@caupulican/pi-adaptative 0.80.99 → 0.80.102

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

@@ -25,6 +25,22 @@ export interface InstalledLocalModel {
     name: string;
     sizeBytes: number;
 }
+/** Pinned ollama release for the managed installer below. Bump here when needed — one constant. */
+export declare const OLLAMA_PINNED_VERSION = "0.31.1";
+export type OllamaAssetKind = "tar-zst" | "tar-gz" | "zip";
+export interface OllamaReleaseAsset {
+    name: string;
+    kind: OllamaAssetKind;
+}
+/**
+ * Maps a platform/arch pair (node's `os.platform()`/`os.arch()`) to the exact ollama release asset
+ * name for {@link OLLAMA_PINNED_VERSION} — verified against the real ollama/ollama GitHub release
+ * and its own install.sh, not guessed: darwin ships a CLI `.tgz` (the `.app`/`.dmg`/`Ollama-*.zip`
+ * assets are the separate GUI-app installer, not what we want), linux ships `.tar.zst` per arch,
+ * windows ships `.zip` per arch. Pure and exported so it's independently testable and reusable
+ * (e.g. by a future doctor-driven managed install) without touching OllamaRuntime.
+ */
+export declare function resolveOllamaAsset(plat: string, architecture: string): OllamaReleaseAsset | undefined;
 export interface LocalRuntimeDeps {
     fetchFn?: typeof fetch;
     spawnFn?: (command: string, args: string[], options: {
@@ -34,6 +50,22 @@ export interface LocalRuntimeDeps {
     envPath?: string;
     homeDir?: string;
     sleepFn?: (ms: number) => Promise<void>;
+    /** os.platform()/os.arch() equivalents — injectable so asset resolution is testable per platform. */
+    platform?: () => string;
+    arch?: () => string;
+    /** Node's built-in zstd decompress transform, when THIS runtime's node:zlib has it — undefined
+     * forces the system-zstd fallback. Injectable so tests can force either path deterministically
+     * instead of depending on whatever Node happens to be running the test. */
+    createZstdDecompress?: () => NodeJS.ReadWriteStream;
+    /** Whether a named command exists on PATH (e.g. system `zstd`), for the extraction fallback. */
+    hasCommand?: (command: string) => boolean;
+    /** Runs the extraction step for a downloaded archive. Injectable so installManaged's
+     * download->extract->verify orchestration is testable without a real tar/zstd/unzip pipeline;
+     * defaults to the real spawn-based extractor. */
+    extractArchive?: (input: NodeJS.ReadableStream, destDir: string, kind: OllamaAssetKind) => Promise<{
+        ok: boolean;
+        error?: string;
+    }>;
 }
 export declare class OllamaRuntime {
     private readonly _agentDir;
@@ -44,6 +76,11 @@ export declare class OllamaRuntime {
     private readonly _envPath;
     private readonly _homeDir;
     private readonly _sleep;
+    private readonly _platform;
+    private readonly _arch;
+    private readonly _createZstdDecompress;
+    private readonly _hasCommand;
+    private readonly _extractArchiveOverride;
     private _child;
     constructor(args: {
         agentDir: string;
@@ -55,16 +92,53 @@ export declare class OllamaRuntime {
     private _findBinary;
     private _serverUp;
     detect(): Promise<LocalRuntimeStatus>;
-    /** GUIDE MODE: exact manual steps, printed, never executed (no `curl | sh`, no sudo). */
+    /**
+     * Manual-steps fallback text — shown when the managed install below wasn't offered (headless),
+     * was declined, or itself failed. #31 reversed the stance for ollama specifically: pi CAN install
+     * it for you now (consent-gated, a controlled download into pi's own runtimes dir, never
+     * `curl | sh`) — these steps are the alternative for when that path isn't taken.
+     */
     installGuide(): string[];
+    /** Which decompressor to use for a `.tar.zst` asset — native (this Node's own zlib) is always
+     * preferred when available, even if a system `zstd` ALSO exists, since it needs no external
+     * dependency at all. Verified per-runtime by feature detection (see the constructor), not by
+     * guessing from a Node version number. */
+    private _chooseZstdStrategy;
+    /**
+     * Managed install (#31): a consent-gated, controlled download of the pinned ollama release into
+     * pi's OWN runtimes dir (`<agentDir>/runtimes/ollama/`) — never `curl | sh`, never a second
+     * server instance, and distinct from {@link start}'s owned MODEL storage (this is the binary
+     * itself). Callers (the router's consent flow, and later a doctor-driven install) own the
+     * consent/UI; this method only does the mechanical download+extract+verify and reports the
+     * outcome honestly. `onProgress` is best-effort UI feedback, not a completion signal.
+     */
+    installManaged(onProgress?: (status: string) => void): Promise<{
+        ok: boolean;
+        error?: string;
+    }>;
+    private _extractArchive;
+    private _extractTar;
+    private _extractZip;
+    private _spawnAndPoll;
     /**
      * Start a pi-managed serve with OWNED storage and the hardened env verified on this class of
-     * hardware. No-op (reported) when a server already responds — pi never double-serves.
+     * hardware. No-op (reported) when a server already responds — pi never double-serves. Used by
+     * `/models add` et al, where isolated per-model-pull storage is the point.
      */
     start(): Promise<{
         started: boolean;
         reason: string;
     }>;
+    /**
+     * Start a serve that REUSES the user's own existing models directory (no `OLLAMA_MODELS`
+     * override — falls through to Ollama's own default, `~/.ollama`), for callers that must see the
+     * user's already-pulled models rather than pi's isolated/owned storage. Same idempotency and
+     * hardened perf env as {@link start}; the only difference is which storage the server sees.
