perf(desktop): remember the bundled CLI version between launches

Every launch spawned the bundled 200 MB CLI executable for --version and
waited ~380 ms for it before the background service, WSL support and,
because windows are created after every layer builds, the window itself.
The version only changes when an update replaces the executable, so keep
it in the settings store keyed by the executable's path, size and mtime
and only spawn when that identity changes.

Packaged build, warm service, 5 launches (median ms since spawn):
renderer process 851 -> 512, window visible 1085 -> 778,
shell visible 1289 -> 961.

AGENTS.md

@@ -184,6 +184,7 @@ const table = sqliteTable("session", {
 - Keep `SessionRunner`, model resolution, tool registry, permissions, and filesystem Location-scoped. Omitted `Location.workspaceID` means implicit-local placement; explicit workspace identity remains reserved for future placement semantics.
 - Preserve one explicit `llm.stream(request)` call per Physical Attempt and reload projected history before durable continuation. A logical Step may use generic pre-output retries, one full-context retry after continuation rejection, incomplete-stream continuation, or one overflow-compaction rebuild. Generic retries retain the logical step number and do not consume another agent-step allowance. Do not delegate orchestration to an in-memory tool loop.
 - Keep local Session drains process-local until clustering is implemented. `SessionRunCoordinator` joins explicit same-Session resumes, coalesces prompt wakeups, and allows different Sessions to run concurrently. A write-ahead execution claim marks a process-local busy period for restart recovery: terminal completion, failure, or user interruption releases it, while shutdown interruption and process death preserve it. Startup recovery resumes claimed top-level Sessions with durable per-execution attempt accounting. The claim is a recovery marker, not clustered ownership, fencing, or an exactly-once guarantee.
+- Keep native compaction mechanisms out of `SessionCompaction`. Plugins register `native` strategies through the `SessionCompaction` editor that turn a prepared request into a replacement window (the built-in `NativeCompactionPlugin` handles `@opencode/ai` compaction operations); later registrations win. Core owns the provider-mode decision, route provenance, the retry policy, overflow recovery, interruption, usage accounting, and checkpoint persistence.
 - Keep delivery vocabulary explicit. Prompts steer by default. At safe step boundaries, steered compaction takes priority up to the first steered move control; other steers retain enqueue order. At an idle boundary, steers take priority; otherwise exactly one queued item delivers before the runner reevaluates continuation. Inbox items may be cancelled or changed between queue and steer before delivery. Promoting new user input resets the selected agent's step allowance; a batch of steers resets it once.
 - One step is one logical LLM call; its durable record covers only the model-visible span. Do not write "provider turn", and do not use bare "turn" for a single call: "turn" is reserved for the future assistant-turn unit containing all steps from prompt promotion until the session would go idle.
 - Keep event replay ownership separate from clustered Session execution ownership.

packages/core/src/plugin/compaction.ts

@@ -0,0 +1,29 @@
+export * as NativeCompactionPlugin from "./compaction.js"
+
+import { LLMClient, Message } from "@opencode/ai"
+import { define } from "@opencode/plugin/effect/plugin"
+import { Effect } from "effect"
+import { SessionCompaction } from "../session/compaction.js"
+import type { PluginInternal } from "./internal.js"
+
+export const Plugin = define({
+  id: "opencode.compaction.native",
+  effect: Effect.fn("NativeCompactionPlugin")(function* () {
+    const llm = yield* LLMClient.Service
+    const compaction = yield* SessionCompaction.Service
+    yield* compaction.transform((editor) => {
+      editor.native((input) => {
+        const request = input.request
+        if (LLMClient.canCompact(request, { mechanism: "trigger" }))
+          return Effect.gen(function* () {
+            const retained = yield* input.retained
+            const result = yield* llm.compact(request, { ...input.options, mechanism: "trigger" })
+            return { replacement: [...retained, Message.assistant(result.checkpoint)], usage: result.usage }
+          })
+        if (LLMClient.canCompact(request))
+          return llm.compact(request, { mechanism: "endpoint", http: input.options.http })
+        return undefined
+      })
+    })
+  }),
+} satisfies PluginInternal.InternalPlugin)

packages/core/src/plugin/internal.ts

@@ -1,5 +1,6 @@
 export * as PluginInternal from "./internal.js"
 
