider-executed tool metadata is flattened using the route key", async () => {
-  const { published, publisher } = capture("openai")
+  const { published, publisher } = capture("openai-responses")
  await Effect.runPromise(
    publisher.publish(
      LLMEvent.toolCall({
@@ -489,7 +495,7 @@ test("provider-executed tool metadata is flattened using the route key", async (
        name: "web_search",
        input: { query: "Effect" },
        providerExecuted: true,
-        providerMetadata: { openai: { itemId: "call" } },
+        providerMetadata: { "openai-responses": { itemId: "call" } },
      }),
    ),
  )
@@ -500,7 +506,7 @@ test("provider-executed tool metadata is flattened using the route key", async (
        name: "web_search",
        result: { type: "json", value: { found: true } },
        providerExecuted: true,
-        providerMetadata: { openai: { itemId: "result" } },
+        providerMetadata: { "openai-responses": { itemId: "result" } },
      }),
    ),
  )
@@ -556,7 +562,7 @@ test("content-filter finish retains failure evidence until step closeout", async
        index: 0,
        reason: { normalized: "content-filter", raw: "refusal" },
        providerMetadata: {
-          anthropic: {
+          "anthropic-messages": {
            stopDetails: { type: "refusal", category: "safety", explanation: "Blocked" },
          },
        },