+     */
+    startReuseExisting(): Promise<{
+        started: boolean;
+        reason: string;
+    }>;
     /** Resource hygiene only: stops the pi-managed serve process; never deletes anything. */
     stop(): {
         stopped: boolean;

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

@@ -1 +1 @@
-{"version":3,"file":"local-runtime.d.ts","sourceRoot":"","sources":["../../../src/core/models/local-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,YAAY,EAAS,MAAM,oBAAoB,CAAC;AAK9D;;;;;;;;;;;GAWG;AAEH,MAAM,WAAW,kBAAkB;IAClC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,UAAU,CAAC;IAC9C,QAAQ,EAAE,OAAO,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,6FAA6F;IAC7F,WAAW,EAAE,OAAO,CAAC;IACrB,uEAAuE;IACvE,cAAc,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,gBAAgB;IAChC,OAAO,CAAC,EAAE,OAAO,KAAK,CAAC;IACvB,OAAO,CAAC,EAAE,CACT,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC,UAAU,CAAA;KAAE,KAC/B,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,MAAM,GAAG,OAAO,GAAG,IAAI,CAAC,CAAC;IACzD,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;IACrC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CACxC;AAOD,qBAAa,aAAa;IACzB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA2C;IAClE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA4B;IACpD,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAgC;IACvD,OAAO,CAAC,MAAM,CAAkE;IAEhF,YAAY,IAAI,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,gBAAgB,CAAA;KAAE,EAShF;IAED,IAAI,OAAO,IAAI,MAAM,CAEpB;IAED,cAAc,IAAI,MAAM,CAEvB;IAED,OAAO,CAAC,WAAW;YAaL,SAAS;IAWjB,MAAM,IAAI,OAAO,CAAC,kBAAkB,CAAC,CAW1C;IAED,yFAAyF;IACzF,YAAY,IAAI,MAAM,EAAE,CASvB;IAED;;;OAGG;IACG,KAAK,IAAI,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CA0B3D;IAED,yFAAyF;IACzF,IAAI,IAAI;QAAE,OAAO,EAAE,OAAO,CAAA;KAAE,CAS3B;IAEK,IAAI,IAAI,OAAO,CAAC,mBAAmB,EAAE,CAAC,CAO3C;IAED,qFAAqF;IAC/E,IAAI,CAAC,GAAG,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAiDvG;IAED,6FAA2F;IACrF,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAiBlE;CACD","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport { existsSync, mkdirSync } from \"node:fs\";\nimport { homedir } from \"node:os\";\nimport { delimiter, join } from \"node:path\";\n\n/**\n * Local model runtime manager (local-model-lifecycle-design.md): Ollama first, interface kept\n * runtime-agnostic. Pi SPAWNS the serve process itself with OWNED model storage\n * (`OLLAMA_MODELS=<agentDir>/models/ollama`) so every downloaded weight lives inside pi's tree —\n * per-model disk accounting is trivial and full cleanup is one directory. If a system server is\n * already running, pi uses it instead of double-installing, with the honest tradeoff surfaced:\n * storage then lives in the system daemon's dir and removal control is limited to `ollama rm`.\n *\n * Hard boundaries (design \"Hard boundaries\"): lifecycle actions are USER commands only — this\n * module is never exposed as a model-invokable tool; install is GUIDE MODE (exact manual steps,\n * never `curl | sh`); removal is explicit, disclosed, and never automatic.\n */\n\nexport interface LocalRuntimeStatus {\n\tbinaryPath?: string;\n\tbinarySource?: \"system\" | \"user\" | \"pi-owned\";\n\tserverUp: boolean;\n\tserverUrl: string;\n\t/** True when the responding server is a child process pi spawned (owned storage applies). */\n\tmanagedByPi: boolean;\n\t/** Owned weights directory (only meaningful for pi-managed serves). */\n\townedModelsDir: string;\n}\n\nexport interface InstalledLocalModel {\n\tname: string;\n\tsizeBytes: number;\n}\n\nexport interface LocalRuntimeDeps {\n\tfetchFn?: typeof fetch;\n\tspawnFn?: (\n\t\tcommand: string,\n\t\targs: string[],\n\t\toptions: { env: NodeJS.ProcessEnv },\n\t) => Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\">;\n\texistsFn?: (path: string) => boolean;\n\tenvPath?: string;\n\thomeDir?: string;\n\tsleepFn?: (ms: number) => Promise<void>;\n}\n\nconst DEFAULT_BASE_URL = \"http://127.0.0.1:11434\";\nconst HEALTH_TIMEOUT_MS = 2_000;\nconst START_POLL_ATTEMPTS = 20;\nconst START_POLL_INTERVAL_MS = 500;\n\nexport class OllamaRuntime {\n\tprivate readonly _agentDir: string;\n\tprivate readonly _baseUrl: string;\n\tprivate readonly _fetch: typeof fetch;\n\tprivate readonly _spawn: NonNullable<LocalRuntimeDeps[\"spawnFn\"]>;\n\tprivate readonly _exists: (path: string) => boolean;\n\tprivate readonly _envPath: string;\n\tprivate readonly _homeDir: string;\n\tprivate readonly _sleep: (ms: number) => Promise<void>;\n\tprivate _child: Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\"> | undefined;\n\n\tconstructor(args: { agentDir: string; baseUrl?: string; deps?: LocalRuntimeDeps }) {\n\t\tthis._agentDir = args.agentDir;\n\t\tthis._baseUrl = (args.baseUrl ?? DEFAULT_BASE_URL).replace(/\\/$/, \"\");\n\t\tthis._fetch = args.deps?.fetchFn ?? fetch;\n\t\tthis._spawn = args.deps?.spawnFn ?? ((command, argv, options) => spawn(command, argv, options));\n\t\tthis._exists = args.deps?.existsFn ?? existsSync;\n\t\tthis._envPath = args.deps?.envPath ?? process.env.PATH ?? \"\";\n\t\tthis._homeDir = args.deps?.homeDir ?? homedir();\n\t\tthis._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));\n\t}\n\n\tget baseUrl(): string {\n\t\treturn this._baseUrl;\n\t}\n\n\townedModelsDir(): string {\n\t\treturn join(this._agentDir, \"models\", \"ollama\");\n\t}\n\n\tprivate _findBinary(): { path: string; source: \"system\" | \"user\" | \"pi-owned\" } | undefined {\n\t\tconst piOwned = join(this._agentDir, \"runtimes\", \"ollama\", \"bin\", \"ollama\");\n\t\tif (this._exists(piOwned)) return { path: piOwned, source: \"pi-owned\" };\n\t\tconst userLevel = join(this._homeDir, \".local\", \"share\", \"ollama-dist\", \"bin\", \"ollama\");\n\t\tif (this._exists(userLevel)) return { path: userLevel, source: \"user\" };\n\t\tfor (const dir of this._envPath.split(delimiter)) {\n\t\t\tif (!dir) continue;\n\t\t\tconst candidate = join(dir, \"ollama\");\n\t\t\tif (this._exists(candidate)) return { path: candidate, source: \"system\" };\n\t\t}\n\t\treturn undefined;\n\t}\n\n\tprivate async _serverUp(): Promise<boolean> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/tags`, {\n\t\t\t\tsignal: AbortSignal.timeout(HEALTH_TIMEOUT_MS),\n\t\t\t});\n\t\t\treturn response.ok;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tasync detect(): Promise<LocalRuntimeStatus> {\n\t\tconst binary = this._findBinary();\n\t\tconst serverUp = await this._serverUp();\n\t\treturn {\n\t\t\tbinaryPath: binary?.path,\n\t\t\tbinarySource: binary?.source,\n\t\t\tserverUp,\n\t\t\tserverUrl: this._baseUrl,\n\t\t\tmanagedByPi: this._child !== undefined && serverUp,\n\t\t\townedModelsDir: this.ownedModelsDir(),\n\t\t};\n\t}\n\n\t/** GUIDE MODE: exact manual steps, printed, never executed (no `curl | sh`, no sudo). */\n\tinstallGuide(): string[] {\n\t\treturn [\n\t\t\t\"Ollama is not installed. Pi never runs installers itself — manual steps (user-level, no sudo):\",\n\t\t\t\"  1. Download the pinned release archive for your platform:\",\n\t\t\t\"     https://github.com/ollama/ollama/releases (asset: ollama-linux-amd64.tar.zst or your platform)\",\n\t\t\t`  2. Extract it to ${join(this._homeDir, \".local\", \"share\", \"ollama-dist\")}`,\n\t\t\t\"  3. Re-run your /models command - pi detects the binary automatically.\",\n\t\t\t\"Alternatively install system-wide from https://ollama.com/download and pi will use the system server.\",\n\t\t];\n\t}\n\n\t/**\n\t * Start a pi-managed serve with OWNED storage and the hardened env verified on this class of\n\t * hardware. No-op (reported) when a server already responds — pi never double-serves.\n\t */\n\tasync start(): Promise<{ started: boolean; reason: string }> {\n\t\tif (await this._serverUp()) {\n\t\t\treturn { started: false, reason: this._child ? \"already_running_managed\" : \"already_running_system\" };\n\t\t}\n\t\tconst binary = this._findBinary();\n\t\tif (!binary) return { started: false, reason: \"binary_missing\" };\n\t\tmkdirSync(this.ownedModelsDir(), { recursive: true });\n\t\tconst host = this._baseUrl.replace(/^https?:\\/\\//, \"\");\n\t\tthis._child = this._spawn(binary.path, [\"serve\"], {\n\t\t\tenv: {\n\t\t\t\t...process.env,\n\t\t\t\tOLLAMA_HOST: host,\n\t\t\t\tOLLAMA_MODELS: this.ownedModelsDir(),\n\t\t\t\tOLLAMA_FLASH_ATTENTION: \"1\",\n\t\t\t\tOLLAMA_KV_CACHE_TYPE: \"q8_0\",\n\t\t\t\tOLLAMA_NUM_PARALLEL: \"1\",\n\t\t\t\tOLLAMA_MAX_LOADED_MODELS: \"3\",\n\t\t\t},\n\t\t});\n\t\tthis._child.unref?.();\n\t\tfor (let attempt = 0; attempt < START_POLL_ATTEMPTS; attempt++) {\n\t\t\tif (await this._serverUp()) return { started: true, reason: \"started\" };\n\t\t\tawait this._sleep(START_POLL_INTERVAL_MS);\n\t\t}\n\t\tthis.stop();\n\t\treturn { started: false, reason: \"health_check_timeout\" };\n\t}\n\n\t/** Resource hygiene only: stops the pi-managed serve process; never deletes anything. */\n\tstop(): { stopped: boolean } {\n\t\tif (!this._child) return { stopped: false };\n\t\ttry {\n\t\t\tthis._child.kill(\"SIGTERM\");\n\t\t} catch {\n\t\t\t// already gone\n\t\t}\n\t\tthis._child = undefined;\n\t\treturn { stopped: true };\n\t}\n\n\tasync list(): Promise<InstalledLocalModel[]> {\n\t\tconst response = await this._fetch(`${this._baseUrl}/api/tags`, { signal: AbortSignal.timeout(10_000) });\n\t\tif (!response.ok) throw new Error(`ollama list failed: HTTP ${response.status}`);\n\t\tconst data = (await response.json()) as { models?: Array<{ name?: string; size?: number }> };\n\t\treturn (data.models ?? [])\n\t\t\t.filter((model): model is { name: string; size?: number } => typeof model.name === \"string\")\n\t\t\t.map((model) => ({ name: model.name, sizeBytes: model.size ?? 