+import { LLMClient } from "@opencode/ai"
 import type { Plugin } from "@opencode/plugin/effect/plugin"
 import { LayerNode } from "@opencode/util/effect/layer-node"
 import { httpClient } from "@opencode/util/effect/app-node-platform"
@@ -12,6 +13,7 @@ import { Provider } from "../provider.js"
 import { Command } from "../command.js"
 import { Config } from "../config.js"
 import { Credential } from "../credential.js"
+import { llmClient } from "../effect/app-node-platform.js"
 import { ConfigAgentPlugin } from "../config/plugin/agent.js"
 import { ConfigCommandPlugin } from "../config/plugin/command.js"
 import { ConfigCompactionPlugin } from "../config/plugin/compaction.js"
@@ -84,6 +86,7 @@ import { WriteTool } from "../tool/plugin/write.js"
 import { AgentPlugin } from "./agent.js"
 import BrowserPlugin from "@opencode/plugin-browser"
 import { CommandPlugin } from "./command.js"
+import { NativeCompactionPlugin } from "./compaction.js"
 import { IdentityPlugin } from "./identity.js"
 import { PlanPlugin } from "./plan.js"
 import { ModelsDevPlugin } from "./models-dev.js"
@@ -120,6 +123,7 @@ const services = [
   Integration.Service,
   Job.Service,
   KV.Service,
+  LLMClient.Service,
   Location.Service,
   ModelsDev.Service,
   Mcp.Service,
@@ -171,6 +175,7 @@ export const requirements = LayerNode.group([
   Integration.node,
   Job.node,
   KV.node,
+  llmClient,
   Location.node,
   ModelsDev.node,
   Mcp.node,
@@ -212,6 +217,7 @@ const pre = [
   SkillPlugin.Plugin,
   VcsHgPlugin.Plugin,
   ModelsDevPlugin,
+  NativeCompactionPlugin.Plugin,
   ...ProviderPlugins,
   ...WebSearchPlugins,
   PatchTool.Plugin,

packages/core/src/session/compaction.ts

@@ -10,7 +10,9 @@ import {
   LLMRequest,
   Message,
   type ContentPart,
+  type Usage,
 } from "@opencode/ai"
+import type { StreamOptions } from "@opencode/ai/route"
 import type { SessionCompactionResult } from "@opencode/plugin/effect/session"
 import { SessionError } from "@opencode/schema/session-error"
 import { Context, Effect, Layer, Stream } from "effect"
@@ -91,8 +93,25 @@ export type Settings = {
   tokens: number
 }
 
+export type NativeInput = {
+  readonly request: LLMRequest
+  readonly options: StreamOptions
+  /** Whole, real user messages within the retained-token allowance, for checkpoint-only mechanisms. */
+  readonly retained: Effect.Effect<ReadonlyArray<Message>>
+}
+
+export type NativeResult = {
+  readonly replacement: ReadonlyArray<Message>
+  readonly usage?: Usage
+}
+
+/** Returns the provider's replacement window, or `undefined` when this strategy has no mechanism for the route. */
+export type NativeStrategy = (input: NativeInput) => Effect.Effect<NativeResult, AIError> | undefined
+
 export type Editor = {
   configure: (settings: Partial<Settings>) => void
+  /** Later registrations take precedence. */
+  native: (strategy: NativeStrategy) => void
 }
 
 export type AutoInput = {
@@ -380,15 +399,18 @@ export const layer = Layer.effect(
     const llm = yield* LLMClient.Service
     const db = (yield* Database.Service).db
 
