@caupulican/pi-adaptative 0.81.13 → 0.81.16

package/dist/core/local-runtime-controller.d.ts

@@ -12,7 +12,7 @@ import type { Api, AssistantMessage, Model } from "@caupulican/pi-ai";
 import type { AgentSessionEvent } from "./agent-session.ts";
 import type { RouteDecision } from "./autonomy/contracts.ts";
 import type { ExtensionUIContext } from "./extensions/index.ts";
-import { type LocalRuntimeDeps, OllamaRuntime } from "./models/local-runtime.ts";
+import { type LocalRuntimeDeps, OllamaRuntime, TransformersRuntime } from "./models/local-runtime.ts";
 export interface LocalRuntimeControllerDeps {
     /** Root directory OllamaRuntime instances are scoped under — fixed for the session's lifetime. */
     agentDir: string;
@@ -34,6 +34,8 @@ export interface LocalRuntimeControllerDeps {
 export declare class LocalRuntimeController {
     /** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */
     private readonly _runtimes;
+    /** Lazy, cached by model+baseUrl so the router and `/models` share one sidecar handle per HF model. */
+    private readonly _transformersRuntimes;
     /** Server URLs confirmed reachable THIS session — skips the health-check round trip on every
      * local-routed turn once warm. Keyed the same way as _runtimes. */
     private readonly _confirmedUp;
@@ -46,31 +48,37 @@ export declare class LocalRuntimeController {
      * own untracked child.
      */
     getLocalRuntime(baseUrl?: string): OllamaRuntime;
+    getTransformersRuntime(modelId: string, baseUrl?: string): TransformersRuntime;
     /** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's
      * own health/boot endpoints are on the Ollama-native server root. */
     deriveOllamaServerUrl(modelBaseUrl: string): string;
+    private deriveOpenAICompatServerUrl;
+    private isManagedLocalProvider;
     /**
      * If the last assistant message in this session was an error from THIS exact local server, a
      * cached "confirmed up" flag would be stale (the server may have died mid-session) — drop it so
      * the next ensure-check is a real one instead of trusting stale state.
      */
+    private confirmationKey;
     private invalidateIfLastCallFailed;
     /**
-     * Ensure a routed model is actually reachable before the turn calls it. No-op (and free) for any
-     * non-local model — this only ever does network/process work for the `ollama` provider. Caches a
-     * "confirmed up this session" flag per server so a steady-state session pays the health-check
-     * round trip once, not on every turn; invalidated above when a prior local call actually failed,
-     * so a server that died mid-session gets re-detected rather than trusted forever. Boots via
-     * `startReuseExisting()` — never owned storage — so the turn sees the user's OWN pulled models,
-     * the same server `/models` commands and the user's own `ollama` CLI already talk to. Never
-     * installs anything itself (installGuide is GUIDE MODE: printed, never executed).
+     * Ensure a routed managed-local model is actually reachable before the turn calls it. No-op (and
+     * free) for non-local/API models. Caches a "confirmed up this session" flag per server (and per
+     * Transformers model) so steady-state routing pays the health-check round trip once; invalidated
+     * above when a prior local call failed so a dead sidecar gets re-detected instead of trusted.
      */
     ensureLocalModelReady(model: Model<Api>): Promise<{
         ready: boolean;
         reason: string;
         installGuide?: string[];
     }>;
+    ensureTransformersModelReady(model: Model<Api>): Promise<{
+        ready: boolean;
+        reason: string;
+        installGuide?: string[];
+    }>;
     private maybeInstallOllamaOnConsent;
+    private maybeInstallTransformersOnConsent;
     /**
      * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct
      * route — both carry tier "cheap") must not dead-end the turn just because ollama isn't up.

package/dist/core/local-runtime-controller.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"local-runtime-controller.d.ts","sourceRoot":"","sources":["../../src/core/local-runtime-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,GAAG,EAAE,gBAAgB,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AACtE,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAC5D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,yBAAyB,CAAC;AAC7D,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,uBAAuB,CAAC;AAEhE,OAAO,EAAE,KAAK,gBAAgB,EAAE,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAUjF,MAAM,WAAW,0BAA0B;IAC1C,oGAAkG;IAClG,QAAQ,EAAE,MAAM,CAAC;IACjB,mGAAmG;IACnG,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC;8BAC0B;IAC1B,uBAAuB,IAAI,gBAAgB,GAAG,SAAS,CAAC;IACxD,mGAAiG;IACjG,YAAY,IAAI,kBAAkB,GAAG,SAAS,CAAC;IAC/C,0GAA0G;IAC1G,IAAI,CAAC,KAAK,EAAE,iBAAiB,GAAG,IAAI,CAAC;IACrC;6CACyC;IACzC,0BAA0B,CAAC,IAAI,EAAE,QAAQ,GAAG,WAAW,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IACjF,uEAAuE;IACvE,WAAW,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;CACvC;AAED,qBAAa,sBAAsB;IAClC,qGAAqG;IACrG,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAoC;IAC9D;uEACmE;IACnE,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAqB;IAElD,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA6B;IAElD,YAAY,IAAI,EAAE,0BAA0B,EAE3C;IAED;;;;;OAKG;IACH,eAAe,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,aAAa,CAQ/C;IAED;yEACqE;IACrE,qBAAqB,CAAC,YAAY,EAAE,MAAM,GAAG,MAAM,CAElD;IAED;;;;OAIG;IACH,OAAO,CAAC,0BAA0B;IAWlC;;;;;;;;;OASG;IACG,qBAAqB,CAC1B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GACf,OAAO,CAAC;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC,CAuBtE;YAgBa,2BAA2B;IAsCzC;;;;;;;;;;;;;;;;OAgBG;IACG,qBAAqB,CAC1B,QAAQ,EAAE;QAAE,QAAQ,EAAE,aAAa,CAAC;QAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAA;KAAE,GAAG,SAAS,GAClE,OAAO,CAAC;QAAE,QAAQ,EAAE,aAAa,CAAC;QAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAA;KAAE,GAAG,SAAS,CAAC,CAuDrE;CACD","sourcesContent":["/**\n * Local-runtime (Ollama) lifecycle controller.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Owns the cached, per-server\n * {@link OllamaRuntime} instances, the \"confirmed up this session\" flag, and the router's readiness\n * gate for a turn routed to a local (`ollama`) model — including the #31 install-on-consent flow and\n * the #27 graceful tier-escalation fallback. Takes narrow deps (agent dir, a last-assistant-message\n * accessor, the session's UI context/event emitter, and the router's own tier resolver) rather than\n * the whole AgentSession.\n */\n\nimport type { Api, AssistantMessage, Model } from \"@caupulican/pi-ai\";\nimport type { AgentSessionEvent } from \"./agent-session.ts\";\nimport type { RouteDecision } from \"./autonomy/contracts.ts\";\nimport type { ExtensionUIContext } from \"./extensions/index.ts\";\nimport { OLLAMA_PROVIDER } from \"./models/local-registration.ts\";\nimport { type LocalRuntimeDeps, OllamaRuntime } from \"./models/local-runtime.ts\";\n\n/** User-facing router tiers in ascending order — \"learning\" is never selected for a user turn, so\n * it has no place in the escalation ladder (#27's ensureRouteModelReady walks this forward only). */\nconst MODEL_ROUTER_TIER_ORDER: readonly (\"cheap\" | \"medium\" | \"expensive\")[] = [\"cheap\", \"medium\", \"expensive\"];\n\n/** How long the #31 \"install ollama now?\" confirm waits before auto-dismissing (same as a \"No\") —\n * long enough to read and decide, short enough that an unattended session doesn't hang a turn on it. */\nconst OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS = 30_000;\n\nexport interface LocalRuntimeControllerDeps {\n\t/** Root directory OllamaRuntime instances are scoped under — fixed for the session's lifetime. */\n\tagentDir: string;\n\t/** Test-injectable seams for OllamaRuntime's own fetch/spawn/exists calls; unset in production. */\n\tlocalRuntimeDeps?: LocalRuntimeDeps;\n\t/** The session's last assistant message, to detect a just-failed local call and drop a stale\n\t * \"confirmed up\" flag. */\n\tgetLastAssistantMessage(): AssistantMessage | undefined;\n\t/** The session's live interactive UI context, if any — undefined in headless/RPC/print modes. */\n\tgetUIContext(): ExtensionUIContext | undefined;\n\t/** Emits a session event (only ever `warning` / `routing_start` / `routing_end` from this controller). */\n\temit(event: AgentSessionEvent): void;\n\t/** Resolves the model configured for a router tier, respecting configured auth — owned by the\n\t * router itself, not this controller. */\n\tresolveConfiguredTierModel(tier: \"medium\" | \"expensive\"): Model<Api> | undefined;\n\t/** `${provider}/${id}` label for a model, for warning/confirm text. */\n\tformatModel(model: Model<Api>): string;\n}\n\nexport class LocalRuntimeController {\n\t/** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */\n\tprivate readonly _runtimes = new Map<string, OllamaRuntime>();\n\t/** Server URLs confirmed reachable THIS session — skips the health-check round trip on every\n\t * local-routed turn once warm. Keyed the same way as _runtimes. */\n\tprivate readonly _confirmedUp = new Set<string>();\n\n\tprivate readonly deps: LocalRuntimeControllerDeps;\n\n\tconstructor(deps: LocalRuntimeControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\t/**\n\t * Shared {@link OllamaRuntime} for a given server, lazily created and cached by baseUrl so every\n\t * caller — the router's readiness gate below and any host UI's own model-lifecycle commands\n\t * (e.g. `/models`) — sees and can stop the SAME pi-managed process instead of each tracking its\n\t * own untracked child.\n\t */\n\tgetLocalRuntime(baseUrl?: string): OllamaRuntime {\n\t\tconst key = baseUrl ?? \"default\";\n\t\tlet runtime = this._runtimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new OllamaRuntime({ agentDir: this.deps.agentDir, baseUrl, deps: this.deps.localRuntimeDeps });\n\t\t\tthis._runtimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\t/** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's\n\t * own health/boot endpoints are on the Ollama-native server root. */\n\tderiveOllamaServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\t/**\n\t * If the last assistant message in this session was an error from THIS exact local server, a\n\t * cached \"confirmed up\" flag would be stale (the server may have died mid-session) — drop it so\n\t * the next ensure-check is a real one instead of trusting stale state.\n\t */\n\tprivate invalidateIfLastCallFailed(model: Model<Api>, serverUrl: string): void {\n\t\tconst lastAssistant = this.deps.getLastAssistantMessage();\n\t\tif (\n\t\t\tlastAssistant?.stopReason === \"error\" &&\n\t\t\tlastAssistant.provider === OLLAMA_PROVIDER &&\n\t\t\tlastAssistant.model === model.id\n\t\t) {\n\t\t\tthis._confirmedUp.delete(serverUrl);\n\t\t}\n\t}\n\n\t/**\n\t * Ensure a routed model is actually reachable before the turn calls it. No-op (and free) for any\n\t * non-local model — this only ever does network/process work for the `ollama` provider. Caches a\n\t * \"confirmed up this session\" flag per server so a steady-state session pays the health-check\n\t * round trip once, not on every turn; invalidated above when a prior local call actually failed,\n\t * so a server that died mid-session gets re-detected rather than trusted forever. Boots via\n\t * `startReuseExisting()` — never owned storage — so the turn sees the user's OWN pulled models,\n\t * the same server `/models` commands and the user's own `ollama` CLI already talk to. Never\n\t * installs anything itself (installGuide is GUIDE MODE: printed, never executed).\n\t */\n\tasync ensureLocalModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== OLLAMA_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_local\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(serverUrl)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(serverUrl);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.binaryPath) {\n\t\t\treturn { ready: false, reason: \"binary_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.startReuseExisting();\n\t\tif (started.started) {\n\t\t\tthis._confirmedUp.add(serverUrl);\n\t\t}\n\t\treturn { ready: started.started, reason: started.reason };\n\t}\n\n\t/**\n\t * #31: the ONE case a routed local model's unreadiness can be fixed automatically is a missing\n\t * ollama binary — an unreachable server can't be helped by installing, so that reason is left to\n\t * the graceful-fallback warning below unchanged. Only offered when there's an interactive UI to\n\t * ask through: headless/RPC/print sessions have no UI context and fall straight through, same as\n\t * declining or timing out (both resolve confirm() to false). Reverses \"pi never runs installers\n\t * itself\" specifically for this one path — the user is asked first, the download is pi's own\n\t * (never curl|sh), and it lands in pi's own runtimes dir (see OllamaRuntime.installManaged).\n\t *\n\t * Pauses/resumes the routing working-indicator around the confirm dialog itself (re-emitting\n\t * routing_end/routing_start — both already idempotent, see interactive-mode.ts's handlers) so an\n\t * animated spinner doesn't fight a dialog the user is trying to read and answer; the indicator\n\t * comes back for the download/extract that follows a \"yes\", which is genuine processing feedback.\n\t */\n\tprivate async maybeInstallOllamaOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"binary_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Ollama?\",\n\t\t\t\t`Ollama isn't installed, so the local model \"${modelLabel}\" can't run. Pi can download and ` +\n\t\t\t\t\t\"install it now (a large one-time download, possibly over 1 GB depending on your platform) \" +\n\t\t\t\t\t\"into its own runtimes folder — never curl|sh, never touching anything outside pi's own \" +\n\t\t\t\t\t\"directory. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"ollama-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"ollama-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\treturn this.ensureLocalModelReady(model);\n\t}\n\n\t/**\n\t * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct\n\t * route — both carry tier \"cheap\") must not dead-end the turn just because ollama isn't up.\n\t * Never a SILENT swap: every fallback is announced in a warning that states (i) the local model\n\t * was unavailable and WHY — binary missing surfaces the install guide inline; any other reason\n\t * gets a \"check that ollama is running\" hint — and (ii) which tier is now handling the turn, so\n\t * the cost shift is never a surprise. Escalates cheap -> medium -> expensive, skipping any\n\t * unconfigured intermediate tier, reusing the router's own existing \"model unavailable\"\n\t * resolution (resolveConfiguredTierModel) rather than inventing a new fallback mechanism.\n\t * Escalation is bounded: tier strictly increases each hop, so it terminates within two hops.\n\t *\n\t * Before the warning/escalation below: #31's consent gate gets one shot at fixing a missing\n\t * binary interactively (see maybeInstallOllamaOnConsent) — declining, timing out, running\n\t * headless, or the install attempt itself failing all fall through here unchanged, just with an\n\t * honest reason (an install that failed is worded as a failed install, not re-labeled as if\n\t * nothing was ever tried).\n\t */\n\tasync ensureRouteModelReady(\n\t\tresolved: { decision: RouteDecision; model: Model<Api> } | undefined,\n\t): Promise<{ decision: RouteDecision; model: Model<Api> } | undefined> {\n\t\tlet current = resolved;\n\t\twhile (current && current.model.provider === OLLAMA_PROVIDER) {\n\t\t\tlet readiness: { ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string } =\n\t\t\t\tawait this.ensureLocalModelReady(current.model);\n\t\t\tif (!readiness.ready) {\n\t\t\t\treadiness = await this.maybeInstallOllamaOnConsent(current.model, readiness);\n\t\t\t}\n\t\t\tif (readiness.ready) return current;\n\n\t\t\t// Walk the remaining tiers in order (never back down to cheap) and take the first one that\n\t\t\t// actually resolves — an unconfigured intermediate tier (e.g. no mediumModel set) must be\n\t\t\t// skipped, not treated as \"no fallback available\".\n\t\t\tconst startIndex = MODEL_ROUTER_TIER_ORDER.indexOf(current.decision.tier as \"cheap\" | \"medium\" | \"expensive\");\n\t\t\tlet escalated: { tier: \"medium\" | \"expensive\"; model: Model<Api> } | undefined;\n\t\t\tfor (let i = startIndex + 1; startIndex !== -1 && i < MODEL_ROUTER_TIER_ORDER.length; i++) {\n\t\t\t\tconst tier = MODEL_ROUTER_TIER_ORDER[i] as \"medium\" | \"expensive\";\n\t\t\t\tconst model = this.deps.resolveConfiguredTierModel(tier);\n\t\t\t\tif (model) {\n\t\t\t\t\tescalated = { tier, model };\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tconst modelLabel = this.deps.formatModel(current.model);\n\t\t\tconst whyText = readiness.installAttemptError\n\t\t\t\t? `pi tried to install it just now, but the install attempt failed: ${readiness.installAttemptError}`\n\t\t\t\t: readiness.installGuide\n\t\t\t\t\t? [\"the ollama binary is not installed.\", ...readiness.installGuide].join(\"\\n\")\n\t\t\t\t\t: `its server is not reachable (${readiness.reason}) — check that ollama is running.`;\n\t\t\tconst fallbackText = escalated\n\t\t\t\t? `Falling back to the ${escalated.tier} tier for this turn.`\n\t\t\t\t: \"No other tier is configured — falling back to the session's default model.\";\n\t\t\tthis.deps.emit({\n\t\t\t\ttype: \"warning\",\n\t\t\t\tmessage: `Local model \"${modelLabel}\" is unavailable: ${whyText}\\n${fallbackText}`,\n\t\t\t});\n\n\t\t\tif (!escalated) return undefined; // no higher tier resolves — caller falls back to the session default\n\t\t\tcurrent = {\n\t\t\t\tmodel: escalated.model,\n\t\t\t\tdecision: {\n\t\t\t\t\t...current.decision,\n\t\t\t\t\ttier: escalated.tier,\n\t\t\t\t\tfallbackFrom: current.decision.tier,\n\t\t\t\t\treasonCode: \"local_model_not_ready_fallback\",\n\t\t\t\t\treasons: [\n\t\t\t\t\t\t...current.decision.reasons,\n\t\t\t\t\t\t`Local model not ready (${readiness.reason}); escalated to ${escalated.tier}`,\n\t\t\t\t\t],\n\t\t\t\t\tmodel: this.deps.formatModel(escalated.model),\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\treturn current;\n\t}\n}\n"]}