0 }));\n\t}\n\n\t/** Pull a model through the server API (weights land in the SERVER's models dir). */\n\tasync pull(ref: string, onProgress?: (status: string) => void): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/pull`, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: { \"content-type\": \"application/json\" },\n\t\t\t\tbody: JSON.stringify({ model: ref, stream: true }),\n\t\t\t});\n\t\t\tif (!response.ok || !response.body) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror: `pull failed: HTTP ${response.status} ${await response.text().catch(() => \"\")}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\tconst reader = response.body.getReader();\n\t\t\tconst decoder = new TextDecoder();\n\t\t\tlet buffer = \"\";\n\t\t\tlet lastStatus = \"\";\n\t\t\tlet errorMessage: string | undefined;\n\t\t\tconst handleLine = (line: string): void => {\n\t\t\t\tif (!line.trim()) return;\n\t\t\t\ttry {\n\t\t\t\t\tconst event = JSON.parse(line) as { status?: string; error?: string };\n\t\t\t\t\tif (event.error) errorMessage = event.error;\n\t\t\t\t\tif (event.status && event.status !== lastStatus) {\n\t\t\t\t\t\tlastStatus = event.status;\n\t\t\t\t\t\tonProgress?.(event.status);\n\t\t\t\t\t}\n\t\t\t\t} catch {\n\t\t\t\t\t// partial line noise\n\t\t\t\t}\n\t\t\t};\n\t\t\tfor (;;) {\n\t\t\t\tconst { done, value } = await reader.read();\n\t\t\t\tif (done) break;\n\t\t\t\tbuffer += decoder.decode(value, { stream: true });\n\t\t\t\tconst lines = buffer.split(\"\\n\");\n\t\t\t\tbuffer = lines.pop() ?? \"\";\n\t\t\t\tfor (const line of lines) handleLine(line);\n\t\t\t}\n\t\t\t// The stream's FINAL line has no trailing newline — it stays in the buffer and holds\n\t\t\t// the terminal \"success\"/error event; dropping it misreports every completed pull.\n\t\t\thandleLine(buffer);\n\t\t\tif (errorMessage) return { ok: false, error: errorMessage };\n\t\t\treturn lastStatus === \"success\"\n\t\t\t\t? { ok: true }\n\t\t\t\t: { ok: false, error: `pull ended with: ${lastStatus || \"unknown\"}` };\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: error instanceof Error ? error.message : String(error) };\n\t\t}\n\t}\n\n\t/** EXPLICIT user action only — callers must have shown what gets deleted and confirmed. */\n\tasync remove(ref: string): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/delete`, {\n\t\t\t\tmethod: \"DELETE\",\n\t\t\t\theaders: { \"content-type\": \"application/json\" },\n\t\t\t\tbody: JSON.stringify({ model: ref }),\n\t\t\t});\n\t\t\tif (!response.ok) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror: `delete failed: HTTP ${response.status} ${await response.text().catch(() => \"\")}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\treturn { ok: true };\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: error instanceof Error ? error.message : String(error) };\n\t\t}\n\t}\n}\n"]}
+{"version":3,"file":"local-runtime.d.ts","sourceRoot":"","sources":["../../../src/core/models/local-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,YAAY,EAAS,MAAM,oBAAoB,CAAC;AAoB9D;;;;;;;;;;;GAWG;AAEH,MAAM,WAAW,kBAAkB;IAClC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,YAAY,CAAC,EAAE,QAAQ,GAAG,MAAM,GAAG,UAAU,CAAC;IAC9C,QAAQ,EAAE,OAAO,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,6FAA6F;IAC7F,WAAW,EAAE,OAAO,CAAC;IACrB,uEAAuE;IACvE,cAAc,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,mBAAmB;IACnC,IAAI,EAAE,MAAM,CAAC;IACb,SAAS,EAAE,MAAM,CAAC;CAClB;AAED,qGAAmG;AACnG,eAAO,MAAM,qBAAqB,WAAW,CAAC;AAE9C,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,QAAQ,GAAG,KAAK,CAAC;AAE3D,MAAM,WAAW,kBAAkB;IAClC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,eAAe,CAAC;CACtB;AAED;;;;;;;GAOG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,kBAAkB,GAAG,SAAS,CAcrG;AAED,MAAM,WAAW,gBAAgB;IAChC,OAAO,CAAC,EAAE,OAAO,KAAK,CAAC;IACvB,OAAO,CAAC,EAAE,CACT,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC,UAAU,CAAA;KAAE,KAC/B,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,MAAM,GAAG,OAAO,GAAG,IAAI,CAAC,CAAC;IACzD,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;IACrC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IACxC,uGAAqG;IACrG,QAAQ,CAAC,EAAE,MAAM,MAAM,CAAC;IACxB,IAAI,CAAC,EAAE,MAAM,MAAM,CAAC;IACpB;;+EAE2E;IAC3E,oBAAoB,CAAC,EAAE,MAAM,MAAM,CAAC,eAAe,CAAC;IACpD,gGAAgG;IAChG,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,OAAO,CAAC;IAC1C;;qDAEiD;IACjD,cAAc,CAAC,EAAE,CAChB,KAAK,EAAE,MAAM,CAAC,cAAc,EAC5B,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,eAAe,KACjB,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CAC9C;AAOD,qBAAa,aAAa;IACzB,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA2C;IAClE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA4B;IACpD,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAgC;IACvD,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAe;IACzC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IACrC,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAA6C;IACnF,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA+B;IAC3D,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAiD;IACzF,OAAO,CAAC,MAAM,CAAkE;IAEhF,YAAY,IAAI,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,gBAAgB,CAAA;KAAE,EA0BhF;IAED,IAAI,OAAO,IAAI,MAAM,CAEpB;IAED,cAAc,IAAI,MAAM,CAEvB;IAED,OAAO,CAAC,WAAW;YAaL,SAAS;IAWjB,MAAM,IAAI,OAAO,CAAC,kBAAkB,CAAC,CAW1C;IAED;;;;;OAKG;IACH,YAAY,IAAI,MAAM,EAAE,CASvB;IAED;;;8CAG0C;IAC1C,OAAO,CAAC,mBAAmB;IAM3B;;;;;;;OAOG;IACG,cAAc,CAAC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CA+BpG;YAKa,eAAe;YAef,WAAW;YAqCX,WAAW;YA2BX,aAAa;IAiB3B;;;;OAIG;IACG,KAAK,IAAI,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAc3D;IAED;;;;;OAKG;IACG,kBAAkB,IAAI,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAYxE;IAED,yFAAyF;IACzF,IAAI,IAAI;QAAE,OAAO,EAAE,OAAO,CAAA;KAAE,CAS3B;IAEK,IAAI,IAAI,OAAO,CAAC,mBAAmB,EAAE,CAAC,CAO3C;IAED,qFAAqF;IAC/E,IAAI,CAAC,GAAG,EAAE,MAAM,EAAE,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,IAAI,GAAG,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAiDvG;IAED,6FAA2F;IACrF,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;QAAE,EAAE,EAAE,OAAO,CAAC;QAAC,KAAK,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAiBlE;CACD","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport { existsSync, mkdirSync } from \"node:fs\";\nimport { homedir, arch as osArch, platform as osPlatform } from \"node:os\";\nimport { delimiter, join } from \"node:path\";\nimport type { Readable, Writable } from \"node:stream\";\nimport * as nodeZlib from \"node:zlib\";\nimport { spawnProcess, spawnProcessSync, waitForChildProcess } from \"../../utils/child-process.ts\";\n\n/**\n * `spawnProcess(..., { stdio: [\"pipe\", \"pipe\", \"pipe\"] })` always yields a non-null `stdin` at\n * runtime, but child-process.ts's typed overload only narrows `.stdin` to non-null for the\n * capture-only (`StdioNull, StdioPipe, StdioPipe`) shape, so the generic overload's `Writable |\n * null` leaks through here. This makes the true invariant explicit and fails loudly (instead of\n * silently misbehaving) if it's ever violated by a future stdio change.\n */\nfunction requireStdin(proc: ChildProcess, label: string): Writable {\n\tif (!proc.stdin) throw new Error(`${label}: no stdin pipe`);\n\treturn proc.stdin;\n}\n\n/**\n * Local model runtime manager (local-model-lifecycle-design.md): Ollama first, interface kept\n * runtime-agnostic. Pi SPAWNS the serve process itself with OWNED model storage\n * (`OLLAMA_MODELS=<agentDir>/models/ollama`) so every downloaded weight lives inside pi's tree —\n * per-model disk accounting is trivial and full cleanup is one directory. If a system server is\n * already running, pi uses it instead of double-installing, with the honest tradeoff surfaced:\n * storage then lives in the system daemon's dir and removal control is limited to `ollama rm`.\n *\n * Hard boundaries (design \"Hard boundaries\"): lifecycle actions are USER commands only — this\n * module is never exposed as a model-invokable tool; install is GUIDE MODE (exact manual steps,\n * never `curl | sh`); removal is explicit, disclosed, and never automatic.\n */\n\nexport interface LocalRuntimeStatus {\n\tbinaryPath?: string;\n\tbinarySource?: \"system\" | \"user\" | \"pi-owned\";\n\tserverUp: boolean;\n\tserverUrl: string;\n\t/** True when the responding server is a child process pi spawned (owned storage applies). */\n\tmanagedByPi: boolean;\n\t/** Owned weights directory (only meaningful for pi-managed serves). */\n\townedModelsDir: string;\n}\n\nexport interface InstalledLocalModel {\n\tname: string;\n\tsizeBytes: number;\n}\n\n/** Pinned ollama release for the managed installer below. Bump here when needed — one constant. */\nexport const OLLAMA_PINNED_VERSION = \"0.31.1\";\n\nexport type OllamaAssetKind = \"tar-zst\" | \"tar-gz\" | \"zip\";\n\nexport interface OllamaReleaseAsset {\n\tname: string;\n\tkind: OllamaAssetKind;\n}\n\n/**\n * Maps a platform/arch pair (node's `os.platform()`/`os.arch()`) to the exact ollama release asset\n * name for {@link OLLAMA_PINNED_VERSION} — verified against the real ollama/ollama GitHub release\n * and its own install.sh, not guessed: darwin ships a CLI `.tgz` (the `.app`/`.dmg`/`Ollama-*.zip`\n * assets are the separate GUI-app installer, not what we want), linux ships `.tar.zst` per arch,\n * windows ships `.zip` per arch. Pure and exported so it's independently testable and reusable\n * (e.g. by a future doctor-driven managed install) without touching OllamaRuntime.\n */\nexport function resolveOllamaAsset(plat: string, architecture: string): OllamaReleaseAsset | undefined {\n\tif (plat === \"darwin\") {\n\t\t// Universal binary — one asset for both arm64 and x64.