-    const state = State.create<Settings, Editor>({
+    const state = State.create<Settings & { readonly native: NativeStrategy[] }, Editor>({
       name: "session-compaction",
-      initial: () => ({ auto: true, buffer: DEFAULT_BUFFER, tokens: DEFAULT_KEEP_TOKENS }),
+      initial: () => ({ auto: true, buffer: DEFAULT_BUFFER, tokens: DEFAULT_KEEP_TOKENS, native: [] }),
       editor: (editor) => ({
         configure: (settings) => {
           if (settings.auto !== undefined) editor.auto = settings.auto
           if (settings.buffer !== undefined) editor.buffer = settings.buffer
           if (settings.tokens !== undefined) editor.tokens = settings.tokens
         },
+        native: (strategy) => {
+          editor.native.push(strategy)
+        },
       }),
     })
     const failed = Effect.fnUntraced(function* (input: SessionEvent.Compaction.Failed["data"]) {
@@ -504,6 +526,23 @@ export const layer = Layer.effect(
         return yield* reject(
           "Provider compaction requires the endpoint in provider/model settings, not a model.request rewrite",
         )
+      const native = state
+        .get()
+        .native.toReversed()
+        .map((strategy) =>
+          strategy({
+            request,
+            options: prepared.options,
+            retained: original(context.session.id).pipe(
+              Effect.map((messages) => retainUsers(messages, context.model, state.get().tokens)),
+            ),
+          }),
+        )
+        .find((effect) => effect !== undefined)
+      if (!native)
+        return yield* reject(
+          `No plugin provides native compaction for ${request.model.provider}/${request.model.route.id}`,
+        )
       const transient = SessionRunnerRetry.transient(yield* SessionRunnerRetry.policy(context.session.id), {
         agent: context.agent.id,
         model: context.model.ref,
@@ -514,25 +553,7 @@ export const layer = Layer.effect(
         Effect.gen(function* () {
           // Transient provider failures retry like any other request; only a known automatic overflow permits
           // local recovery, and nothing is installed until the provider returns a checkpoint.
-          const result = yield* restore(
-            Effect.gen(function* () {
-              if (LLMClient.canCompact(request, { mechanism: "trigger" })) {
-                const retained = retainUsers(yield* original(context.session.id), context.model, state.get().tokens)
-                const result = yield* llm
-                  .compact(request, { ...prepared.options, mechanism: "trigger" })
-                  .pipe(transient)
-                return { replacement: [...retained, Message.assistant(result.checkpoint)], usage: result.usage }
-              }
-              if (LLMClient.canCompact(request))
-                return yield* llm
-                  .compact(request, { mechanism: "endpoint", http: prepared.options.http })
-                  .pipe(transient)
-              // Model resolution admits provider policies only for routes with a compaction operation.
-              return yield* Effect.die(
-                new Error(`${request.model.provider}/${request.model.route.id} has no compaction operation`),
-              )
-            }),
-          )
+          const result = yield* restore(native.pipe(transient))
           const usage = result.usage ? SessionUsage.record(result.usage, context.model.cost) : undefined
           if (usage)
             yield* bus.publish(SessionEvent.UsageRecorded, {

packages/core/test/session-native-compaction.test.ts

@@ -1,5 +1,5 @@
 import { expect, test } from "bun:test"
-import { LLMClient, LanguageModel, Message, ToolDefinition } from "@opencode/ai"
+import { LLMClient, LanguageModel, Message, ToolDefinition, Usage } from "@opencode/ai"
 import { OpenAI } from "@opencode/ai/providers"
 import { Agent } from "@opencode/core/agent"
 import { Bus } from "@opencode/core/bus"
@@ -7,6 +7,7 @@ import { Database } from "@opencode/core/database/database"
 import { AppNodeBuilder } from "@opencode/core/effect/app-node-builder"
 import { llmClient } from "@opencode/core/effect/app-node-platform"
 import { Instructions } from "@opencode/core/instructions/index"
+import { NativeCompactionPlugin } from "@opencode/core/plugin/compaction"
 import { PluginHooks } from "@opencode/core/plugin/hooks"
 import { Project } from "@opencode/core/project"
 import { ProjectTable } from "@opencode/core/project/sql"
@@ -26,6 +27,7 @@ import { SessionStore } from "@opencode/core/session/store"
 import { LayerNode } from "@opencode/util/effect/layer-node"
 import { DateTime, Deferred, Effect, Fiber, Schema } from "effect"
 import { testEffect } from "./lib/effect"
+import { host } from "./plugin/host"
 
 const it = testEffect(
   AppNodeBuilder.build(
@@ -44,7 +46,8 @@ const it = testEffect(
   ),
 )
 