+{"version":3,"file":"local-runtime-controller.d.ts","sourceRoot":"","sources":["../../src/core/local-runtime-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,GAAG,EAAE,gBAAgB,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AACtE,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AAC5D,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,yBAAyB,CAAC;AAC7D,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,uBAAuB,CAAC;AAEhE,OAAO,EACN,KAAK,gBAAgB,EACrB,aAAa,EAEb,mBAAmB,EACnB,MAAM,2BAA2B,CAAC;AAUnC,MAAM,WAAW,0BAA0B;IAC1C,oGAAkG;IAClG,QAAQ,EAAE,MAAM,CAAC;IACjB,mGAAmG;IACnG,gBAAgB,CAAC,EAAE,gBAAgB,CAAC;IACpC;8BAC0B;IAC1B,uBAAuB,IAAI,gBAAgB,GAAG,SAAS,CAAC;IACxD,mGAAiG;IACjG,YAAY,IAAI,kBAAkB,GAAG,SAAS,CAAC;IAC/C,0GAA0G;IAC1G,IAAI,CAAC,KAAK,EAAE,iBAAiB,GAAG,IAAI,CAAC;IACrC;6CACyC;IACzC,0BAA0B,CAAC,IAAI,EAAE,QAAQ,GAAG,WAAW,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IACjF,uEAAuE;IACvE,WAAW,CAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC;CACvC;AAED,qBAAa,sBAAsB;IAClC,qGAAqG;IACrG,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAoC;IAC9D,uGAAuG;IACvG,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAA0C;IAChF;uEACmE;IACnE,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAqB;IAElD,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA6B;IAElD,YAAY,IAAI,EAAE,0BAA0B,EAE3C;IAED;;;;;OAKG;IACH,eAAe,CAAC,OAAO,CAAC,EAAE,MAAM,GAAG,aAAa,CAQ/C;IAED,sBAAsB,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,mBAAmB,CAc7E;IAED;yEACqE;IACrE,qBAAqB,CAAC,YAAY,EAAE,MAAM,GAAG,MAAM,CAElD;IAED,OAAO,CAAC,2BAA2B;IAInC,OAAO,CAAC,sBAAsB;IAI9B;;;;OAIG;IACH,OAAO,CAAC,eAAe;IAIvB,OAAO,CAAC,0BAA0B;IAWlC;;;;;OAKG;IACG,qBAAqB,CAC1B,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GACf,OAAO,CAAC;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC,CAwBtE;IAEK,4BAA4B,CACjC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,GACf,OAAO,CAAC;QAAE,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;KAAE,CAAC,CAyBtE;YAgBa,2BAA2B;YAsC3B,iCAAiC;IA+C/C;;;;;;;;;;;;;;;;OAgBG;IACG,qBAAqB,CAC1B,QAAQ,EAAE;QAAE,QAAQ,EAAE,aAAa,CAAC;QAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAA;KAAE,GAAG,SAAS,GAClE,OAAO,CAAC;QAAE,QAAQ,EAAE,aAAa,CAAC;QAAC,KAAK,EAAE,KAAK,CAAC,GAAG,CAAC,CAAA;KAAE,GAAG,SAAS,CAAC,CAoErE;CACD","sourcesContent":["/**\n * Local-runtime (Ollama) lifecycle controller.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Owns the cached, per-server\n * {@link OllamaRuntime} instances, the \"confirmed up this session\" flag, and the router's readiness\n * gate for a turn routed to a local (`ollama`) model — including the #31 install-on-consent flow and\n * the #27 graceful tier-escalation fallback. Takes narrow deps (agent dir, a last-assistant-message\n * accessor, the session's UI context/event emitter, and the router's own tier resolver) rather than\n * the whole AgentSession.\n */\n\nimport type { Api, AssistantMessage, Model } from \"@caupulican/pi-ai\";\nimport type { AgentSessionEvent } from \"./agent-session.ts\";\nimport type { RouteDecision } from \"./autonomy/contracts.ts\";\nimport type { ExtensionUIContext } from \"./extensions/index.ts\";\nimport { HF_TRANSFORMERS_PROVIDER, OLLAMA_PROVIDER } from \"./models/local-registration.ts\";\nimport {\n\ttype LocalRuntimeDeps,\n\tOllamaRuntime,\n\tresolveTransformersBaseUrl,\n\tTransformersRuntime,\n} from \"./models/local-runtime.ts\";\n\n/** User-facing router tiers in ascending order — \"learning\" is never selected for a user turn, so\n * it has no place in the escalation ladder (#27's ensureRouteModelReady walks this forward only). */\nconst MODEL_ROUTER_TIER_ORDER: readonly (\"cheap\" | \"medium\" | \"expensive\")[] = [\"cheap\", \"medium\", \"expensive\"];\n\n/** How long the #31 \"install ollama now?\" confirm waits before auto-dismissing (same as a \"No\") —\n * long enough to read and decide, short enough that an unattended session doesn't hang a turn on it. */\nconst OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS = 30_000;\n\nexport interface LocalRuntimeControllerDeps {\n\t/** Root directory OllamaRuntime instances are scoped under — fixed for the session's lifetime. */\n\tagentDir: string;\n\t/** Test-injectable seams for OllamaRuntime's own fetch/spawn/exists calls; unset in production. */\n\tlocalRuntimeDeps?: LocalRuntimeDeps;\n\t/** The session's last assistant message, to detect a just-failed local call and drop a stale\n\t * \"confirmed up\" flag. */\n\tgetLastAssistantMessage(): AssistantMessage | undefined;\n\t/** The session's live interactive UI context, if any — undefined in headless/RPC/print modes. */\n\tgetUIContext(): ExtensionUIContext | undefined;\n\t/** Emits a session event (only ever `warning` / `routing_start` / `routing_end` from this controller). */\n\temit(event: AgentSessionEvent): void;\n\t/** Resolves the model configured for a router tier, respecting configured auth — owned by the\n\t * router itself, not this controller. */\n\tresolveConfiguredTierModel(tier: \"medium\" | \"expensive\"): Model<Api> | undefined;\n\t/** `${provider}/${id}` label for a model, for warning/confirm text. */\n\tformatModel(model: Model<Api>): string;\n}\n\nexport class LocalRuntimeController {\n\t/** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */\n\tprivate readonly _runtimes = new Map<string, OllamaRuntime>();\n\t/** Lazy, cached by model+baseUrl so the router and `/models` share one sidecar handle per HF model. */\n\tprivate readonly _transformersRuntimes = new Map<string, TransformersRuntime>();\n\t/** Server URLs confirmed reachable THIS session — skips the health-check round trip on every\n\t * local-routed turn once warm. Keyed the same way as _runtimes. */\n\tprivate readonly _confirmedUp = new Set<string>();\n\n\tprivate readonly deps: LocalRuntimeControllerDeps;\n\n\tconstructor(deps: LocalRuntimeControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\t/**\n\t * Shared {@link OllamaRuntime} for a given server, lazily created and cached by baseUrl so every\n\t * caller — the router's readiness gate below and any host UI's own model-lifecycle commands\n\t * (e.g. `/models`) — sees and can stop the SAME pi-managed process instead of each tracking its\n\t * own untracked child.\n\t */\n\tgetLocalRuntime(baseUrl?: string): OllamaRuntime {\n\t\tconst key = baseUrl ?? \"default\";\n\t\tlet runtime = this._runtimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new OllamaRuntime({ agentDir: this.deps.agentDir, baseUrl, deps: this.deps.localRuntimeDeps });\n\t\t\tthis._runtimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\tgetTransformersRuntime(modelId: string, baseUrl?: string): TransformersRuntime {\n\t\tconst resolvedBaseUrl = baseUrl?.replace(/\\/$/, \"\") ?? resolveTransformersBaseUrl(modelId);\n\t\tconst key = `${modelId}\\0${resolvedBaseUrl}`;\n\t\tlet runtime = this._transformersRuntimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new TransformersRuntime({\n\t\t\t\tagentDir: this.deps.agentDir,\n\t\t\t\tmodelId,\n\t\t\t\tbaseUrl: resolvedBaseUrl,\n\t\t\t\tdeps: this.deps.localRuntimeDeps,\n\t\t\t});\n\t\t\tthis._transformersRuntimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\t/** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's\n\t * own health/boot endpoints are on the Ollama-native server root. */\n\tderiveOllamaServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\tprivate deriveOpenAICompatServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\tprivate isManagedLocalProvider(provider: string): boolean {\n\t\treturn provider === OLLAMA_PROVIDER || provider === HF_TRANSFORMERS_PROVIDER;\n\t}\n\n\t/**\n\t * If the last assistant message in this session was an error from THIS exact local server, a\n\t * cached \"confirmed up\" flag would be stale (the server may have died mid-session) — drop it so\n\t * the next ensure-check is a real one instead of trusting stale state.\n\t */\n\tprivate confirmationKey(model: Model<Api>, serverUrl: string): string {\n\t\treturn model.provider === HF_TRANSFORMERS_PROVIDER ? `${serverUrl}\\0${model.id}` : serverUrl;\n\t}\n\n\tprivate invalidateIfLastCallFailed(model: Model<Api>, serverUrl: string): void {\n\t\tconst lastAssistant = this.deps.getLastAssistantMessage();\n\t\tif (\n\t\t\tlastAssistant?.stopReason === \"error\" &&\n\t\t\tlastAssistant.provider === model.provider &&\n\t\t\tlastAssistant.model === model.id\n\t\t) {\n\t\t\tthis._confirmedUp.delete(this.confirmationKey(model, serverUrl));\n\t\t}\n\t}\n\n\t/**\n\t * Ensure a routed managed-local model is actually reachable before the turn calls it. No-op (and\n\t * free) for non-local/API models. Caches a \"confirmed up this session\" flag per server (and per\n\t * Transformers model) so steady-state routing pays the health-check round trip once; invalidated\n\t * above when a prior local call failed so a dead sidecar gets re-detected instead of trusted.\n\t */\n\tasync ensureLocalModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== OLLAMA_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_local\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst confirmedKey = this.confirmationKey(model, serverUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(confirmedKey)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.binaryPath) {\n\t\t\treturn { ready: false, reason: \"binary_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.startReuseExisting();\n\t\tif (started.started) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t}\n\t\treturn { ready: started.started, reason: started.reason };\n\t}\n\n\tasync ensureTransformersModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== HF_TRANSFORMERS_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_transformers\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);\n\t\tconst confirmedKey = this.confirmationKey(model, serverUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(confirmedKey)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getTransformersRuntime(model.id, serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.runtimeInstalled) {\n\t\t\treturn { ready: false, reason: \"runtime_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.start();\n\t\tif (started.started || started.reason === \"already_running\") {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: started.reason };\n\t\t}\n\t\treturn { ready: false, reason: started.reason };\n\t}\n\n\t/**\n\t * #31: the ONE case a routed local model's unreadiness can be fixed automatically is a missing\n\t * ollama binary — an unreachable server can't be helped by installing, so that reason is left to\n\t * the graceful-fallback warning below unchanged. Only offered when there's an interactive UI to\n\t * ask through: headless/RPC/print sessions have no UI context and fall straight through, same as\n\t * declining or timing out (both resolve confirm() to false). Reverses \"pi never runs installers\n\t * itself\" specifically for this one path — the user is asked first, the download is pi's own\n\t * (never curl|sh), and it lands in pi's own runtimes dir (see OllamaRuntime.installManaged).\n\t *\n\t * Pauses/resumes the routing working-indicator around the confirm dialog itself (re-emitting\n\t * routing_end/routing_start — both already idempotent, see interactive-mode.ts's handlers) so an\n\t * animated spinner doesn't fight a dialog the user is trying to read and answer; the indicator\n\t * comes back for the download/extract that follows a \"yes\", which is genuine processing feedback.\n\t */\n\tprivate async maybeInstallOllamaOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"binary_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Ollama?\",\n\t\t\t\t`Ollama isn't installed, so the local model \"${modelLabel}\" can't run. Pi can download and ` +\n\t\t\t\t\t\"install it now (a large one-time download, possibly over 1 GB depending on your platform) \" +\n\t\t\t\t\t\"into its own runtimes folder — never curl|sh, never touching anything outside pi's own \" +\n\t\t\t\t\t\"directory. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"ollama-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"ollama-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\treturn this.ensureLocalModelReady(model);\n\t}\n\n\tprivate async maybeInstallTransformersOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"runtime_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Transformers runtime?\",\n\t\t\t\t`The Hugging Face model \"${modelLabel}\" needs a pi-managed Python venv with Transformers ` +\n\t\t\t\t\t\"and CPU PyTorch before it can run. Pi will install those packages into its own runtimes \" +\n\t\t\t\t\t\"folder, download the model into a pi-owned Hugging Face cache, and leave system Python, \" +\n\t\t\t\t\t\"your Ollama models, and your user HF cache untouched. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);\n\t\tconst runtime = this.getTransformersRuntime(model.id, serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"transformers-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"transformers-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\tlet downloadResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tdownloadResult = await runtime.downloadModel((status) => ui.setStatus(\"transformers-download\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"transformers-download\", undefined);\n\t\t}\n\t\tif (!downloadResult.ok) {\n\t\t\treturn { ready: false, reason: \"download_failed\", installAttemptError: downloadResult.error };\n\t\t}\n\t\treturn this.ensureTransformersModelReady(model);\n\t}\n\n\t/**\n\t * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct\n\t * route — both carry tier \"cheap\") must not dead-end the turn just because ollama isn't up.\n\t * Never a SILENT swap: every fallback is announced in a warning that states (i) the local model\n\t * was unavailable and WHY — binary missing surfaces the install guide inline; any other reason\n\t * gets a \"check that ollama is running\" hint — and (ii) which tier is now handling the turn, so\n\t * the cost shift is never a surprise. Escalates cheap -> medium -> expensive, skipping any\n\t * unconfigured intermediate tier, reusing the router's own existing \"model unavailable\"\n\t * resolution (resolveConfiguredTierModel) rather than inventing a new fallback mechanism.\n\t * Escalation is bounded: tier strictly increases each hop, so it terminates within two hops.\n\t *\n\t * Before the warning/escalation below: #31's consent gate gets one shot at fixing a missing\n\t * binary interactively (see maybeInstallOllamaOnConsent) — declining, timing out, running\n\t * headless, or the install attempt itself failing all fall through here unchanged, just with an\n\t * honest reason (an install that failed is worded as a failed install, not re-labeled as if\n\t * nothing was ever tried).\n\t */\n\tasync ensureRouteModelReady(\n\t\tresolved: { decision: RouteDecision; model: Model<Api> } | undefined,\n\t): Promise<{ decision: RouteDecision; model: Model<Api> } | undefined> {\n\t\tlet current = resolved;\n\t\twhile (current && this.isManagedLocalProvider(current.model.provider)) {\n\t\t\tlet readiness: { ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string } =\n\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t? await this.ensureLocalModelReady(current.model)\n\t\t\t\t\t: await this.ensureTransformersModelReady(current.model);\n\t\t\tif (!readiness.ready) {\n\t\t\t\treadiness =\n\t\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t? await this.maybeInstallOllamaOnConsent(current.model, readiness)\n\t\t\t\t\t\t: await this.maybeInstallTransformersOnConsent(current.model, readiness);\n\t\t\t}\n\t\t\tif (readiness.ready) return current;\n\n\t\t\t// Walk the remaining tiers in order (never back down to cheap) and take the first one that\n\t\t\t// actually resolves — an unconfigured intermediate tier (e.g. no mediumModel set) must be\n\t\t\t// skipped, not treated as \"no fallback available\".\n\t\t\tconst startIndex = MODEL_ROUTER_TIER_ORDER.indexOf(current.decision.tier as \"cheap\" | \"medium\" | \"expensive\");\n\t\t\tlet escalated: { tier: \"medium\" | \"expensive\"; model: Model<Api> } | undefined;\n\t\t\tfor (let i = startIndex + 1; startIndex !== -1 && i < MODEL_ROUTER_TIER_ORDER.length; i++) {\n\t\t\t\tconst tier = MODEL_ROUTER_TIER_ORDER[i] as \"medium\" | \"expensive\";\n\t\t\t\tconst model = this.deps.resolveConfiguredTierModel(tier);\n\t\t\t\tif (model) {\n\t\t\t\t\tescalated = { tier, model };\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tconst modelLabel = this.deps.formatModel(current.model);\n\t\t\tconst localRuntimeName = current.model.provider === OLLAMA_PROVIDER ? \"ollama\" : \"Transformers\";\n\t\t\tconst whyText = readiness.installAttemptError\n\t\t\t\t? `pi tried to install it just now, but the install attempt failed: ${readiness.installAttemptError}`\n\t\t\t\t: readiness.installGuide\n\t\t\t\t\t? [\n\t\t\t\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t\t\t? \"the ollama binary is not installed.\"\n\t\t\t\t\t\t\t\t: \"the pi-managed Transformers runtime is not installed.\",\n\t\t\t\t\t\t\t...readiness.installGuide,\n\t\t\t\t\t\t].join(\"\\n\")\n\t\t\t\t\t: current.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t? `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that ollama is running.`\n\t\t\t\t\t\t: `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that the runtime is running.`;\n\t\t\tconst fallbackText = escalated\n\t\t\t\t? `Falling back to the ${escalated.tier} tier for this turn.`\n\t\t\t\t: \"No other tier is configured — falling back to the session's default model.\";\n\t\t\tthis.deps.emit({\n\t\t\t\ttype: \"warning\",\n\t\t\t\tmessage: `Local model \"${modelLabel}\" is unavailable: ${whyText}\\n${fallbackText}`,\n\t\t\t});\n\n\t\t\tif (!escalated) return undefined; // no higher tier resolves — caller falls back to the session default\n\t\t\tcurrent = {\n\t\t\t\tmodel: escalated.model,\n\t\t\t\tdecision: {\n\t\t\t\t\t...current.decision,\n\t\t\t\t\ttier: escalated.tier,\n\t\t\t\t\tfallbackFrom: current.decision.tier,\n\t\t\t\t\treasonCode: \"local_model_not_ready_fallback\",\n\t\t\t\t\treasons: [\n\t\t\t\t\t\t...current.decision.reasons,\n\t\t\t\t\t\t`Local model not ready (${readiness.reason}); escalated to ${escalated.tier}`,\n\t\t\t\t\t],\n\t\t\t\t\tmodel: this.deps.formatModel(escalated.model),\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\treturn current;\n\t}\n}\n"]}

package/dist/core/local-runtime-controller.js

@@ -8,8 +8,8 @@
  * accessor, the session's UI context/event emitter, and the router's own tier resolver) rather than
  * the whole AgentSession.