\n\t\treturn { name: \"ollama-darwin.tgz\", kind: \"tar-gz\" };\n\t}\n\tif (plat === \"linux\") {\n\t\tconst archName = architecture === \"arm64\" ? \"arm64\" : \"amd64\";\n\t\treturn { name: `ollama-linux-${archName}.tar.zst`, kind: \"tar-zst\" };\n\t}\n\tif (plat === \"win32\") {\n\t\tconst archName = architecture === \"arm64\" ? \"arm64\" : \"amd64\";\n\t\treturn { name: `ollama-windows-${archName}.zip`, kind: \"zip\" };\n\t}\n\treturn undefined;\n}\n\nexport interface LocalRuntimeDeps {\n\tfetchFn?: typeof fetch;\n\tspawnFn?: (\n\t\tcommand: string,\n\t\targs: string[],\n\t\toptions: { env: NodeJS.ProcessEnv },\n\t) => Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\">;\n\texistsFn?: (path: string) => boolean;\n\tenvPath?: string;\n\thomeDir?: string;\n\tsleepFn?: (ms: number) => Promise<void>;\n\t/** os.platform()/os.arch() equivalents — injectable so asset resolution is testable per platform. */\n\tplatform?: () => string;\n\tarch?: () => string;\n\t/** Node's built-in zstd decompress transform, when THIS runtime's node:zlib has it — undefined\n\t * forces the system-zstd fallback. Injectable so tests can force either path deterministically\n\t * instead of depending on whatever Node happens to be running the test. */\n\tcreateZstdDecompress?: () => NodeJS.ReadWriteStream;\n\t/** Whether a named command exists on PATH (e.g. system `zstd`), for the extraction fallback. */\n\thasCommand?: (command: string) => boolean;\n\t/** Runs the extraction step for a downloaded archive. Injectable so installManaged's\n\t * download->extract->verify orchestration is testable without a real tar/zstd/unzip pipeline;\n\t * defaults to the real spawn-based extractor. */\n\textractArchive?: (\n\t\tinput: NodeJS.ReadableStream,\n\t\tdestDir: string,\n\t\tkind: OllamaAssetKind,\n\t) => Promise<{ ok: boolean; error?: string }>;\n}\n\nconst DEFAULT_BASE_URL = \"http://127.0.0.1:11434\";\nconst HEALTH_TIMEOUT_MS = 2_000;\nconst START_POLL_ATTEMPTS = 20;\nconst START_POLL_INTERVAL_MS = 500;\n\nexport class OllamaRuntime {\n\tprivate readonly _agentDir: string;\n\tprivate readonly _baseUrl: string;\n\tprivate readonly _fetch: typeof fetch;\n\tprivate readonly _spawn: NonNullable<LocalRuntimeDeps[\"spawnFn\"]>;\n\tprivate readonly _exists: (path: string) => boolean;\n\tprivate readonly _envPath: string;\n\tprivate readonly _homeDir: string;\n\tprivate readonly _sleep: (ms: number) => Promise<void>;\n\tprivate readonly _platform: () => string;\n\tprivate readonly _arch: () => string;\n\tprivate readonly _createZstdDecompress: (() => NodeJS.ReadWriteStream) | undefined;\n\tprivate readonly _hasCommand: (command: string) => boolean;\n\tprivate readonly _extractArchiveOverride: LocalRuntimeDeps[\"extractArchive\"] | undefined;\n\tprivate _child: Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\"> | undefined;\n\n\tconstructor(args: { agentDir: string; baseUrl?: string; deps?: LocalRuntimeDeps }) {\n\t\tthis._agentDir = args.agentDir;\n\t\tthis._baseUrl = (args.baseUrl ?? DEFAULT_BASE_URL).replace(/\\/$/, \"\");\n\t\tthis._fetch = args.deps?.fetchFn ?? fetch;\n\t\tthis._spawn = args.deps?.spawnFn ?? ((command, argv, options) => spawn(command, argv, options));\n\t\tthis._exists = args.deps?.existsFn ?? existsSync;\n\t\tthis._envPath = args.deps?.envPath ?? process.env.PATH ?? \"\";\n\t\tthis._homeDir = args.deps?.homeDir ?? homedir();\n\t\tthis._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));\n\t\tthis._platform = args.deps?.platform ?? osPlatform;\n\t\tthis._arch = args.deps?.arch ?? osArch;\n\t\t// Feature-detect (not version-sniff): recent Node added zstd natively to node:zlib. Verified\n\t\t// present on this repo's minimum supported Node (>=22.19.0); still detected at runtime rather\n\t\t// than assumed, so an older/different Node correctly falls through to the system-zstd path.\n\t\t// `\"createZstdDecompress\" in deps` (rather than `??`) so a test can force the fallback path by\n\t\t// passing the key as explicitly undefined, distinct from omitting it (real auto-detection).\n\t\tthis._createZstdDecompress =\n\t\t\targs.deps && \"createZstdDecompress\" in args.deps\n\t\t\t\t? args.deps.createZstdDecompress\n\t\t\t\t: typeof nodeZlib.createZstdDecompress === \"function\"\n\t\t\t\t\t? () => nodeZlib.createZstdDecompress()\n\t\t\t\t\t: undefined;\n\t\tthis._hasCommand =\n\t\t\targs.deps?.hasCommand ??\n\t\t\t((command) => spawnProcessSync(command, [\"--version\"], { encoding: \"utf8\" }).error === undefined);\n\t\tthis._extractArchiveOverride = args.deps?.extractArchive;\n\t}\n\n\tget baseUrl(): string {\n\t\treturn this._baseUrl;\n\t}\n\n\townedModelsDir(): string {\n\t\treturn join(this._agentDir, \"models\", \"ollama\");\n\t}\n\n\tprivate _findBinary(): { path: string; source: \"system\" | \"user\" | \"pi-owned\" } | undefined {\n\t\tconst piOwned = join(this._agentDir, \"runtimes\", \"ollama\", \"bin\", \"ollama\");\n\t\tif (this._exists(piOwned)) return { path: piOwned, source: \"pi-owned\" };\n\t\tconst userLevel = join(this._homeDir, \".local\", \"share\", \"ollama-dist\", \"bin\", \"ollama\");\n\t\tif (this._exists(userLevel)) return { path: userLevel, source: \"user\" };\n\t\tfor (const dir of this._envPath.split(delimiter)) {\n\t\t\tif (!dir) continue;\n\t\t\tconst candidate = join(dir, \"ollama\");\n\t\t\tif (this._exists(candidate)) return { path: candidate, source: \"system\" };\n\t\t}\n\t\treturn undefined;\n\t}\n\n\tprivate async _serverUp(): Promise<boolean> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/tags`, {\n\t\t\t\tsignal: AbortSignal.timeout(HEALTH_TIMEOUT_MS),\n\t\t\t});\n\t\t\treturn response.ok;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tasync detect(): Promise<LocalRuntimeStatus> {\n\t\tconst binary = this._findBinary();\n\t\tconst serverUp = await this._serverUp();\n\t\treturn {\n\t\t\tbinaryPath: binary?.path,\n\t\t\tbinarySource: binary?.source,\n\t\t\tserverUp,\n\t\t\tserverUrl: this._baseUrl,\n\t\t\tmanagedByPi: this._child !== undefined && serverUp,\n\t\t\townedModelsDir: this.ownedModelsDir(),\n\t\t};\n\t}\n\n\t/**\n\t * Manual-steps fallback text — shown when the managed install below wasn't offered (headless),\n\t * was declined, or itself failed. #31 reversed the stance for ollama specifically: pi CAN install\n\t * it for you now (consent-gated, a controlled download into pi's own runtimes dir, never\n\t * `curl | sh`) — these steps are the alternative for when that path isn't taken.\n\t */\n\tinstallGuide(): string[] {\n\t\treturn [\n\t\t\t\"Ollama is not installed. Pi can install it for you now (asks first, never curl|sh) — or follow these manual steps yourself:\",\n\t\t\t\"  1. Download the pinned release archive for your platform:\",\n\t\t\t\"     https://github.com/ollama/ollama/releases (asset: ollama-linux-amd64.tar.zst or your platform)\",\n\t\t\t`  2. Extract it to ${join(this._homeDir, \".local\", \"share\", \"ollama-dist\")}`,\n\t\t\t\"  3. Re-run your /models command - pi detects the binary automatically.\",\n\t\t\t\"Alternatively install system-wide from https://ollama.com/download and pi will use the system server.\",\n\t\t];\n\t}\n\n\t/** Which decompressor to use for a `.tar.zst` asset — native (this Node's own zlib) is always\n\t * preferred when available, even if a system `zstd` ALSO exists, since it needs no external\n\t * dependency at all. Verified per-runtime by feature detection (see the constructor), not by\n\t * guessing from a Node version number. */\n\tprivate _chooseZstdStrategy(): { kind: \"native\" } | { kind: \"system\" } | { kind: \"unavailable\" } {\n\t\tif (this._createZstdDecompress) return { kind: \"native\" };\n\t\tif (this._hasCommand(\"zstd\")) return { kind: \"system\" };\n\t\treturn { kind: \"unavailable\" };\n\t}\n\n\t/**\n\t * Managed install (#31): a consent-gated, controlled download of the pinned ollama release into\n\t * pi's OWN runtimes dir (`<agentDir>/runtimes/ollama/`) — never `curl | sh`, never a second\n\t * server instance, and distinct from {@link start}'s owned MODEL storage (this is the binary\n\t * itself). Callers (the router's consent flow, and later a doctor-driven install) own the\n\t * consent/UI; this method only does the mechanical download+extract+verify and reports the\n\t * outcome honestly. `onProgress` is best-effort UI feedback, not a completion signal.\n\t */\n\tasync installManaged(onProgress?: (status: string) => void): Promise<{ ok: boolean; error?: string }> {\n\t\tconst asset = resolveOllamaAsset(this._platform(), this._arch());\n\t\tif (!asset) return { ok: false, error: \"unsupported-platform\" };\n\n\t\tonProgress?.(`Downloading ${asset.name}…`);\n\t\tconst downloadUrl = `https://github.com/ollama/ollama/releases/download/v${OLLAMA_PINNED_VERSION}/${asset.name}`;\n\t\tlet response: Response;\n\t\ttry {\n\t\t\tresponse = await this._fetch(downloadUrl);\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: `download-fail: ${error instanceof Error ? error.message : String(error)}` };\n\t\t}\n\t\tif (!response.ok || !response.body) {\n\t\t\treturn { ok: false, error: `download-fail: HTTP ${response.status}` };\n\t\t}\n\n\t\tconst destDir = join(this._agentDir, \"runtimes\", \"ollama\");\n\t\tmkdirSync(destDir, { recursive: true });\n\n\t\tonProgress?.(`Extracting ${asset.name}…`);\n\t\tconst extract = this._extractArchiveOverride ?? ((input, dest, kind) => this._extractArchive(input, dest, kind));\n\t\tconst extracted = await extract(response.body as unknown as NodeJS.ReadableStream, destDir, asset.kind);\n\t\tif (!extracted.ok) return extracted;\n\n\t\tconst binaryName = this._platform() === \"win32\" ? \"ollama.exe\" : \"ollama\";\n\t\tconst binaryPath = join(destDir, \"bin\", binaryName);\n\t\tif (!this._exists(binaryPath)) {\n\t\t\treturn { ok: false, error: \"extract-fail: binary not found after extraction\" };\n\t\t}\n\t\tonProgress?.(\"Ollama installed.