-const setup = Effect.fnUntraced(function* (endpoint = false) {
+const setup = Effect.fnUntraced(function* (options: { endpoint?: boolean; plugin?: boolean } = {}) {
+  const endpoint = options.endpoint ?? false
   const db = (yield* Database.Service).db
   const bus = yield* Bus.Service
   const inbox = yield* SessionInbox.Service
@@ -185,6 +188,7 @@ const setup = Effect.fnUntraced(function* (endpoint = false) {
     render: { initial: String, changed: (_previous, value) => value, removed: () => "removed" },
   })
   yield* InstructionState.prepare(db, bus, instructions, sessionID)
+  if (options.plugin !== false) yield* NativeCompactionPlugin.Plugin.effect(host())
   yield* hooks.register("session", "model.request", (event) =>
     Effect.sync(() => {
       event.headers["x-test-hook"] = event.kind
@@ -261,6 +265,7 @@ const setup = Effect.fnUntraced(function* (endpoint = false) {
     store,
     hooks,
     model,
+    compaction,
   }
 })
 
@@ -348,7 +353,7 @@ it.live(
 
 it.live("manual and automatic endpoint compaction keep the provider replacement unchanged", () =>
   Effect.gen(function* () {
-    const fixture = yield* setup(true)
+    const fixture = yield* setup({ endpoint: true })
     yield* fixture.prompt("Original user")
     expect(yield* fixture.compact).toEqual({ status: "completed" })
     expect(yield* fixture.automatic).toEqual({ status: "completed" })
@@ -445,6 +450,31 @@ it.live("rejects request-hook route rewrites before provider compaction", () =>
   }),
 )
 
+it.live("provider compaction fails without a native strategy and persists a registered strategy's window", () =>
+  Effect.gen(function* () {
+    const fixture = yield* setup({ plugin: false })
+    yield* fixture.prompt("Original user")
+    expect(yield* fixture.compact).toMatchObject({
+      status: "failed",
+      error: { type: "provider.unsupported-operation", message: expect.stringContaining("openai/openai-responses") },
+    })
+    yield* fixture.compaction.transform((editor) => {
+      editor.native(() =>
+        Effect.succeed({
+          replacement: [Message.assistant("plugin window")],
+          usage: new Usage({ nonCachedInputTokens: 20, outputTokens: 4 }),
+        }),
+      )
+    })
+    expect(yield* fixture.compact).toEqual({ status: "completed" })
+    expect(fixture.state.calls).toBe(0)
+    const installed = yield* fixture.checkpoint
+    expect(installed.provenance).toEqual(SessionProviderContext.provenance(fixture.model)!)
+    expect(SessionProviderContext.decode(installed)).toEqual([Message.assistant("plugin window")])
+    expect(yield* fixture.store.get(fixture.sessionID)).toMatchObject({ tokens: { input: 20, output: 4 } })
+  }),
+)
+
 test("retained user budget counts attachments and drops whole oldest messages", () => {
   const model = SessionRunnerModel.resolved(OpenAI.responses("gpt-5.4-mini"), {
     capabilities: { tools: true, input: ["text", "image"], output: ["text"] },

packages/core/test/session-runner.test.ts

@@ -56,6 +56,7 @@ import { Plugin } from "@opencode/core/plugin"
 import { PluginHooks } from "@opencode/core/plugin/hooks"
 import { OptimizePlugin } from "@opencode/core/plugin/optimize"
 import { IdentityPlugin } from "@opencode/core/plugin/identity"
+import { NativeCompactionPlugin } from "@opencode/core/plugin/compaction"
 import { QuestionTool } from "@opencode/core/tool/plugin/question"
 import { Agent } from "@opencode/core/agent"
 import { Config } from "@opencode/core/config"
@@ -470,6 +471,7 @@ const layer = Layer.unwrap(
         Config.node,
         Snapshot.node,
         SessionCompaction.node,
+        LayerNodePlatform.llmClient,
         SessionRunnerLLM.node,
         SessionExecution.node,
         Session.node,
@@ -523,6 +525,7 @@ const setup = Effect.gen(function* () {
     discard: true,
   })
   yield* IdentityPlugin.Plugin.effect(pluginHost)
+  yield* NativeCompactionPlugin.Plugin.effect(pluginHost)
   yield* agents.transform((editor) =>
     editor.update(Agent.ID.make("build"), (agent) => {
       agent.mode = "primary"