  */
-import { OLLAMA_PROVIDER } from "./models/local-registration.js";
-import { OllamaRuntime } from "./models/local-runtime.js";
+import { HF_TRANSFORMERS_PROVIDER, OLLAMA_PROVIDER } from "./models/local-registration.js";
+import { OllamaRuntime, resolveTransformersBaseUrl, TransformersRuntime, } from "./models/local-runtime.js";
 /** User-facing router tiers in ascending order — "learning" is never selected for a user turn, so
  * it has no place in the escalation ladder (#27's ensureRouteModelReady walks this forward only). */
 const MODEL_ROUTER_TIER_ORDER = ["cheap", "medium", "expensive"];
@@ -19,6 +19,8 @@ const OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS = 30_000;
 export class LocalRuntimeController {
     /** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */
     _runtimes = new Map();
+    /** Lazy, cached by model+baseUrl so the router and `/models` share one sidecar handle per HF model. */
+    _transformersRuntimes = new Map();
     /** Server URLs confirmed reachable THIS session — skips the health-check round trip on every
      * local-routed turn once warm. Keyed the same way as _runtimes. */
     _confirmedUp = new Set();
@@ -41,47 +43,68 @@ export class LocalRuntimeController {
         }
         return runtime;
     }
+    getTransformersRuntime(modelId, baseUrl) {
+        const resolvedBaseUrl = baseUrl?.replace(/\/$/, "") ?? resolveTransformersBaseUrl(modelId);
+        const key = `${modelId}\0${resolvedBaseUrl}`;
+        let runtime = this._transformersRuntimes.get(key);
+        if (!runtime) {
+            runtime = new TransformersRuntime({
+                agentDir: this.deps.agentDir,
+                modelId,
+                baseUrl: resolvedBaseUrl,
+                deps: this.deps.localRuntimeDeps,
+            });
+            this._transformersRuntimes.set(key, runtime);
+        }
+        return runtime;
+    }
     /** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's
      * own health/boot endpoints are on the Ollama-native server root. */
     deriveOllamaServerUrl(modelBaseUrl) {
         return modelBaseUrl.replace(/\/v1\/?$/, "");
     }
+    deriveOpenAICompatServerUrl(modelBaseUrl) {
+        return modelBaseUrl.replace(/\/v1\/?$/, "");
+    }
+    isManagedLocalProvider(provider) {
+        return provider === OLLAMA_PROVIDER || provider === HF_TRANSFORMERS_PROVIDER;
+    }
     /**
      * If the last assistant message in this session was an error from THIS exact local server, a
      * cached "confirmed up" flag would be stale (the server may have died mid-session) — drop it so
      * the next ensure-check is a real one instead of trusting stale state.
      */
+    confirmationKey(model, serverUrl) {
+        return model.provider === HF_TRANSFORMERS_PROVIDER ? `${serverUrl}\0${model.id}` : serverUrl;
+    }
     invalidateIfLastCallFailed(model, serverUrl) {
         const lastAssistant = this.deps.getLastAssistantMessage();
         if (lastAssistant?.stopReason === "error" &&
-            lastAssistant.provider === OLLAMA_PROVIDER &&
+            lastAssistant.provider === model.provider &&
             lastAssistant.model === model.id) {
-            this._confirmedUp.delete(serverUrl);
+            this._confirmedUp.delete(this.confirmationKey(model, serverUrl));
         }
     }
     /**
-     * Ensure a routed model is actually reachable before the turn calls it. No-op (and free) for any
-     * non-local model — this only ever does network/process work for the `ollama` provider. Caches a
-     * "confirmed up this session" flag per server so a steady-state session pays the health-check
-     * round trip once, not on every turn; invalidated above when a prior local call actually failed,
-     * so a server that died mid-session gets re-detected rather than trusted forever. Boots via
-     * `startReuseExisting()` — never owned storage — so the turn sees the user's OWN pulled models,
-     * the same server `/models` commands and the user's own `ollama` CLI already talk to. Never
-     * installs anything itself (installGuide is GUIDE MODE: printed, never executed).
+     * Ensure a routed managed-local model is actually reachable before the turn calls it. No-op (and
+     * free) for non-local/API models. Caches a "confirmed up this session" flag per server (and per
+     * Transformers model) so steady-state routing pays the health-check round trip once; invalidated
+     * above when a prior local call failed so a dead sidecar gets re-detected instead of trusted.
      */
     async ensureLocalModelReady(model) {
         if (model.provider !== OLLAMA_PROVIDER) {
             return { ready: true, reason: "not_local" };
         }
         const serverUrl = this.deriveOllamaServerUrl(model.baseUrl);
+        const confirmedKey = this.confirmationKey(model, serverUrl);
         this.invalidateIfLastCallFailed(model, serverUrl);
-        if (this._confirmedUp.has(serverUrl)) {
+        if (this._confirmedUp.has(confirmedKey)) {
             return { ready: true, reason: "confirmed_up_cached" };
         }
         const runtime = this.getLocalRuntime(serverUrl);
         const status = await runtime.detect();
         if (status.serverUp) {
-            this._confirmedUp.add(serverUrl);
+            this._confirmedUp.add(confirmedKey);
             return { ready: true, reason: "already_running" };
         }
         if (!status.binaryPath) {
@@ -89,10 +112,36 @@ export class LocalRuntimeController {
         }
         const started = await runtime.startReuseExisting();
         if (started.started) {
-            this._confirmedUp.add(serverUrl);
+            this._confirmedUp.add(confirmedKey);
         }
         return { ready: started.started, reason: started.reason };
     }
+    async ensureTransformersModelReady(model) {
+        if (model.provider !== HF_TRANSFORMERS_PROVIDER) {
+            return { ready: true, reason: "not_transformers" };
+        }
+        const serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);
+        const confirmedKey = this.confirmationKey(model, serverUrl);
+        this.invalidateIfLastCallFailed(model, serverUrl);
+        if (this._confirmedUp.has(confirmedKey)) {
+            return { ready: true, reason: "confirmed_up_cached" };
+        }
+        const runtime = this.getTransformersRuntime(model.id, serverUrl);
+        const status = await runtime.detect();
+        if (status.serverUp) {
+            this._confirmedUp.add(confirmedKey);
+            return { ready: true, reason: "already_running" };
+        }
+        if (!status.runtimeInstalled) {
+            return { ready: false, reason: "runtime_missing", installGuide: runtime.installGuide() };
+        }
+        const started = await runtime.start();
+        if (started.started || started.reason === "already_running") {
+            this._confirmedUp.add(confirmedKey);
+            return { ready: true, reason: started.reason };
+        }
+        return { ready: false, reason: started.reason };
+    }
     /**
      * #31: the ONE case a routed local model's unreadiness can be fixed automatically is a missing
      * ollama binary — an unreachable server can't be helped by installing, so that reason is left to
@@ -139,6 +188,48 @@ export class LocalRuntimeController {
         }
         return this.ensureLocalModelReady(model);
     }
+    async maybeInstallTransformersOnConsent(model, readiness) {
+        const ui = this.deps.getUIContext();
+        if (!ui || readiness.ready || readiness.reason !== "runtime_missing")
+            return readiness;
+        const modelLabel = this.deps.formatModel(model);
+        this.deps.emit({ type: "routing_end" });
+        let confirmed;
+        try {
+            confirmed = await ui.confirm("Install Transformers runtime?", `The Hugging Face model "${modelLabel}" needs a pi-managed Python venv with Transformers ` +
+                "and CPU PyTorch before it can run. Pi will install those packages into its own runtimes " +
+                "folder, download the model into a pi-owned Hugging Face cache, and leave system Python, " +
+                "your Ollama models, and your user HF cache untouched. Install it now?", { timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS });
+        }
+        finally {
+            this.deps.emit({ type: "routing_start" });
+        }
+        if (!confirmed)
+            return readiness;
+        const serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);
+        const runtime = this.getTransformersRuntime(model.id, serverUrl);
+        let installResult;
+        try {
+            installResult = await runtime.installManaged((status) => ui.setStatus("transformers-install", status));
+        }
+        finally {
+            ui.setStatus("transformers-install", undefined);
+        }
+        if (!installResult.ok) {
+            return { ready: false, reason: "install_failed", installAttemptError: installResult.error };
+        }
+        let downloadResult;
+        try {
+            downloadResult = await runtime.downloadModel((status) => ui.setStatus("transformers-download", status));
+        }
+        finally {
+            ui.setStatus("transformers-download", undefined);
+        }
+        if (!downloadResult.ok) {
+            return { ready: false, reason: "download_failed", installAttemptError: downloadResult.error };
+        }
+        return this.ensureTransformersModelReady(model);
+    }
     /**
      * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct
      * route — both carry tier "cheap") must not dead-end the turn just because ollama isn't up.
@@ -158,10 +249,15 @@ export class LocalRuntimeController {
      */
     async ensureRouteModelReady(resolved) {
         let current = resolved;
-        while (current && current.model.provider === OLLAMA_PROVIDER) {
-            let readiness = await this.ensureLocalModelReady(current.model);
+        while (current && this.isManagedLocalProvider(current.model.provider)) {
+            let readiness = current.model.provider === OLLAMA_PROVIDER
+                ? await this.ensureLocalModelReady(current.model)
+                : await this.ensureTransformersModelReady(current.model);
             if (!readiness.ready) {
-                readiness = await this.maybeInstallOllamaOnConsent(current.model, readiness);
+                readiness =
+                    current.model.provider === OLLAMA_PROVIDER
+                        ? await this.maybeInstallOllamaOnConsent(current.model, readiness)
+                        : await this.maybeInstallTransformersOnConsent(current.model, readiness);
             }
             if (readiness.ready)
                 return current;
@@ -179,11 +275,19 @@ export class LocalRuntimeController {
                 }
             }
             const modelLabel = this.deps.formatModel(current.model);
+            const localRuntimeName = current.model.provider === OLLAMA_PROVIDER ? "ollama" : "Transformers";
             const whyText = readiness.installAttemptError
                 ? `pi tried to install it just now, but the install attempt failed: ${readiness.installAttemptError}`
                 : readiness.installGuide
-                    ? ["the ollama binary is not installed.", ...readiness.installGuide].join("\n")
-                    : `its server is not reachable (${readiness.reason}) — check that ollama is running.`;
+                    ? [
+                        current.model.provider === OLLAMA_PROVIDER
+                            ? "the ollama binary is not installed."