packages/desktop/src/main/service/desktop-cli.ts

@@ -3,9 +3,11 @@ export * as DesktopCli from "./desktop-cli"
 import { execFile, spawn } from "node:child_process"
 import { promisify } from "node:util"
 import { app } from "electron"
-import { Context, Effect, FileSystem, Layer, Path } from "effect"
+import { Context, Effect, FileSystem, Layer, Option, Path } from "effect"
 import installer from "../../../../../install?raw"
 import { DesktopPaths } from "../paths"
+import { BUNDLED_CLI_VERSION_KEY } from "../storage/keys"
+import { getStore } from "../storage/store"
 import { parseCliVersion } from "./cli-version"
 
 const execFileAsync = promisify(execFile)
@@ -79,11 +81,36 @@ const resolveBundledCli = Effect.fn("DesktopCli.resolveBundled")(function* (isol
     ? path.join(process.resourcesPath, executableName())
     : path.join(paths.developmentResourcesRoot, isolated ? developmentExecutableName() : executableName())
   yield* Effect.logInfo("v2 CLI executable resolved", { bundled, packaged: app.isPackaged })
-  const version = parseCliVersion(yield* run(bundled, ["--version"]))
+  const version = yield* bundledVersion(bundled)
   const binary = app.isPackaged || isolated ? yield* installCli(bundled, version) : bundled
   return { version, binary, command: [binary] }
 })
 
+// Spawning the bundled executable for `--version` costs ~400 ms of startup on a 200 MB binary, so
+// the answer is remembered per executable identity and only re-read after an update replaces it.
+const bundledVersion = Effect.fn("DesktopCli.bundledVersion")(function* (bundled: string) {
+  const fs = yield* FileSystem.FileSystem
+  const stat = yield* fs.stat(bundled).pipe(Effect.orElseSucceed(() => undefined))
+  const identity = stat ? `${stat.size}:${Option.getOrUndefined(stat.mtime)?.getTime() ?? ""}` : undefined
+  const store = getStore()
+  const cached = store.get(BUNDLED_CLI_VERSION_KEY)
+  if (identity && isVersionCache(cached) && cached.path === bundled && cached.identity === identity) {
+    yield* Effect.logInfo("v2 CLI version reused", { version: cached.version })
+    return cached.version
+  }
+  const version = parseCliVersion(yield* run(bundled, ["--version"]))
+  if (identity) store.set(BUNDLED_CLI_VERSION_KEY, { path: bundled, identity, version } satisfies VersionCache)
+  return version
+})
+
+type VersionCache = { path: string; identity: string; version: string }
+
+function isVersionCache(value: unknown): value is VersionCache {
+  if (!value || typeof value !== "object") return false
+  const cache = value as Record<string, unknown>
+  return typeof cache.path === "string" && typeof cache.identity === "string" && typeof cache.version === "string"
+}
+
 export const cleanStages = Effect.fn("DesktopCli.cleanStages")(function* (binary: string) {
   const fs = yield* FileSystem.FileSystem
   const path = yield* Path.Path

packages/desktop/src/main/storage/keys.ts

@@ -5,3 +5,4 @@ export const WSL_SERVERS_KEY = "wslServers"
 export const PINCH_ZOOM_ENABLED_KEY = "pinchZoomEnabled"
 export const BACKGROUND_COLOR_KEY = "backgroundColor"
 export const WINDOW_IDS_KEY = "windowIds"
+export const BUNDLED_CLI_VERSION_KEY = "bundledCliVersion"