+                            : "the pi-managed Transformers runtime is not installed.",
+                        ...readiness.installGuide,
+                    ].join("\n")
+                    : current.model.provider === OLLAMA_PROVIDER
+                        ? `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that ollama is running.`
+                        : `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that the runtime is running.`;
             const fallbackText = escalated
                 ? `Falling back to the ${escalated.tier} tier for this turn.`
                 : "No other tier is configured — falling back to the session's default model.";

package/dist/core/local-runtime-controller.js.map

@@ -1 +1 @@
-{"version":3,"file":"local-runtime-controller.js","sourceRoot":"","sources":["../../src/core/local-runtime-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAMH,OAAO,EAAE,eAAe,EAAE,MAAM,gCAAgC,CAAC;AACjE,OAAO,EAAyB,aAAa,EAAE,MAAM,2BAA2B,CAAC;AAEjF;qGACqG;AACrG,MAAM,uBAAuB,GAAkD,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC;AAEhH;wGACwG;AACxG,MAAM,iCAAiC,GAAG,MAAM,CAAC;AAqBjD,MAAM,OAAO,sBAAsB;IAClC,qGAAqG;IACpF,SAAS,GAAG,IAAI,GAAG,EAAyB,CAAC;IAC9D;uEACmE;IAClD,YAAY,GAAG,IAAI,GAAG,EAAU,CAAC;IAEjC,IAAI,CAA6B;IAElD,YAAY,IAAgC,EAAE;QAC7C,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IAAA,CACjB;IAED;;;;;OAKG;IACH,eAAe,CAAC,OAAgB,EAAiB;QAChD,MAAM,GAAG,GAAG,OAAO,IAAI,SAAS,CAAC;QACjC,IAAI,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACtC,IAAI,CAAC,OAAO,EAAE,CAAC;YACd,OAAO,GAAG,IAAI,aAAa,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC,CAAC;YACzG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QAClC,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED;yEACqE;IACrE,qBAAqB,CAAC,YAAoB,EAAU;QACnD,OAAO,YAAY,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAED;;;;OAIG;IACK,0BAA0B,CAAC,KAAiB,EAAE,SAAiB,EAAQ;QAC9E,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,uBAAuB,EAAE,CAAC;QAC1D,IACC,aAAa,EAAE,UAAU,KAAK,OAAO;YACrC,aAAa,CAAC,QAAQ,KAAK,eAAe;YAC1C,aAAa,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,EAC/B,CAAC;YACF,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QACrC,CAAC;IAAA,CACD;IAED;;;;;;;;;OASG;IACH,KAAK,CAAC,qBAAqB,CAC1B,KAAiB,EACsD;QACvE,IAAI,KAAK,CAAC,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;QAC7C,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC5D,IAAI,CAAC,0BAA0B,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;QAClD,IAAI,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;YACtC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,qBAAqB,EAAE,CAAC;QACvD,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE,CAAC;QACtC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;YACjC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,iBAAiB,EAAE,CAAC;QACnD,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,CAAC;YACxB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,EAAE,YAAY,EAAE,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC;QACzF,CAAC;QACD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,kBAAkB,EAAE,CAAC;QACnD,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YACrB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAClC,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;IAAA,CAC1D;IAED;;;;;;;;;;;;;OAaG;IACK,KAAK,CAAC,2BAA2B,CACxC,KAAiB,EACjB,SAAsE,EAC+B;QACrG,MAAM,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;QACpC,IAAI,CAAC,EAAE,IAAI,SAAS,CAAC,KAAK,IAAI,SAAS,CAAC,MAAM,KAAK,gBAAgB;YAAE,OAAO,SAAS,CAAC;QAEtF,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAChD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC,CAAC;QACxC,IAAI,SAAkB,CAAC;QACvB,IAAI,CAAC;YACJ,SAAS,GAAG,MAAM,EAAE,CAAC,OAAO,CAC3B,iBAAiB,EACjB,+CAA+C,UAAU,mCAAmC;gBAC3F,4FAA4F;gBAC5F,2FAAyF;gBACzF,4BAA4B,EAC7B,EAAE,OAAO,EAAE,iCAAiC,EAAE,CAC9C,CAAC;QACH,CAAC;gBAAS,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,SAAS;YAAE,OAAO,SAAS,CAAC;QAEjC,MAAM,SAAS,GAAG,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC5D,MAAM,OAAO,GAAG,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QAChD,IAAI,aAA8C,CAAC;QACnD,IAAI,CAAC;YACJ,aAAa,GAAG,MAAM,OAAO,CAAC,cAAc,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,SAAS,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;QAClG,CAAC;gBAAS,CAAC;YACV,EAAE,CAAC,SAAS,CAAC,gBAAgB,EAAE,SAAS,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,aAAa,CAAC,EAAE,EAAE,CAAC;YACvB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,aAAa,CAAC,KAAK,EAAE,CAAC;QAC7F,CAAC;QACD,OAAO,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAAA,CACzC;IAED;;;;;;;;;;;;;;;;OAgBG;IACH,KAAK,CAAC,qBAAqB,CAC1B,QAAoE,EACE;QACtE,IAAI,OAAO,GAAG,QAAQ,CAAC;QACvB,OAAO,OAAO,IAAI,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe,EAAE,CAAC;YAC9D,IAAI,SAAS,GACZ,MAAM,IAAI,CAAC,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YACjD,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;gBACtB,SAAS,GAAG,MAAM,IAAI,CAAC,2BAA2B,CAAC,OAAO,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;YAC9E,CAAC;YACD,IAAI,SAAS,CAAC,KAAK;gBAAE,OAAO,OAAO,CAAC;YAEpC,2FAA2F;YAC3F,4FAA0F;YAC1F,mDAAmD;YACnD,MAAM,UAAU,GAAG,uBAAuB,CAAC,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAwC,CAAC,CAAC;YAC9G,IAAI,SAA0E,CAAC;YAC/E,KAAK,IAAI,CAAC,GAAG,UAAU,GAAG,CAAC,EAAE,UAAU,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,uBAAuB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC3F,MAAM,IAAI,GAAG,uBAAuB,CAAC,CAAC,CAA2B,CAAC;gBAClE,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,CAAC;gBACzD,IAAI,KAAK,EAAE,CAAC;oBACX,SAAS,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC;oBAC5B,MAAM;gBACP,CAAC;YACF,CAAC;YAED,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YACxD,MAAM,OAAO,GAAG,SAAS,CAAC,mBAAmB;gBAC5C,CAAC,CAAC,oEAAoE,SAAS,CAAC,mBAAmB,EAAE;gBACrG,CAAC,CAAC,SAAS,CAAC,YAAY;oBACvB,CAAC,CAAC,CAAC,qCAAqC,EAAE,GAAG,SAAS,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;oBAC/E,CAAC,CAAC,gCAAgC,SAAS,CAAC,MAAM,qCAAmC,CAAC;YACxF,MAAM,YAAY,GAAG,SAAS;gBAC7B,CAAC,CAAC,uBAAuB,SAAS,CAAC,IAAI,sBAAsB;gBAC7D,CAAC,CAAC,8EAA4E,CAAC;YAChF,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;gBACd,IAAI,EAAE,SAAS;gBACf,OAAO,EAAE,gBAAgB,UAAU,qBAAqB,OAAO,KAAK,YAAY,EAAE;aAClF,CAAC,CAAC;YAEH,IAAI,CAAC,SAAS;gBAAE,OAAO,SAAS,CAAC,CAAC,uEAAqE;YACvG,OAAO,GAAG;gBACT,KAAK,EAAE,SAAS,CAAC,KAAK;gBACtB,QAAQ,EAAE;oBACT,GAAG,OAAO,CAAC,QAAQ;oBACnB,IAAI,EAAE,SAAS,CAAC,IAAI;oBACpB,YAAY,EAAE,OAAO,CAAC,QAAQ,CAAC,IAAI;oBACnC,UAAU,EAAE,gCAAgC;oBAC5C,OAAO,EAAE;wBACR,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO;wBAC3B,0BAA0B,SAAS,CAAC,MAAM,mBAAmB,SAAS,CAAC,IAAI,EAAE;qBAC7E;oBACD,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,KAAK,CAAC;iBAC7C;aACD,CAAC;QACH,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;CACD","sourcesContent":["/**\n * Local-runtime (Ollama) lifecycle controller.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Owns the cached, per-server\n * {@link OllamaRuntime} instances, the \"confirmed up this session\" flag, and the router's readiness\n * gate for a turn routed to a local (`ollama`) model — including the #31 install-on-consent flow and\n * the #27 graceful tier-escalation fallback. Takes narrow deps (agent dir, a last-assistant-message\n * accessor, the session's UI context/event emitter, and the router's own tier resolver) rather than\n * the whole AgentSession.\n */\n\nimport type { Api, AssistantMessage, Model } from \"@caupulican/pi-ai\";\nimport type { AgentSessionEvent } from \"./agent-session.ts\";\nimport type { RouteDecision } from \"./autonomy/contracts.ts\";\nimport type { ExtensionUIContext } from \"./extensions/index.ts\";\nimport { OLLAMA_PROVIDER } from \"./models/local-registration.ts\";\nimport { type LocalRuntimeDeps, OllamaRuntime } from \"./models/local-runtime.ts\";\n\n/** User-facing router tiers in ascending order — \"learning\" is never selected for a user turn, so\n * it has no place in the escalation ladder (#27's ensureRouteModelReady walks this forward only). */\nconst MODEL_ROUTER_TIER_ORDER: readonly (\"cheap\" | \"medium\" | \"expensive\")[] = [\"cheap\", \"medium\", \"expensive\"];\n\n/** How long the #31 \"install ollama now?\" confirm waits before auto-dismissing (same as a \"No\") —\n * long enough to read and decide, short enough that an unattended session doesn't hang a turn on it. */\nconst OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS = 30_000;\n\nexport interface LocalRuntimeControllerDeps {\n\t/** Root directory OllamaRuntime instances are scoped under — fixed for the session's lifetime. */\n\tagentDir: string;\n\t/** Test-injectable seams for OllamaRuntime's own fetch/spawn/exists calls; unset in production. */\n\tlocalRuntimeDeps?: LocalRuntimeDeps;\n\t/** The session's last assistant message, to detect a just-failed local call and drop a stale\n\t * \"confirmed up\" flag. */\n\tgetLastAssistantMessage(): AssistantMessage | undefined;\n\t/** The session's live interactive UI context, if any — undefined in headless/RPC/print modes. */\n\tgetUIContext(): ExtensionUIContext | undefined;\n\t/** Emits a session event (only ever `warning` / `routing_start` / `routing_end` from this controller). */\n\temit(event: AgentSessionEvent): void;\n\t/** Resolves the model configured for a router tier, respecting configured auth — owned by the\n\t * router itself, not this controller. */\n\tresolveConfiguredTierModel(tier: \"medium\" | \"expensive\"): Model<Api> | undefined;\n\t/** `${provider}/${id}` label for a model, for warning/confirm text. */\n\tformatModel(model: Model<Api>): string;\n}\n\nexport class LocalRuntimeController {\n\t/** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */\n\tprivate readonly _runtimes = new Map<string, OllamaRuntime>();\n\t/** Server URLs confirmed reachable THIS session — skips the health-check round trip on every\n\t * local-routed turn once warm. Keyed the same way as _runtimes. */\n\tprivate readonly _confirmedUp = new Set<string>();\n\n\tprivate readonly deps: LocalRuntimeControllerDeps;\n\n\tconstructor(deps: LocalRuntimeControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\t/**\n\t * Shared {@link OllamaRuntime} for a given server, lazily created and cached by baseUrl so every\n\t * caller — the router's readiness gate below and any host UI's own model-lifecycle commands\n\t * (e.g. `/models`) — sees and can stop the SAME pi-managed process instead of each tracking its\n\t * own untracked child.\n\t */\n\tgetLocalRuntime(baseUrl?: string): OllamaRuntime {\n\t\tconst key = baseUrl ?? \"default\";\n\t\tlet runtime = this._runtimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new OllamaRuntime({ agentDir: this.deps.agentDir, baseUrl, deps: this.deps.localRuntimeDeps });\n\t\t\tthis._runtimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\t/** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's\n\t * own health/boot endpoints are on the Ollama-native server root. */\n\tderiveOllamaServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\t/**\n\t * If the last assistant message in this session was an error from THIS exact local server, a\n\t * cached \"confirmed up\" flag would be stale (the server may have died mid-session) — drop it so\n\t * the next ensure-check is a real one instead of trusting stale state.\n\t */\n\tprivate invalidateIfLastCallFailed(model: Model<Api>, serverUrl: string): void {\n\t\tconst lastAssistant = this.deps.getLastAssistantMessage();\n\t\tif (\n\t\t\tlastAssistant?.stopReason === \"error\" &&\n\t\t\tlastAssistant.provider === OLLAMA_PROVIDER &&\n\t\t\tlastAssistant.model === model.id\n\t\t) {\n\t\t\tthis._confirmedUp.delete(serverUrl);\n\t\t}\n\t}\n\n\t/**\n\t * Ensure a routed model is actually reachable before the turn calls it. No-op (and free) for any\n\t * non-local model — this only ever does network/process work for the `ollama` provider. Caches a\n\t * \"confirmed up this session\" flag per server so a steady-state session pays the health-check\n\t * round trip once, not on every turn; invalidated above when a prior local call actually failed,\n\t * so a server that died mid-session gets re-detected rather than trusted forever. Boots via\n\t * `startReuseExisting()` — never owned storage — so the turn sees the user's OWN pulled models,\n\t * the same server `/models` commands and the user's own `ollama` CLI already talk to. Never\n\t * installs anything itself (installGuide is GUIDE MODE: printed, never executed).\n\t */\n\tasync ensureLocalModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== OLLAMA_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_local\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(serverUrl)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(serverUrl);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.binaryPath) {\n\t\t\treturn { ready: false, reason: \"binary_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.startReuseExisting();\n\t\tif (started.started) {\n\t\t\tthis._confirmedUp.add(serverUrl);\n\t\t}\n\t\treturn { ready: started.started, reason: started.reason };\n\t}\n\n\t/**\n\t * #31: the ONE case a routed local model's unreadiness can be fixed automatically is a missing\n\t * ollama binary — an unreachable server can't be helped by installing, so that reason is left to\n\t * the graceful-fallback warning below unchanged. Only offered when there's an interactive UI to\n\t * ask through: headless/RPC/print sessions have no UI context and fall straight through, same as\n\t * declining or timing out (both resolve confirm() to false). Reverses \"pi never runs installers\n\t * itself\" specifically for this one path — the user is asked first, the download is pi's own\n\t * (never curl|sh), and it lands in pi's own runtimes dir (see OllamaRuntime.installManaged).\n\t *\n\t * Pauses/resumes the routing working-indicator around the confirm dialog itself (re-emitting\n\t * routing_end/routing_start — both already idempotent, see interactive-mode.ts's handlers) so an\n\t * animated spinner doesn't fight a dialog the user is trying to read and answer; the indicator\n\t * comes back for the download/extract that follows a \"yes\", which is genuine processing feedback.\n\t */\n\tprivate async maybeInstallOllamaOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"binary_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Ollama?\",\n\t\t\t\t`Ollama isn't installed, so the local model \"${modelLabel}\" can't run. Pi can download and ` +\n\t\t\t\t\t\"install it now (a large one-time download, possibly over 1 GB depending on your platform) \" +\n\t\t\t\t\t\"into its own runtimes folder — never curl|sh, never touching anything outside pi's own \" +\n\t\t\t\t\t\"directory. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"ollama-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"ollama-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\treturn this.ensureLocalModelReady(model);\n\t}\n\n\t/**\n\t * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct\n\t * route — both carry tier \"cheap\") must not dead-end the turn just because ollama isn't up.\n\t * Never a SILENT swap: every fallback is announced in a warning that states (i) the local model\n\t * was unavailable and WHY — binary missing surfaces the install guide inline; any other reason\n\t * gets a \"check that ollama is running\" hint — and (ii) which tier is now handling the turn, so\n\t * the cost shift is never a surprise. Escalates cheap -> medium -> expensive, skipping any\n\t * unconfigured intermediate tier, reusing the router's own existing \"model unavailable\"\n\t * resolution (resolveConfiguredTierModel) rather than inventing a new fallback mechanism.\n\t * Escalation is bounded: tier strictly increases each hop, so it terminates within two hops.\n\t *\n\t * Before the warning/escalation below: #31's consent gate gets one shot at fixing a missing\n\t * binary interactively (see maybeInstallOllamaOnConsent) — declining, timing out, running\n\t * headless, or the install attempt itself failing all fall through here unchanged, just with an\n\t * honest reason (an install that failed is worded as a failed install, not re-labeled as if\n\t * nothing was ever tried).\n\t */\n\tasync ensureRouteModelReady(\n\t\tresolved: { decision: RouteDecision; model: Model<Api> } | undefined,\n\t): Promise<{ decision: RouteDecision; model: Model<Api> } | undefined> {\n\t\tlet current = resolved;\n\t\twhile (current && current.model.provider === OLLAMA_PROVIDER) {\n\t\t\tlet readiness: { ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string } =\n\t\t\t\tawait this.ensureLocalModelReady(current.model);\n\t\t\tif (!readiness.ready) {\n\t\t\t\treadiness = await this.maybeInstallOllamaOnConsent(current.model, readiness);\n\t\t\t}\n\t\t\tif (readiness.ready) return current;\n\n\t\t\t// Walk the remaining tiers in order (never back down to cheap) and take the first one that\n\t\t\t// actually resolves — an unconfigured intermediate tier (e.g. no mediumModel set) must be\n\t\t\t// skipped, not treated as \"no fallback available\".\n\t\t\tconst startIndex = MODEL_ROUTER_TIER_ORDER.indexOf(current.decision.tier as \"cheap\" | \"medium\" | \"expensive\");\n\t\t\tlet escalated: { tier: \"medium\" | \"expensive\"; model: Model<Api> } | undefined;\n\t\t\tfor (let i = startIndex + 1; startIndex !== -1 && i < MODEL_ROUTER_TIER_ORDER.length; i++) {\n\t\t\t\tconst tier = MODEL_ROUTER_TIER_ORDER[i] as \"medium\" | \"expensive\";\n\t\t\t\tconst model = this.deps.resolveConfiguredTierModel(tier);\n\t\t\t\tif (model) {\n\t\t\t\t\tescalated = { tier, model };\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tconst modelLabel = this.deps.formatModel(current.model);\n\t\t\tconst whyText = readiness.installAttemptError\n\t\t\t\t? `pi tried to install it just now, but the install attempt failed: ${readiness.installAttemptError}`\n\t\t\t\t: readiness.installGuide\n\t\t\t\t\t? [\"the ollama binary is not installed.\", ...readiness.installGuide].join(\"\\n\")\n\t\t\t\t\t: `its server is not reachable (${readiness.reason}) — check that ollama is running.`;\n\t\t\tconst fallbackText = escalated\n\t\t\t\t? `Falling back to the ${escalated.tier} tier for this turn.`\n\t\t\t\t: \"No other tier is configured — falling back to the session's default model.\";\n\t\t\tthis.deps.emit({\n\t\t\t\ttype: \"warning\",\n\t\t\t\tmessage: `Local model \"${modelLabel}\" is unavailable: ${whyText}\\n${fallbackText}`,\n\t\t\t});\n\n\t\t\tif (!escalated) return undefined; // no higher tier resolves — caller falls back to the session default\n\t\t\tcurrent = {\n\t\t\t\tmodel: escalated.model,\n\t\t\t\tdecision: {\n\t\t\t\t\t...current.decision,\n\t\t\t\t\ttier: escalated.tier,\n\t\t\t\t\tfallbackFrom: current.decision.tier,\n\t\t\t\t\treasonCode: \"local_model_not_ready_fallback\",\n\t\t\t\t\treasons: [\n\t\t\t\t\t\t...current.decision.reasons,\n\t\t\t\t\t\t`Local model not ready (${readiness.reason}); escalated to ${escalated.tier}`,\n\t\t\t\t\t],\n\t\t\t\t\tmodel: this.deps.formatModel(escalated.model),\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\treturn current;\n\t}\n}\n"]}
+{"version":3,"file":"local-runtime-controller.js","sourceRoot":"","sources":["../../src/core/local-runtime-controller.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAMH,OAAO,EAAE,wBAAwB,EAAE,eAAe,EAAE,MAAM,gCAAgC,CAAC;AAC3F,OAAO,EAEN,aAAa,EACb,0BAA0B,EAC1B,mBAAmB,GACnB,MAAM,2BAA2B,CAAC;AAEnC;qGACqG;AACrG,MAAM,uBAAuB,GAAkD,CAAC,OAAO,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC;AAEhH;wGACwG;AACxG,MAAM,iCAAiC,GAAG,MAAM,CAAC;AAqBjD,MAAM,OAAO,sBAAsB;IAClC,qGAAqG;IACpF,SAAS,GAAG,IAAI,GAAG,EAAyB,CAAC;IAC9D,uGAAuG;IACtF,qBAAqB,GAAG,IAAI,GAAG,EAA+B,CAAC;IAChF;uEACmE;IAClD,YAAY,GAAG,IAAI,GAAG,EAAU,CAAC;IAEjC,IAAI,CAA6B;IAElD,YAAY,IAAgC,EAAE;QAC7C,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IAAA,CACjB;IAED;;;;;OAKG;IACH,eAAe,CAAC,OAAgB,EAAiB;QAChD,MAAM,GAAG,GAAG,OAAO,IAAI,SAAS,CAAC;QACjC,IAAI,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACtC,IAAI,CAAC,OAAO,EAAE,CAAC;YACd,OAAO,GAAG,IAAI,aAAa,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC,CAAC;YACzG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QAClC,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED,sBAAsB,CAAC,OAAe,EAAE,OAAgB,EAAuB;QAC9E,MAAM,eAAe,GAAG,OAAO,EAAE,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,0BAA0B,CAAC,OAAO,CAAC,CAAC;QAC3F,MAAM,GAAG,GAAG,GAAG,OAAO,KAAK,eAAe,EAAE,CAAC;QAC7C,IAAI,OAAO,GAAG,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAClD,IAAI,CAAC,OAAO,EAAE,CAAC;YACd,OAAO,GAAG,IAAI,mBAAmB,CAAC;gBACjC,QAAQ,EAAE,IAAI,CAAC,IAAI,CAAC,QAAQ;gBAC5B,OAAO;gBACP,OAAO,EAAE,eAAe;gBACxB,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,gBAAgB;aAChC,CAAC,CAAC;YACH,IAAI,CAAC,qBAAqB,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,CAAC;QAC9C,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;IAED;yEACqE;IACrE,qBAAqB,CAAC,YAAoB,EAAU;QACnD,OAAO,YAAY,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,2BAA2B,CAAC,YAAoB,EAAU;QACjE,OAAO,YAAY,CAAC,OAAO,CAAC,UAAU,EAAE,EAAE,CAAC,CAAC;IAAA,CAC5C;IAEO,sBAAsB,CAAC,QAAgB,EAAW;QACzD,OAAO,QAAQ,KAAK,eAAe,IAAI,QAAQ,KAAK,wBAAwB,CAAC;IAAA,CAC7E;IAED;;;;OAIG;IACK,eAAe,CAAC,KAAiB,EAAE,SAAiB,EAAU;QACrE,OAAO,KAAK,CAAC,QAAQ,KAAK,wBAAwB,CAAC,CAAC,CAAC,GAAG,SAAS,KAAK,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IAAA,CAC7F;IAEO,0BAA0B,CAAC,KAAiB,EAAE,SAAiB,EAAQ;QAC9E,MAAM,aAAa,GAAG,IAAI,CAAC,IAAI,CAAC,uBAAuB,EAAE,CAAC;QAC1D,IACC,aAAa,EAAE,UAAU,KAAK,OAAO;YACrC,aAAa,CAAC,QAAQ,KAAK,KAAK,CAAC,QAAQ;YACzC,aAAa,CAAC,KAAK,KAAK,KAAK,CAAC,EAAE,EAC/B,CAAC;YACF,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,IAAI,CAAC,eAAe,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC,CAAC;QAClE,CAAC;IAAA,CACD;IAED;;;;;OAKG;IACH,KAAK,CAAC,qBAAqB,CAC1B,KAAiB,EACsD;QACvE,IAAI,KAAK,CAAC,QAAQ,KAAK,eAAe,EAAE,CAAC;YACxC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC;QAC7C,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC5D,MAAM,YAAY,GAAG,IAAI,CAAC,eAAe,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;QAC5D,IAAI,CAAC,0BAA0B,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;QAClD,IAAI,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,EAAE,CAAC;YACzC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,qBAAqB,EAAE,CAAC;QACvD,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE,CAAC;QACtC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YACpC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,iBAAiB,EAAE,CAAC;QACnD,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,CAAC;YACxB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,EAAE,YAAY,EAAE,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC;QACzF,CAAC;QACD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,kBAAkB,EAAE,CAAC;QACnD,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;YACrB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;QACrC,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;IAAA,CAC1D;IAED,KAAK,CAAC,4BAA4B,CACjC,KAAiB,EACsD;QACvE,IAAI,KAAK,CAAC,QAAQ,KAAK,wBAAwB,EAAE,CAAC;YACjD,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,kBAAkB,EAAE,CAAC;QACpD,CAAC;QACD,MAAM,SAAS,GAAG,IAAI,CAAC,2BAA2B,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAClE,MAAM,YAAY,GAAG,IAAI,CAAC,eAAe,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;QAC5D,IAAI,CAAC,0BAA0B,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;QAClD,IAAI,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,EAAE,CAAC;YACzC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,qBAAqB,EAAE,CAAC;QACvD,CAAC;QACD,MAAM,OAAO,GAAG,IAAI,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,EAAE,SAAS,CAAC,CAAC;QACjE,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,MAAM,EAAE,CAAC;QACtC,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrB,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YACpC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,iBAAiB,EAAE,CAAC;QACnD,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,gBAAgB,EAAE,CAAC;YAC9B,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,iBAAiB,EAAE,YAAY,EAAE,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC;QAC1F,CAAC;QACD,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;QACtC,IAAI,OAAO,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,KAAK,iBAAiB,EAAE,CAAC;YAC7D,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC;YACpC,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;QAChD,CAAC;QACD,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;IAAA,CAChD;IAED;;;;;;;;;;;;;OAaG;IACK,KAAK,CAAC,2BAA2B,CACxC,KAAiB,EACjB,SAAsE,EAC+B;QACrG,MAAM,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;QACpC,IAAI,CAAC,EAAE,IAAI,SAAS,CAAC,KAAK,IAAI,SAAS,CAAC,MAAM,KAAK,gBAAgB;YAAE,OAAO,SAAS,CAAC;QAEtF,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAChD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC,CAAC;QACxC,IAAI,SAAkB,CAAC;QACvB,IAAI,CAAC;YACJ,SAAS,GAAG,MAAM,EAAE,CAAC,OAAO,CAC3B,iBAAiB,EACjB,+CAA+C,UAAU,mCAAmC;gBAC3F,4FAA4F;gBAC5F,2FAAyF;gBACzF,4BAA4B,EAC7B,EAAE,OAAO,EAAE,iCAAiC,EAAE,CAC9C,CAAC;QACH,CAAC;gBAAS,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,SAAS;YAAE,OAAO,SAAS,CAAC;QAEjC,MAAM,SAAS,GAAG,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAC5D,MAAM,OAAO,GAAG,IAAI,CAAC,eAAe,CAAC,SAAS,CAAC,CAAC;QAChD,IAAI,aAA8C,CAAC;QACnD,IAAI,CAAC;YACJ,aAAa,GAAG,MAAM,OAAO,CAAC,cAAc,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,SAAS,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;QAClG,CAAC;gBAAS,CAAC;YACV,EAAE,CAAC,SAAS,CAAC,gBAAgB,EAAE,SAAS,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,aAAa,CAAC,EAAE,EAAE,CAAC;YACvB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,aAAa,CAAC,KAAK,EAAE,CAAC;QAC7F,CAAC;QACD,OAAO,IAAI,CAAC,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAAA,CACzC;IAEO,KAAK,CAAC,iCAAiC,CAC9C,KAAiB,EACjB,SAAsE,EAC+B;QACrG,MAAM,EAAE,GAAG,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;QACpC,IAAI,CAAC,EAAE,IAAI,SAAS,CAAC,KAAK,IAAI,SAAS,CAAC,MAAM,KAAK,iBAAiB;YAAE,OAAO,SAAS,CAAC;QAEvF,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC;QAChD,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,aAAa,EAAE,CAAC,CAAC;QACxC,IAAI,SAAkB,CAAC;QACvB,IAAI,CAAC;YACJ,SAAS,GAAG,MAAM,EAAE,CAAC,OAAO,CAC3B,+BAA+B,EAC/B,2BAA2B,UAAU,qDAAqD;gBACzF,0FAA0F;gBAC1F,0FAA0F;gBAC1F,uEAAuE,EACxE,EAAE,OAAO,EAAE,iCAAiC,EAAE,CAC9C,CAAC;QACH,CAAC;gBAAS,CAAC;YACV,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC,CAAC;QAC3C,CAAC;QACD,IAAI,CAAC,SAAS;YAAE,OAAO,SAAS,CAAC;QAEjC,MAAM,SAAS,GAAG,IAAI,CAAC,2BAA2B,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAClE,MAAM,OAAO,GAAG,IAAI,CAAC,sBAAsB,CAAC,KAAK,CAAC,EAAE,EAAE,SAAS,CAAC,CAAC;QACjE,IAAI,aAA8C,CAAC;QACnD,IAAI,CAAC;YACJ,aAAa,GAAG,MAAM,OAAO,CAAC,cAAc,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,SAAS,CAAC,sBAAsB,EAAE,MAAM,CAAC,CAAC,CAAC;QACxG,CAAC;gBAAS,CAAC;YACV,EAAE,CAAC,SAAS,CAAC,sBAAsB,EAAE,SAAS,CAAC,CAAC;QACjD,CAAC;QACD,IAAI,CAAC,aAAa,CAAC,EAAE,EAAE,CAAC;YACvB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,gBAAgB,EAAE,mBAAmB,EAAE,aAAa,CAAC,KAAK,EAAE,CAAC;QAC7F,CAAC;QACD,IAAI,cAA+C,CAAC;QACpD,IAAI,CAAC;YACJ,cAAc,GAAG,MAAM,OAAO,CAAC,aAAa,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,SAAS,CAAC,uBAAuB,EAAE,MAAM,CAAC,CAAC,CAAC;QACzG,CAAC;gBAAS,CAAC;YACV,EAAE,CAAC,SAAS,CAAC,uBAAuB,EAAE,SAAS,CAAC,CAAC;QAClD,CAAC;QACD,IAAI,CAAC,cAAc,CAAC,EAAE,EAAE,CAAC;YACxB,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,iBAAiB,EAAE,mBAAmB,EAAE,cAAc,CAAC,KAAK,EAAE,CAAC;QAC/F,CAAC;QACD,OAAO,IAAI,CAAC,4BAA4B,CAAC,KAAK,CAAC,CAAC;IAAA,CAChD;IAED;;;;;;;;;;;;;;;;OAgBG;IACH,KAAK,CAAC,qBAAqB,CAC1B,QAAoE,EACE;QACtE,IAAI,OAAO,GAAG,QAAQ,CAAC;QACvB,OAAO,OAAO,IAAI,IAAI,CAAC,sBAAsB,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,EAAE,CAAC;YACvE,IAAI,SAAS,GACZ,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe;gBACzC,CAAC,CAAC,MAAM,IAAI,CAAC,qBAAqB,CAAC,OAAO,CAAC,KAAK,CAAC;gBACjD,CAAC,CAAC,MAAM,IAAI,CAAC,4BAA4B,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YAC3D,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;gBACtB,SAAS;oBACR,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe;wBACzC,CAAC,CAAC,MAAM,IAAI,CAAC,2BAA2B,CAAC,OAAO,CAAC,KAAK,EAAE,SAAS,CAAC;wBAClE,CAAC,CAAC,MAAM,IAAI,CAAC,iCAAiC,CAAC,OAAO,CAAC,KAAK,EAAE,SAAS,CAAC,CAAC;YAC5E,CAAC;YACD,IAAI,SAAS,CAAC,KAAK;gBAAE,OAAO,OAAO,CAAC;YAEpC,2FAA2F;YAC3F,4FAA0F;YAC1F,mDAAmD;YACnD,MAAM,UAAU,GAAG,uBAAuB,CAAC,OAAO,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAwC,CAAC,CAAC;YAC9G,IAAI,SAA0E,CAAC;YAC/E,KAAK,IAAI,CAAC,GAAG,UAAU,GAAG,CAAC,EAAE,UAAU,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,uBAAuB,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC3F,MAAM,IAAI,GAAG,uBAAuB,CAAC,CAAC,CAA2B,CAAC;gBAClE,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,0BAA0B,CAAC,IAAI,CAAC,CAAC;gBACzD,IAAI,KAAK,EAAE,CAAC;oBACX,SAAS,GAAG,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC;oBAC5B,MAAM;gBACP,CAAC;YACF,CAAC;YAED,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;YACxD,MAAM,gBAAgB,GAAG,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,cAAc,CAAC;YAChG,MAAM,OAAO,GAAG,SAAS,CAAC,mBAAmB;gBAC5C,CAAC,CAAC,oEAAoE,SAAS,CAAC,mBAAmB,EAAE;gBACrG,CAAC,CAAC,SAAS,CAAC,YAAY;oBACvB,CAAC,CAAC;wBACA,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe;4BACzC,CAAC,CAAC,qCAAqC;4BACvC,CAAC,CAAC,uDAAuD;wBAC1D,GAAG,SAAS,CAAC,YAAY;qBACzB,CAAC,IAAI,CAAC,IAAI,CAAC;oBACb,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,QAAQ,KAAK,eAAe;wBAC3C,CAAC,CAAC,OAAO,gBAAgB,6BAA6B,SAAS,CAAC,MAAM,qCAAmC;wBACzG,CAAC,CAAC,OAAO,gBAAgB,6BAA6B,SAAS,CAAC,MAAM,0CAAwC,CAAC;YAClH,MAAM,YAAY,GAAG,SAAS;gBAC7B,CAAC,CAAC,uBAAuB,SAAS,CAAC,IAAI,sBAAsB;gBAC7D,CAAC,CAAC,8EAA4E,CAAC;YAChF,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC;gBACd,IAAI,EAAE,SAAS;gBACf,OAAO,EAAE,gBAAgB,UAAU,qBAAqB,OAAO,KAAK,YAAY,EAAE;aAClF,CAAC,CAAC;YAEH,IAAI,CAAC,SAAS;gBAAE,OAAO,SAAS,CAAC,CAAC,uEAAqE;YACvG,OAAO,GAAG;gBACT,KAAK,EAAE,SAAS,CAAC,KAAK;gBACtB,QAAQ,EAAE;oBACT,GAAG,OAAO,CAAC,QAAQ;oBACnB,IAAI,EAAE,SAAS,CAAC,IAAI;oBACpB,YAAY,EAAE,OAAO,CAAC,QAAQ,CAAC,IAAI;oBACnC,UAAU,EAAE,gCAAgC;oBAC5C,OAAO,EAAE;wBACR,GAAG,OAAO,CAAC,QAAQ,CAAC,OAAO;wBAC3B,0BAA0B,SAAS,CAAC,MAAM,mBAAmB,SAAS,CAAC,IAAI,EAAE;qBAC7E;oBACD,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,SAAS,CAAC,KAAK,CAAC;iBAC7C;aACD,CAAC;QACH,CAAC;QACD,OAAO,OAAO,CAAC;IAAA,CACf;CACD","sourcesContent":["/**\n * Local-runtime (Ollama) lifecycle controller.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Owns the cached, per-server\n * {@link OllamaRuntime} instances, the \"confirmed up this session\" flag, and the router's readiness\n * gate for a turn routed to a local (`ollama`) model — including the #31 install-on-consent flow and\n * the #27 graceful tier-escalation fallback. Takes narrow deps (agent dir, a last-assistant-message\n * accessor, the session's UI context/event emitter, and the router's own tier resolver) rather than\n * the whole AgentSession.\n */\n\nimport type { Api, AssistantMessage, Model } from \"@caupulican/pi-ai\";\nimport type { AgentSessionEvent } from \"./agent-session.ts\";\nimport type { RouteDecision } from \"./autonomy/contracts.ts\";\nimport type { ExtensionUIContext } from \"./extensions/index.ts\";\nimport { HF_TRANSFORMERS_PROVIDER, OLLAMA_PROVIDER } from \"./models/local-registration.ts\";\nimport {\n\ttype LocalRuntimeDeps,\n\tOllamaRuntime,\n\tresolveTransformersBaseUrl,\n\tTransformersRuntime,\n} from \"./models/local-runtime.ts\";\n\n/** User-facing router tiers in ascending order — \"learning\" is never selected for a user turn, so\n * it has no place in the escalation ladder (#27's ensureRouteModelReady walks this forward only). */\nconst MODEL_ROUTER_TIER_ORDER: readonly (\"cheap\" | \"medium\" | \"expensive\")[] = [\"cheap\", \"medium\", \"expensive\"];\n\n/** How long the #31 \"install ollama now?\" confirm waits before auto-dismissing (same as a \"No\") —\n * long enough to read and decide, short enough that an unattended session doesn't hang a turn on it. */\nconst OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS = 30_000;\n\nexport interface LocalRuntimeControllerDeps {\n\t/** Root directory OllamaRuntime instances are scoped under — fixed for the session's lifetime. */\n\tagentDir: string;\n\t/** Test-injectable seams for OllamaRuntime's own fetch/spawn/exists calls; unset in production. */\n\tlocalRuntimeDeps?: LocalRuntimeDeps;\n\t/** The session's last assistant message, to detect a just-failed local call and drop a stale\n\t * \"confirmed up\" flag. */\n\tgetLastAssistantMessage(): AssistantMessage | undefined;\n\t/** The session's live interactive UI context, if any — undefined in headless/RPC/print modes. */\n\tgetUIContext(): ExtensionUIContext | undefined;\n\t/** Emits a session event (only ever `warning` / `routing_start` / `routing_end` from this controller). */\n\temit(event: AgentSessionEvent): void;\n\t/** Resolves the model configured for a router tier, respecting configured auth — owned by the\n\t * router itself, not this controller. */\n\tresolveConfiguredTierModel(tier: \"medium\" | \"expensive\"): Model<Api> | undefined;\n\t/** `${provider}/${id}` label for a model, for warning/confirm text. */\n\tformatModel(model: Model<Api>): string;\n}\n\nexport class LocalRuntimeController {\n\t/** Lazy, cached by baseUrl so the router path and any other caller share one instance per server. */\n\tprivate readonly _runtimes = new Map<string, OllamaRuntime>();\n\t/** Lazy, cached by model+baseUrl so the router and `/models` share one sidecar handle per HF model. */\n\tprivate readonly _transformersRuntimes = new Map<string, TransformersRuntime>();\n\t/** Server URLs confirmed reachable THIS session — skips the health-check round trip on every\n\t * local-routed turn once warm. Keyed the same way as _runtimes. */\n\tprivate readonly _confirmedUp = new Set<string>();\n\n\tprivate readonly deps: LocalRuntimeControllerDeps;\n\n\tconstructor(deps: LocalRuntimeControllerDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\t/**\n\t * Shared {@link OllamaRuntime} for a given server, lazily created and cached by baseUrl so every\n\t * caller — the router's readiness gate below and any host UI's own model-lifecycle commands\n\t * (e.g. `/models`) — sees and can stop the SAME pi-managed process instead of each tracking its\n\t * own untracked child.\n\t */\n\tgetLocalRuntime(baseUrl?: string): OllamaRuntime {\n\t\tconst key = baseUrl ?? \"default\";\n\t\tlet runtime = this._runtimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new OllamaRuntime({ agentDir: this.deps.agentDir, baseUrl, deps: this.deps.localRuntimeDeps });\n\t\t\tthis._runtimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\tgetTransformersRuntime(modelId: string, baseUrl?: string): TransformersRuntime {\n\t\tconst resolvedBaseUrl = baseUrl?.replace(/\\/$/, \"\") ?? resolveTransformersBaseUrl(modelId);\n\t\tconst key = `${modelId}\\0${resolvedBaseUrl}`;\n\t\tlet runtime = this._transformersRuntimes.get(key);\n\t\tif (!runtime) {\n\t\t\truntime = new TransformersRuntime({\n\t\t\t\tagentDir: this.deps.agentDir,\n\t\t\t\tmodelId,\n\t\t\t\tbaseUrl: resolvedBaseUrl,\n\t\t\t\tdeps: this.deps.localRuntimeDeps,\n\t\t\t});\n\t\t\tthis._transformersRuntimes.set(key, runtime);\n\t\t}\n\t\treturn runtime;\n\t}\n\n\t/** models.json registers a local model's baseUrl as `<server>/v1` (OpenAI-compat); the runtime's\n\t * own health/boot endpoints are on the Ollama-native server root. */\n\tderiveOllamaServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\tprivate deriveOpenAICompatServerUrl(modelBaseUrl: string): string {\n\t\treturn modelBaseUrl.replace(/\\/v1\\/?$/, \"\");\n\t}\n\n\tprivate isManagedLocalProvider(provider: string): boolean {\n\t\treturn provider === OLLAMA_PROVIDER || provider === HF_TRANSFORMERS_PROVIDER;\n\t}\n\n\t/**\n\t * If the last assistant message in this session was an error from THIS exact local server, a\n\t * cached \"confirmed up\" flag would be stale (the server may have died mid-session) — drop it so\n\t * the next ensure-check is a real one instead of trusting stale state.\n\t */\n\tprivate confirmationKey(model: Model<Api>, serverUrl: string): string {\n\t\treturn model.provider === HF_TRANSFORMERS_PROVIDER ? `${serverUrl}\\0${model.id}` : serverUrl;\n\t}\n\n\tprivate invalidateIfLastCallFailed(model: Model<Api>, serverUrl: string): void {\n\t\tconst lastAssistant = this.deps.getLastAssistantMessage();\n\t\tif (\n\t\t\tlastAssistant?.stopReason === \"error\" &&\n\t\t\tlastAssistant.provider === model.provider &&\n\t\t\tlastAssistant.model === model.id\n\t\t) {\n\t\t\tthis._confirmedUp.delete(this.confirmationKey(model, serverUrl));\n\t\t}\n\t}\n\n\t/**\n\t * Ensure a routed managed-local model is actually reachable before the turn calls it. No-op (and\n\t * free) for non-local/API models. Caches a \"confirmed up this session\" flag per server (and per\n\t * Transformers model) so steady-state routing pays the health-check round trip once; invalidated\n\t * above when a prior local call failed so a dead sidecar gets re-detected instead of trusted.\n\t */\n\tasync ensureLocalModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== OLLAMA_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_local\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst confirmedKey = this.confirmationKey(model, serverUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(confirmedKey)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.binaryPath) {\n\t\t\treturn { ready: false, reason: \"binary_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.startReuseExisting();\n\t\tif (started.started) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t}\n\t\treturn { ready: started.started, reason: started.reason };\n\t}\n\n\tasync ensureTransformersModelReady(\n\t\tmodel: Model<Api>,\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[] }> {\n\t\tif (model.provider !== HF_TRANSFORMERS_PROVIDER) {\n\t\t\treturn { ready: true, reason: \"not_transformers\" };\n\t\t}\n\t\tconst serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);\n\t\tconst confirmedKey = this.confirmationKey(model, serverUrl);\n\t\tthis.invalidateIfLastCallFailed(model, serverUrl);\n\t\tif (this._confirmedUp.has(confirmedKey)) {\n\t\t\treturn { ready: true, reason: \"confirmed_up_cached\" };\n\t\t}\n\t\tconst runtime = this.getTransformersRuntime(model.id, serverUrl);\n\t\tconst status = await runtime.detect();\n\t\tif (status.serverUp) {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: \"already_running\" };\n\t\t}\n\t\tif (!status.runtimeInstalled) {\n\t\t\treturn { ready: false, reason: \"runtime_missing\", installGuide: runtime.installGuide() };\n\t\t}\n\t\tconst started = await runtime.start();\n\t\tif (started.started || started.reason === \"already_running\") {\n\t\t\tthis._confirmedUp.add(confirmedKey);\n\t\t\treturn { ready: true, reason: started.reason };\n\t\t}\n\t\treturn { ready: false, reason: started.reason };\n\t}\n\n\t/**\n\t * #31: the ONE case a routed local model's unreadiness can be fixed automatically is a missing\n\t * ollama binary — an unreachable server can't be helped by installing, so that reason is left to\n\t * the graceful-fallback warning below unchanged. Only offered when there's an interactive UI to\n\t * ask through: headless/RPC/print sessions have no UI context and fall straight through, same as\n\t * declining or timing out (both resolve confirm() to false). Reverses \"pi never runs installers\n\t * itself\" specifically for this one path — the user is asked first, the download is pi's own\n\t * (never curl|sh), and it lands in pi's own runtimes dir (see OllamaRuntime.installManaged).\n\t *\n\t * Pauses/resumes the routing working-indicator around the confirm dialog itself (re-emitting\n\t * routing_end/routing_start — both already idempotent, see interactive-mode.ts's handlers) so an\n\t * animated spinner doesn't fight a dialog the user is trying to read and answer; the indicator\n\t * comes back for the download/extract that follows a \"yes\", which is genuine processing feedback.\n\t */\n\tprivate async maybeInstallOllamaOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"binary_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Ollama?\",\n\t\t\t\t`Ollama isn't installed, so the local model \"${modelLabel}\" can't run. Pi can download and ` +\n\t\t\t\t\t\"install it now (a large one-time download, possibly over 1 GB depending on your platform) \" +\n\t\t\t\t\t\"into its own runtimes folder — never curl|sh, never touching anything outside pi's own \" +\n\t\t\t\t\t\"directory. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOllamaServerUrl(model.baseUrl);\n\t\tconst runtime = this.getLocalRuntime(serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"ollama-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"ollama-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\treturn this.ensureLocalModelReady(model);\n\t}\n\n\tprivate async maybeInstallTransformersOnConsent(\n\t\tmodel: Model<Api>,\n\t\treadiness: { ready: boolean; reason: string; installGuide?: string[] },\n\t): Promise<{ ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string }> {\n\t\tconst ui = this.deps.getUIContext();\n\t\tif (!ui || readiness.ready || readiness.reason !== \"runtime_missing\") return readiness;\n\n\t\tconst modelLabel = this.deps.formatModel(model);\n\t\tthis.deps.emit({ type: \"routing_end\" });\n\t\tlet confirmed: boolean;\n\t\ttry {\n\t\t\tconfirmed = await ui.confirm(\n\t\t\t\t\"Install Transformers runtime?\",\n\t\t\t\t`The Hugging Face model \"${modelLabel}\" needs a pi-managed Python venv with Transformers ` +\n\t\t\t\t\t\"and CPU PyTorch before it can run. Pi will install those packages into its own runtimes \" +\n\t\t\t\t\t\"folder, download the model into a pi-owned Hugging Face cache, and leave system Python, \" +\n\t\t\t\t\t\"your Ollama models, and your user HF cache untouched. Install it now?\",\n\t\t\t\t{ timeout: OLLAMA_INSTALL_CONFIRM_TIMEOUT_MS },\n\t\t\t);\n\t\t} finally {\n\t\t\tthis.deps.emit({ type: \"routing_start\" });\n\t\t}\n\t\tif (!confirmed) return readiness;\n\n\t\tconst serverUrl = this.deriveOpenAICompatServerUrl(model.baseUrl);\n\t\tconst runtime = this.getTransformersRuntime(model.id, serverUrl);\n\t\tlet installResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tinstallResult = await runtime.installManaged((status) => ui.setStatus(\"transformers-install\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"transformers-install\", undefined);\n\t\t}\n\t\tif (!installResult.ok) {\n\t\t\treturn { ready: false, reason: \"install_failed\", installAttemptError: installResult.error };\n\t\t}\n\t\tlet downloadResult: { ok: boolean; error?: string };\n\t\ttry {\n\t\t\tdownloadResult = await runtime.downloadModel((status) => ui.setStatus(\"transformers-download\", status));\n\t\t} finally {\n\t\t\tui.setStatus(\"transformers-download\", undefined);\n\t\t}\n\t\tif (!downloadResult.ok) {\n\t\t\treturn { ready: false, reason: \"download_failed\", installAttemptError: downloadResult.error };\n\t\t}\n\t\treturn this.ensureTransformersModelReady(model);\n\t}\n\n\t/**\n\t * Router-swap gate (#27): a turn routed to a local model (any tier, including an executor-direct\n\t * route — both carry tier \"cheap\") must not dead-end the turn just because ollama isn't up.\n\t * Never a SILENT swap: every fallback is announced in a warning that states (i) the local model\n\t * was unavailable and WHY — binary missing surfaces the install guide inline; any other reason\n\t * gets a \"check that ollama is running\" hint — and (ii) which tier is now handling the turn, so\n\t * the cost shift is never a surprise. Escalates cheap -> medium -> expensive, skipping any\n\t * unconfigured intermediate tier, reusing the router's own existing \"model unavailable\"\n\t * resolution (resolveConfiguredTierModel) rather than inventing a new fallback mechanism.\n\t * Escalation is bounded: tier strictly increases each hop, so it terminates within two hops.\n\t *\n\t * Before the warning/escalation below: #31's consent gate gets one shot at fixing a missing\n\t * binary interactively (see maybeInstallOllamaOnConsent) — declining, timing out, running\n\t * headless, or the install attempt itself failing all fall through here unchanged, just with an\n\t * honest reason (an install that failed is worded as a failed install, not re-labeled as if\n\t * nothing was ever tried).\n\t */\n\tasync ensureRouteModelReady(\n\t\tresolved: { decision: RouteDecision; model: Model<Api> } | undefined,\n\t): Promise<{ decision: RouteDecision; model: Model<Api> } | undefined> {\n\t\tlet current = resolved;\n\t\twhile (current && this.isManagedLocalProvider(current.model.provider)) {\n\t\t\tlet readiness: { ready: boolean; reason: string; installGuide?: string[]; installAttemptError?: string } =\n\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t? await this.ensureLocalModelReady(current.model)\n\t\t\t\t\t: await this.ensureTransformersModelReady(current.model);\n\t\t\tif (!readiness.ready) {\n\t\t\t\treadiness =\n\t\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t? await this.maybeInstallOllamaOnConsent(current.model, readiness)\n\t\t\t\t\t\t: await this.maybeInstallTransformersOnConsent(current.model, readiness);\n\t\t\t}\n\t\t\tif (readiness.ready) return current;\n\n\t\t\t// Walk the remaining tiers in order (never back down to cheap) and take the first one that\n\t\t\t// actually resolves — an unconfigured intermediate tier (e.g. no mediumModel set) must be\n\t\t\t// skipped, not treated as \"no fallback available\".\n\t\t\tconst startIndex = MODEL_ROUTER_TIER_ORDER.indexOf(current.decision.tier as \"cheap\" | \"medium\" | \"expensive\");\n\t\t\tlet escalated: { tier: \"medium\" | \"expensive\"; model: Model<Api> } | undefined;\n\t\t\tfor (let i = startIndex + 1; startIndex !== -1 && i < MODEL_ROUTER_TIER_ORDER.length; i++) {\n\t\t\t\tconst tier = MODEL_ROUTER_TIER_ORDER[i] as \"medium\" | \"expensive\";\n\t\t\t\tconst model = this.deps.resolveConfiguredTierModel(tier);\n\t\t\t\tif (model) {\n\t\t\t\t\tescalated = { tier, model };\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tconst modelLabel = this.deps.formatModel(current.model);\n\t\t\tconst localRuntimeName = current.model.provider === OLLAMA_PROVIDER ? \"ollama\" : \"Transformers\";\n\t\t\tconst whyText = readiness.installAttemptError\n\t\t\t\t? `pi tried to install it just now, but the install attempt failed: ${readiness.installAttemptError}`\n\t\t\t\t: readiness.installGuide\n\t\t\t\t\t? [\n\t\t\t\t\t\t\tcurrent.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t\t\t? \"the ollama binary is not installed.\"\n\t\t\t\t\t\t\t\t: \"the pi-managed Transformers runtime is not installed.\",\n\t\t\t\t\t\t\t...readiness.installGuide,\n\t\t\t\t\t\t].join(\"\\n\")\n\t\t\t\t\t: current.model.provider === OLLAMA_PROVIDER\n\t\t\t\t\t\t? `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that ollama is running.`\n\t\t\t\t\t\t: `its ${localRuntimeName} server is not reachable (${readiness.reason}) — check that the runtime is running.`;\n\t\t\tconst fallbackText = escalated\n\t\t\t\t? `Falling back to the ${escalated.tier} tier for this turn.`\n\t\t\t\t: \"No other tier is configured — falling back to the session's default model.\";\n\t\t\tthis.deps.emit({\n\t\t\t\ttype: \"warning\",\n\t\t\t\tmessage: `Local model \"${modelLabel}\" is unavailable: ${whyText}\\n${fallbackText}`,\n\t\t\t});\n\n\t\t\tif (!escalated) return undefined; // no higher tier resolves — caller falls back to the session default\n\t\t\tcurrent = {\n\t\t\t\tmodel: escalated.model,\n\t\t\t\tdecision: {\n\t\t\t\t\t...current.decision,\n\t\t\t\t\ttier: escalated.tier,\n\t\t\t\t\tfallbackFrom: current.decision.tier,\n\t\t\t\t\treasonCode: \"local_model_not_ready_fallback\",\n\t\t\t\t\treasons: [\n\t\t\t\t\t\t...current.decision.reasons,\n\t\t\t\t\t\t`Local model not ready (${readiness.reason}); escalated to ${escalated.tier}`,\n\t\t\t\t\t],\n\t\t\t\t\tmodel: this.deps.formatModel(escalated.model),\n\t\t\t\t},\n\t\t\t};\n\t\t}\n\t\treturn current;\n\t}\n}\n"]}

package/dist/core/models/adaptation-store.d.ts

@@ -17,11 +17,13 @@ export type ModelProtocolCalibration = {
     variantsTried: string[];
 };
 export type ModelToolProbeVerdict = "native" | "text-protocol" | "none";
+export type NativeToolProbeGrade = "task" | "echo-only" | "absent";
 export interface ModelToolProbe {
     version: number;
     status: ModelToolProbeVerdict;
     probedAt: string;
     variant?: string;
+    nativeGrade?: NativeToolProbeGrade;
     diagnostic?: string;
 }
 export interface ModelTeachStats {

package/dist/core/models/adaptation-store.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"adaptation-store.d.ts","sourceRoot":"","sources":["../../../src/core/models/adaptation-store.ts"],"names":[],"mappings":"AAEA,OAAO,EAA0B,KAAK,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAMlF,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,wBAAwB,GACjC;IACA,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,YAAY,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACpB,GACD;IACA,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,QAAQ,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACvB,CAAC;AAEL,MAAM,MAAM,qBAAqB,GAAG,QAAQ,GAAG,eAAe,GAAG,MAAM,CAAC;AAExE,MAAM,WAAW,cAAc;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,qBAAqB,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,eAAe;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,sBAAsB;IACtC,KAAK,EAAE,mBAAmB,EAAE,CAAC;IAC7B,QAAQ,CAAC,EAAE,wBAAwB,CAAC;IACpC,SAAS,CAAC,EAAE,cAAc,CAAC;IAC3B,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;CAC5C;AAED,MAAM,WAAW,qBAAqB;IACrC,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,sBAAsB,CAAC;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;CACtB;AAmGD,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,EAG9E;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,GAAG,oBAAoB,CAE5G;IAED,OAAO,CAAC,IAAI;IAaZ,OAAO,CAAC,KAAK;IAKb,OAAO,CAAC,KAAK;IASb,2FAA2F;IAC3F,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,sBAAsB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAEvF;IAED,sFAAsF;IACtF,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,GAAE,IAAiB,GAAG,sBAAsB,CAWjE;IAED,qEAAqE;IACrE,OAAO,CACN,KAAK,EAAE,MAAM,EACb,IAAI,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE,EAC5E,GAAG,OAAa,GACd,qBAAqB,CAUvB;IAED,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,OAAO,CAMhE;IAED,yEAAyE;IACzE,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,qBAAqB,GAAG,SAAS,CAO7F;IAED,WAAW,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,wBAAwB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIjG;IAED,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIzF;IAED,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,OAAO,CAMtD;IAED,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,eAAe,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIrG;IAED,sEAAsE;IACtE,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,qBAAqB,EAAE,CAMnD;IAED,6EAA6E;IAC7E,MAAM,IAAI,qBAAqB,EAAE,CAKhC;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { currentHostFingerprint, type HostFingerprint } from \"./fitness-store.ts\";\n\nconst STORE_VERSION = 1;\nconst MAX_RULES_PER_MODEL = 5;\nconst RETIRE_AFTER_MS = 30 * 24 * 60 * 60 * 1000;\n\nexport interface ModelAdaptationRule {\n\tmode: string;\n\ttext: string;\n\taddedAt: string;\n\tlastFiredAt: string;\n}\n\nexport type ModelProtocolCalibration =\n\t| {\n\t\t\tversion: number;\n\t\t\tstatus?: \"calibrated\";\n\t\t\tvariant: string;\n\t\t\tcalibratedAt: string;\n\t  }\n\t| {\n\t\t\tversion: number;\n\t\t\tstatus: \"failed\";\n\t\t\tattemptedAt: string;\n\t\t\tvariantsTried: string[];\n\t  };\n\nexport type ModelToolProbeVerdict = \"native\" | \"text-protocol\" | \"none\";\n\nexport interface ModelToolProbe {\n\tversion: number;\n\tstatus: ModelToolProbeVerdict;\n\tprobedAt: string;\n\tvariant?: string;\n\tdiagnostic?: string;\n}\n\nexport interface ModelTeachStats {\n\ttaught: number;\n\trecurrenceBefore: number;\n\trecurrenceAfter: number;\n}\n\nexport interface ModelAdaptationProfile {\n\trules: ModelAdaptationRule[];\n\tprotocol?: ModelProtocolCalibration;\n\ttoolProbe?: ModelToolProbe;\n\tteachStats: Record<string, ModelTeachStats>;\n}\n\nexport interface StoredModelAdaptation {\n\tmodel: string;\n\tprofile: ModelAdaptationProfile;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface AdaptationStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored adaptation profile. */\n\thosts: Record<string, Record<string, StoredModelAdaptation>>;\n}\n\nfunction emptyProfile(): ModelAdaptationProfile {\n\treturn { rules: [], teachStats: {} };\n}\n\nfunction normalizeProfile(profile: Partial<ModelAdaptationProfile> | undefined): ModelAdaptationProfile {\n\treturn {\n\t\trules: Array.isArray(profile?.rules) ? profile.rules.filter(isRule) : [],\n\t\t...(isProtocol(profile?.protocol) && { protocol: profile.protocol }),\n\t\t...(isToolProbe(profile?.toolProbe) && { toolProbe: profile.toolProbe }),\n\t\tteachStats: isRecord(profile?.teachStats) ? filterTeachStats(profile.teachStats) : {},\n\t};\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value);\n}\n\nfunction isRule(value: unknown): value is ModelAdaptationRule {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.mode === \"string\" &&\n\t\ttypeof value.text === \"string\" &&\n\t\ttypeof value.addedAt === \"string\" &&\n\t\ttypeof value.lastFiredAt === \"string\"\n\t);\n}\n\nfunction isProtocol(value: unknown): value is ModelProtocolCalibration {\n\tif (!isRecord(value) || typeof value.version !== \"number\") return false;\n\tif (value.status === \"failed\") {\n\t\treturn (\n\t\t\ttypeof value.attemptedAt === \"string\" &&\n\t\t\tArray.isArray(value.variantsTried) &&\n\t\t\tvalue.variantsTried.every((variant) => typeof variant === \"string\")\n\t\t);\n\t}\n\treturn (\n\t\t(value.status === undefined || value.status === \"calibrated\") &&\n\t\ttypeof value.variant === \"string\" &&\n\t\ttypeof value.calibratedAt === \"string\"\n\t);\n}\n\nfunction isToolProbe(value: unknown): value is ModelToolProbe {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.version === \"number\" &&\n\t\t(value.status === \"native\" || value.status === \"text-protocol\" || value.status === \"none\") &&\n\t\ttypeof value.probedAt === \"string\" &&\n\t\t(value.variant === undefined || typeof value.variant === \"string\") &&\n\t\t(value.diagnostic === undefined || typeof value.diagnostic === \"string\")\n\t);\n}\n\nfunction isTeachStats(value: unknown): value is ModelTeachStats {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.taught === \"number\" &&\n\t\ttypeof value.recurrenceBefore === \"number\" &&\n\t\ttypeof value.recurrenceAfter === \"number\"\n\t);\n}\n\nfunction filterTeachStats(value: Record<string, unknown>): Record<string, ModelTeachStats> {\n\treturn Object.fromEntries(\n\t\tObject.entries(value).filter((entry): entry is [string, ModelTeachStats] => isTeachStats(entry[1])),\n\t);\n}\n\nfunction ruleRecency(rule: ModelAdaptationRule): number {\n\tconst lastFired = Date.parse(rule.lastFiredAt);\n\tif (Number.isFinite(lastFired)) return lastFired;\n\tconst added = Date.parse(rule.addedAt);\n\treturn Number.isFinite(added) ? added : 0;\n}\n\nfunction pruneRetiredRules(rules: readonly ModelAdaptationRule[], now: Date): ModelAdaptationRule[] {\n\tconst cutoff = now.getTime() - RETIRE_AFTER_MS;\n\treturn rules.filter((rule) => ruleRecency(rule) >= cutoff);\n}\n\nfunction enforceRuleCap(rules: readonly ModelAdaptationRule[]): ModelAdaptationRule[] {\n\tif (rules.length <= MAX_RULES_PER_MODEL) return [...rules];\n\treturn [...rules].sort((a, b) => ruleRecency(b) - ruleRecency(a)).slice(0, MAX_RULES_PER_MODEL);\n}\n\nfunction mergeRule(rules: readonly ModelAdaptationRule[], rule: ModelAdaptationRule): ModelAdaptationRule[] {\n\tconst withoutSameMode = rules.filter((existing) => existing.mode !== rule.mode);\n\treturn enforceRuleCap([...withoutSameMode, rule]);\n}\n\nexport class ModelAdaptationStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): ModelAdaptationStore {\n\t\treturn new ModelAdaptationStore(join(agentDir, \"state\", \"model-adaptation.json\"), options);\n\t}\n\n\tprivate load(): AdaptationStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: STORE_VERSION, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as AdaptationStoreFile;\n\t\t\tif (parsed && parsed.version === STORE_VERSION && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: STORE_VERSION, hosts: {} };\n\t}\n\n\tprivate write(file: AdaptationStoreFile): void {\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\tprivate store(model: string, profile: ModelAdaptationProfile, at: string): StoredModelAdaptation {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredModelAdaptation = { model, profile: normalizeProfile(profile), at, host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tthis.write(file);\n\t\treturn entry;\n\t}\n\n\t/** Persist the profile for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, profile: ModelAdaptationProfile, at?: string): StoredModelAdaptation {\n\t\treturn this.store(model, profile, at ?? new Date().toISOString());\n\t}\n\n\t/** Profile for a model on the current host; prunes retired rules before returning. */\n\tget(model: string, now: Date = new Date()): ModelAdaptationProfile {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tconst entry = file.hosts[host.id]?.[model];\n\t\tif (!entry) return emptyProfile();\n\t\tconst profile = normalizeProfile(entry.profile);\n\t\tconst prunedRules = pruneRetiredRules(profile.rules, now);\n\t\tif (prunedRules.length !== profile.rules.length) {\n\t\t\treturn this.store(model, { ...profile, rules: prunedRules }, now.toISOString()).profile;\n\t\t}\n\t\treturn profile;\n\t}\n\n\t/** Add or replace one standing rule, enforcing the per-model cap. */\n\taddRule(\n\t\tmodel: string,\n\t\trule: { mode: string; text: string; addedAt?: string; lastFiredAt?: string },\n\t\tnow = new Date(),\n\t): StoredModelAdaptation {\n\t\tconst profile = this.get(model, now);\n\t\tconst at = now.toISOString();\n\t\tconst nextRule: ModelAdaptationRule = {\n\t\t\tmode: rule.mode,\n\t\t\ttext: rule.text,\n\t\t\taddedAt: rule.addedAt ?? at,\n\t\t\tlastFiredAt: rule.lastFiredAt ?? at,\n\t\t};\n\t\treturn this.store(model, { ...profile, rules: mergeRule(profile.rules, nextRule) }, at);\n\t}\n\n\tremoveRule(model: string, mode: string, at = new Date()): boolean {\n\t\tconst profile = this.get(model, at);\n\t\tconst rules = profile.rules.filter((rule) => rule.mode !== mode);\n\t\tif (rules.length === profile.rules.length) return false;\n\t\tthis.store(model, { ...profile, rules }, at.toISOString());\n\t\treturn true;\n\t}\n\n\t/** Update last-fired recency for an existing rule. No-op when absent. */\n\tmarkRuleFired(model: string, mode: string, at = new Date()): StoredModelAdaptation | undefined {\n\t\tconst profile = this.get(model, at);\n\t\tconst rules = profile.rules.map((rule) =>\n\t\t\trule.mode === mode ? { ...rule, lastFiredAt: at.toISOString() } : rule,\n\t\t);\n\t\tif (rules.every((rule, index) => rule === profile.rules[index])) return undefined;\n\t\treturn this.store(model, { ...profile, rules }, at.toISOString());\n\t}\n\n\tsetProtocol(model: string, protocol: ModelProtocolCalibration, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? (protocol.status === \"failed\" ? protocol.attemptedAt : protocol.calibratedAt);\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, protocol }, now);\n\t}\n\n\tsetToolProbe(model: string, toolProbe: ModelToolProbe, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? toolProbe.probedAt;\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, toolProbe }, now);\n\t}\n\n\tremoveProtocol(model: string, at = new Date()): boolean {\n\t\tconst profile = this.get(model, at);\n\t\tif (!profile.protocol) return false;\n\t\tconst { protocol: _protocol, ...rest } = profile;\n\t\tthis.store(model, rest, at.toISOString());\n\t\treturn true;\n\t}\n\n\tsetTeachStats(model: string, mode: string, stats: ModelTeachStats, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? new Date().toISOString();\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, teachStats: { ...profile.teachStats, [mode]: stats } }, now);\n\t}\n\n\t/** Profiles for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredModelAdaptation[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {}).map((entry) => ({\n\t\t\t...entry,\n\t\t\tprofile: normalizeProfile(entry.profile),\n\t\t}));\n\t}\n\n\t/** Every stored profile across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredModelAdaptation[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) =>\n\t\t\tObject.values(models).map((entry) => ({ ...entry, profile: normalizeProfile(entry.profile) })),\n\t\t);\n\t}\n}\n"]}
+{"version":3,"file":"adaptation-store.d.ts","sourceRoot":"","sources":["../../../src/core/models/adaptation-store.ts"],"names":[],"mappings":"AAEA,OAAO,EAA0B,KAAK,eAAe,EAAE,MAAM,oBAAoB,CAAC;AAMlF,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,MAAM,wBAAwB,GACjC;IACA,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,CAAC,EAAE,YAAY,CAAC;IACtB,OAAO,EAAE,MAAM,CAAC;IAChB,YAAY,EAAE,MAAM,CAAC;CACpB,GACD;IACA,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,QAAQ,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACvB,CAAC;AAEL,MAAM,MAAM,qBAAqB,GAAG,QAAQ,GAAG,eAAe,GAAG,MAAM,CAAC;AACxE,MAAM,MAAM,oBAAoB,GAAG,MAAM,GAAG,WAAW,GAAG,QAAQ,CAAC;AAEnE,MAAM,WAAW,cAAc;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,qBAAqB,CAAC;IAC9B,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,WAAW,CAAC,EAAE,oBAAoB,CAAC;IACnC,UAAU,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,eAAe;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,sBAAsB;IACtC,KAAK,EAAE,mBAAmB,EAAE,CAAC;IAC7B,QAAQ,CAAC,EAAE,wBAAwB,CAAC;IACpC,SAAS,CAAC,EAAE,cAAc,CAAC;IAC3B,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;CAC5C;AAED,MAAM,WAAW,qBAAqB;IACrC,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,sBAAsB,CAAC;IAChC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,eAAe,CAAC;CACtB;AAuGD,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAwB;IAEpD,YAAY,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,EAG9E;IAED,MAAM,CAAC,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE;QAAE,WAAW,CAAC,EAAE,MAAM,eAAe,CAAA;KAAE,GAAG,oBAAoB,CAE5G;IAED,OAAO,CAAC,IAAI;IAaZ,OAAO,CAAC,KAAK;IAKb,OAAO,CAAC,KAAK;IASb,2FAA2F;IAC3F,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,EAAE,sBAAsB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAEvF;IAED,sFAAsF;IACtF,GAAG,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,GAAE,IAAiB,GAAG,sBAAsB,CAWjE;IAED,qEAAqE;IACrE,OAAO,CACN,KAAK,EAAE,MAAM,EACb,IAAI,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAA;KAAE,EAC5E,GAAG,OAAa,GACd,qBAAqB,CAUvB;IAED,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,OAAO,CAMhE;IAED,yEAAyE;IACzE,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,qBAAqB,GAAG,SAAS,CAO7F;IAED,WAAW,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,wBAAwB,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIjG;IAED,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,cAAc,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIzF;IAED,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,OAAa,GAAG,OAAO,CAMtD;IAED,aAAa,CAAC,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,eAAe,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,qBAAqB,CAIrG;IAED,sEAAsE;IACtE,UAAU,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,qBAAqB,EAAE,CAMnD;IAED,6EAA6E;IAC7E,MAAM,IAAI,qBAAqB,EAAE,CAKhC;CACD","sourcesContent":["import { existsSync, mkdirSync, readFileSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport { currentHostFingerprint, type HostFingerprint } from \"./fitness-store.ts\";\n\nconst STORE_VERSION = 1;\nconst MAX_RULES_PER_MODEL = 5;\nconst RETIRE_AFTER_MS = 30 * 24 * 60 * 60 * 1000;\n\nexport interface ModelAdaptationRule {\n\tmode: string;\n\ttext: string;\n\taddedAt: string;\n\tlastFiredAt: string;\n}\n\nexport type ModelProtocolCalibration =\n\t| {\n\t\t\tversion: number;\n\t\t\tstatus?: \"calibrated\";\n\t\t\tvariant: string;\n\t\t\tcalibratedAt: string;\n\t  }\n\t| {\n\t\t\tversion: number;\n\t\t\tstatus: \"failed\";\n\t\t\tattemptedAt: string;\n\t\t\tvariantsTried: string[];\n\t  };\n\nexport type ModelToolProbeVerdict = \"native\" | \"text-protocol\" | \"none\";\nexport type NativeToolProbeGrade = \"task\" | \"echo-only\" | \"absent\";\n\nexport interface ModelToolProbe {\n\tversion: number;\n\tstatus: ModelToolProbeVerdict;\n\tprobedAt: string;\n\tvariant?: string;\n\tnativeGrade?: NativeToolProbeGrade;\n\tdiagnostic?: string;\n}\n\nexport interface ModelTeachStats {\n\ttaught: number;\n\trecurrenceBefore: number;\n\trecurrenceAfter: number;\n}\n\nexport interface ModelAdaptationProfile {\n\trules: ModelAdaptationRule[];\n\tprotocol?: ModelProtocolCalibration;\n\ttoolProbe?: ModelToolProbe;\n\tteachStats: Record<string, ModelTeachStats>;\n}\n\nexport interface StoredModelAdaptation {\n\tmodel: string;\n\tprofile: ModelAdaptationProfile;\n\tat: string;\n\thost: HostFingerprint;\n}\n\ninterface AdaptationStoreFile {\n\tversion: 1;\n\t/** hostId -> modelRef -> latest stored adaptation profile. */\n\thosts: Record<string, Record<string, StoredModelAdaptation>>;\n}\n\nfunction emptyProfile(): ModelAdaptationProfile {\n\treturn { rules: [], teachStats: {} };\n}\n\nfunction normalizeProfile(profile: Partial<ModelAdaptationProfile> | undefined): ModelAdaptationProfile {\n\treturn {\n\t\trules: Array.isArray(profile?.rules) ? profile.rules.filter(isRule) : [],\n\t\t...(isProtocol(profile?.protocol) && { protocol: profile.protocol }),\n\t\t...(isToolProbe(profile?.toolProbe) && { toolProbe: profile.toolProbe }),\n\t\tteachStats: isRecord(profile?.teachStats) ? filterTeachStats(profile.teachStats) : {},\n\t};\n}\n\nfunction isRecord(value: unknown): value is Record<string, unknown> {\n\treturn typeof value === \"object\" && value !== null && !Array.isArray(value);\n}\n\nfunction isRule(value: unknown): value is ModelAdaptationRule {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.mode === \"string\" &&\n\t\ttypeof value.text === \"string\" &&\n\t\ttypeof value.addedAt === \"string\" &&\n\t\ttypeof value.lastFiredAt === \"string\"\n\t);\n}\n\nfunction isProtocol(value: unknown): value is ModelProtocolCalibration {\n\tif (!isRecord(value) || typeof value.version !== \"number\") return false;\n\tif (value.status === \"failed\") {\n\t\treturn (\n\t\t\ttypeof value.attemptedAt === \"string\" &&\n\t\t\tArray.isArray(value.variantsTried) &&\n\t\t\tvalue.variantsTried.every((variant) => typeof variant === \"string\")\n\t\t);\n\t}\n\treturn (\n\t\t(value.status === undefined || value.status === \"calibrated\") &&\n\t\ttypeof value.variant === \"string\" &&\n\t\ttypeof value.calibratedAt === \"string\"\n\t);\n}\n\nfunction isToolProbe(value: unknown): value is ModelToolProbe {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.version === \"number\" &&\n\t\t(value.status === \"native\" || value.status === \"text-protocol\" || value.status === \"none\") &&\n\t\ttypeof value.probedAt === \"string\" &&\n\t\t(value.variant === undefined || typeof value.variant === \"string\") &&\n\t\t(value.nativeGrade === undefined ||\n\t\t\tvalue.nativeGrade === \"task\" ||\n\t\t\tvalue.nativeGrade === \"echo-only\" ||\n\t\t\tvalue.nativeGrade === \"absent\") &&\n\t\t(value.diagnostic === undefined || typeof value.diagnostic === \"string\")\n\t);\n}\n\nfunction isTeachStats(value: unknown): value is ModelTeachStats {\n\treturn (\n\t\tisRecord(value) &&\n\t\ttypeof value.taught === \"number\" &&\n\t\ttypeof value.recurrenceBefore === \"number\" &&\n\t\ttypeof value.recurrenceAfter === \"number\"\n\t);\n}\n\nfunction filterTeachStats(value: Record<string, unknown>): Record<string, ModelTeachStats> {\n\treturn Object.fromEntries(\n\t\tObject.entries(value).filter((entry): entry is [string, ModelTeachStats] => isTeachStats(entry[1])),\n\t);\n}\n\nfunction ruleRecency(rule: ModelAdaptationRule): number {\n\tconst lastFired = Date.parse(rule.lastFiredAt);\n\tif (Number.isFinite(lastFired)) return lastFired;\n\tconst added = Date.parse(rule.addedAt);\n\treturn Number.isFinite(added) ? added : 0;\n}\n\nfunction pruneRetiredRules(rules: readonly ModelAdaptationRule[], now: Date): ModelAdaptationRule[] {\n\tconst cutoff = now.getTime() - RETIRE_AFTER_MS;\n\treturn rules.filter((rule) => ruleRecency(rule) >= cutoff);\n}\n\nfunction enforceRuleCap(rules: readonly ModelAdaptationRule[]): ModelAdaptationRule[] {\n\tif (rules.length <= MAX_RULES_PER_MODEL) return [...rules];\n\treturn [...rules].sort((a, b) => ruleRecency(b) - ruleRecency(a)).slice(0, MAX_RULES_PER_MODEL);\n}\n\nfunction mergeRule(rules: readonly ModelAdaptationRule[], rule: ModelAdaptationRule): ModelAdaptationRule[] {\n\tconst withoutSameMode = rules.filter((existing) => existing.mode !== rule.mode);\n\treturn enforceRuleCap([...withoutSameMode, rule]);\n}\n\nexport class ModelAdaptationStore {\n\tprivate readonly filePath: string;\n\tprivate readonly fingerprint: () => HostFingerprint;\n\n\tconstructor(filePath: string, options?: { fingerprint?: () => HostFingerprint }) {\n\t\tthis.filePath = filePath;\n\t\tthis.fingerprint = options?.fingerprint ?? currentHostFingerprint;\n\t}\n\n\tstatic forAgentDir(agentDir: string, options?: { fingerprint?: () => HostFingerprint }): ModelAdaptationStore {\n\t\treturn new ModelAdaptationStore(join(agentDir, \"state\", \"model-adaptation.json\"), options);\n\t}\n\n\tprivate load(): AdaptationStoreFile {\n\t\ttry {\n\t\t\tif (!existsSync(this.filePath)) return { version: STORE_VERSION, hosts: {} };\n\t\t\tconst parsed = JSON.parse(readFileSync(this.filePath, \"utf-8\")) as AdaptationStoreFile;\n\t\t\tif (parsed && parsed.version === STORE_VERSION && parsed.hosts && typeof parsed.hosts === \"object\") {\n\t\t\t\treturn parsed;\n\t\t\t}\n\t\t} catch {\n\t\t\t// Unreadable/corrupt store: start fresh in memory; the next save rewrites the file.\n\t\t}\n\t\treturn { version: STORE_VERSION, hosts: {} };\n\t}\n\n\tprivate write(file: AdaptationStoreFile): void {\n\t\tmkdirSync(dirname(this.filePath), { recursive: true });\n\t\twriteFileSync(this.filePath, `${JSON.stringify(file, null, \"\\t\")}\\n`, \"utf-8\");\n\t}\n\n\tprivate store(model: string, profile: ModelAdaptationProfile, at: string): StoredModelAdaptation {\n\t\tconst host = this.fingerprint();\n\t\tconst entry: StoredModelAdaptation = { model, profile: normalizeProfile(profile), at, host };\n\t\tconst file = this.load();\n\t\tfile.hosts[host.id] = { ...(file.hosts[host.id] ?? {}), [model]: entry };\n\t\tthis.write(file);\n\t\treturn entry;\n\t}\n\n\t/** Persist the profile for a model on the CURRENT host. Best-effort, returns the entry. */\n\tsave(model: string, profile: ModelAdaptationProfile, at?: string): StoredModelAdaptation {\n\t\treturn this.store(model, profile, at ?? new Date().toISOString());\n\t}\n\n\t/** Profile for a model on the current host; prunes retired rules before returning. */\n\tget(model: string, now: Date = new Date()): ModelAdaptationProfile {\n\t\tconst host = this.fingerprint();\n\t\tconst file = this.load();\n\t\tconst entry = file.hosts[host.id]?.[model];\n\t\tif (!entry) return emptyProfile();\n\t\tconst profile = normalizeProfile(entry.profile);\n\t\tconst prunedRules = pruneRetiredRules(profile.rules, now);\n\t\tif (prunedRules.length !== profile.rules.length) {\n\t\t\treturn this.store(model, { ...profile, rules: prunedRules }, now.toISOString()).profile;\n\t\t}\n\t\treturn profile;\n\t}\n\n\t/** Add or replace one standing rule, enforcing the per-model cap. */\n\taddRule(\n\t\tmodel: string,\n\t\trule: { mode: string; text: string; addedAt?: string; lastFiredAt?: string },\n\t\tnow = new Date(),\n\t): StoredModelAdaptation {\n\t\tconst profile = this.get(model, now);\n\t\tconst at = now.toISOString();\n\t\tconst nextRule: ModelAdaptationRule = {\n\t\t\tmode: rule.mode,\n\t\t\ttext: rule.text,\n\t\t\taddedAt: rule.addedAt ?? at,\n\t\t\tlastFiredAt: rule.lastFiredAt ?? at,\n\t\t};\n\t\treturn this.store(model, { ...profile, rules: mergeRule(profile.rules, nextRule) }, at);\n\t}\n\n\tremoveRule(model: string, mode: string, at = new Date()): boolean {\n\t\tconst profile = this.get(model, at);\n\t\tconst rules = profile.rules.filter((rule) => rule.mode !== mode);\n\t\tif (rules.length === profile.rules.length) return false;\n\t\tthis.store(model, { ...profile, rules }, at.toISOString());\n\t\treturn true;\n\t}\n\n\t/** Update last-fired recency for an existing rule. No-op when absent. */\n\tmarkRuleFired(model: string, mode: string, at = new Date()): StoredModelAdaptation | undefined {\n\t\tconst profile = this.get(model, at);\n\t\tconst rules = profile.rules.map((rule) =>\n\t\t\trule.mode === mode ? { ...rule, lastFiredAt: at.toISOString() } : rule,\n\t\t);\n\t\tif (rules.every((rule, index) => rule === profile.rules[index])) return undefined;\n\t\treturn this.store(model, { ...profile, rules }, at.toISOString());\n\t}\n\n\tsetProtocol(model: string, protocol: ModelProtocolCalibration, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? (protocol.status === \"failed\" ? protocol.attemptedAt : protocol.calibratedAt);\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, protocol }, now);\n\t}\n\n\tsetToolProbe(model: string, toolProbe: ModelToolProbe, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? toolProbe.probedAt;\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, toolProbe }, now);\n\t}\n\n\tremoveProtocol(model: string, at = new Date()): boolean {\n\t\tconst profile = this.get(model, at);\n\t\tif (!profile.protocol) return false;\n\t\tconst { protocol: _protocol, ...rest } = profile;\n\t\tthis.store(model, rest, at.toISOString());\n\t\treturn true;\n\t}\n\n\tsetTeachStats(model: string, mode: string, stats: ModelTeachStats, at?: string): StoredModelAdaptation {\n\t\tconst now = at ?? new Date().toISOString();\n\t\tconst profile = this.get(model, new Date(now));\n\t\treturn this.store(model, { ...profile, teachStats: { ...profile.teachStats, [mode]: stats } }, now);\n\t}\n\n\t/** Profiles for the current host (default) or an explicit host id. */\n\tgetForHost(hostId?: string): StoredModelAdaptation[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts[hostId ?? this.fingerprint().id] ?? {}).map((entry) => ({\n\t\t\t...entry,\n\t\t\tprofile: normalizeProfile(entry.profile),\n\t\t}));\n\t}\n\n\t/** Every stored profile across all hosts (for cross-machine comparisons). */\n\tgetAll(): StoredModelAdaptation[] {\n\t\tconst file = this.load();\n\t\treturn Object.values(file.hosts).flatMap((models) =>\n\t\t\tObject.values(models).map((entry) => ({ ...entry, profile: normalizeProfile(entry.profile) })),\n\t\t);\n\t}\n}\n"]}

package/dist/core/models/adaptation-store.js

@@ -43,6 +43,10 @@ function isToolProbe(value) {
         (value.status === "native" || value.status === "text-protocol" || value.status === "none") &&
         typeof value.probedAt === "string" &&
         (value.variant === undefined || typeof value.variant === "string") &&
+        (value.nativeGrade === undefined ||
+            value.nativeGrade === "task" ||
+            value.nativeGrade === "echo-only" ||
+            value.nativeGrade === "absent") &&
         (value.diagnostic === undefined || typeof value.diagnostic === "string"));
 }
 function isTeachStats(value) {