\");\n\t\treturn { ok: true };\n\t}\n\n\t/** Real extraction: `tar-gz` lets `tar` itself gunzip; `tar-zst` decompresses first (native\n\t * node:zlib preferred, system `zstd` as fallback, an honest error if neither exists — see\n\t * {@link _chooseZstdStrategy}); `zip` needs a seekable file, so it's buffered to disk first. */\n\tprivate async _extractArchive(\n\t\tinput: NodeJS.ReadableStream,\n\t\tdestDir: string,\n\t\tkind: OllamaAssetKind,\n\t): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\tif (kind === \"zip\") {\n\t\t\t\treturn await this._extractZip(input, destDir);\n\t\t\t}\n\t\t\treturn await this._extractTar(input, destDir, kind);\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: `extract-fail: ${error instanceof Error ? error.message : String(error)}` };\n\t\t}\n\t}\n\n\tprivate async _extractTar(\n\t\tinput: NodeJS.ReadableStream,\n\t\tdestDir: string,\n\t\tkind: \"tar-gz\" | \"tar-zst\",\n\t): Promise<{ ok: boolean; error?: string }> {\n\t\tlet decompressed: Readable = input as unknown as Readable;\n\t\tconst tarArgs = [\"-xf\", \"-\", \"-C\", destDir];\n\t\tif (kind === \"tar-zst\") {\n\t\t\tconst strategy = this._chooseZstdStrategy();\n\t\t\tif (strategy.kind === \"unavailable\") {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror: \"zstd-missing: this archive needs zstd to decompress and neither Node's built-in support nor a system `zstd` binary was found. Install zstd (e.g. `apt-get install zstd`, `dnf install zstd`, `pacman -S zstd`) and try again.\",\n\t\t\t\t};\n\t\t\t}\n\t\t\tif (strategy.kind === \"native\") {\n\t\t\t\tdecompressed = decompressed.pipe(\n\t\t\t\t\tthis._createZstdDecompress?.() as unknown as NodeJS.ReadWriteStream & Readable,\n\t\t\t\t);\n\t\t\t} else {\n\t\t\t\tconst zstdProc = spawnProcess(\"zstd\", [\"-d\"], { stdio: [\"pipe\", \"pipe\", \"pipe\"] });\n\t\t\t\tdecompressed.pipe(requireStdin(zstdProc, \"zstd\"));\n\t\t\t\tdecompressed = zstdProc.stdout as unknown as Readable;\n\t\t\t}\n\t\t} else {\n\t\t\t// tar itself handles gzip via -z — no separate decompression step needed.\n\t\t\ttarArgs[0] = \"-xzf\";\n\t\t}\n\t\tconst tarProc = spawnProcess(\"tar\", tarArgs, { stdio: [\"pipe\", \"pipe\", \"pipe\"] });\n\t\tdecompressed.pipe(requireStdin(tarProc, \"tar\"));\n\t\tconst code = await waitForChildProcess(tarProc);\n\t\tif (code !== 0) {\n\t\t\treturn { ok: false, error: `extract-fail: tar exited with code ${code}` };\n\t\t}\n\t\treturn { ok: true };\n\t}\n\n\tprivate async _extractZip(input: NodeJS.ReadableStream, destDir: string): Promise<{ ok: boolean; error?: string }> {\n\t\t// Zip's central directory needs seekable file access — buffer to a temp file first.\n\t\tconst { createWriteStream } = await import(\"node:fs\");\n\t\tconst { pipeline } = await import(\"node:stream/promises\");\n\t\tconst zipPath = join(destDir, \"..\", `ollama-download-${process.pid}-${Date.now()}.zip`);\n\t\tawait pipeline(input as unknown as Readable, createWriteStream(zipPath));\n\t\ttry {\n\t\t\tconst extractCommand = this._platform() === \"win32\" && this._hasCommand(\"tar\") ? \"tar\" : \"unzip\";\n\t\t\tconst args = extractCommand === \"tar\" ? [\"-xf\", zipPath, \"-C\", destDir] : [\"-q\", zipPath, \"-d\", destDir];\n\t\t\tconst proc = spawnProcess(extractCommand, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\t\tconst code = await waitForChildProcess(proc);\n\t\t\tif (code !== 0) {\n\t\t\t\treturn { ok: false, error: `extract-fail: ${extractCommand} exited with code ${code}` };\n\t\t\t}\n\t\t\treturn { ok: true };\n\t\t} finally {\n\t\t\ttry {\n\t\t\t\tconst { rmSync } = await import(\"node:fs\");\n\t\t\t\trmSync(zipPath, { force: true });\n\t\t\t} catch {\n\t\t\t\t// best-effort cleanup\n\t\t\t}\n\t\t}\n\t}\n\n\t/** Shared spawn-then-health-poll for both start modes below; `extraEnv` is the only thing that\n\t * differs between them (owned storage vs reusing the user's own). */\n\tprivate async _spawnAndPoll(\n\t\tbinary: { path: string },\n\t\textraEnv: NodeJS.ProcessEnv,\n\t): Promise<{ started: boolean; reason: string }> {\n\t\tconst host = this._baseUrl.replace(/^https?:\\/\\//, \"\");\n\t\tthis._child = this._spawn(binary.path, [\"serve\"], {\n\t\t\tenv: { ...process.env, OLLAMA_HOST: host, ...extraEnv },\n\t\t});\n\t\tthis._child.unref?.();\n\t\tfor (let attempt = 0; attempt < START_POLL_ATTEMPTS; attempt++) {\n\t\t\tif (await this._serverUp()) return { started: true, reason: \"started\" };\n\t\t\tawait this._sleep(START_POLL_INTERVAL_MS);\n\t\t}\n\t\tthis.stop();\n\t\treturn { started: false, reason: \"health_check_timeout\" };\n\t}\n\n\t/**\n\t * Start a pi-managed serve with OWNED storage and the hardened env verified on this class of\n\t * hardware. No-op (reported) when a server already responds — pi never double-serves. Used by\n\t * `/models add` et al, where isolated per-model-pull storage is the point.\n\t */\n\tasync start(): Promise<{ started: boolean; reason: string }> {\n\t\tif (await this._serverUp()) {\n\t\t\treturn { started: false, reason: this._child ? \"already_running_managed\" : \"already_running_system\" };\n\t\t}\n\t\tconst binary = this._findBinary();\n\t\tif (!binary) return { started: false, reason: \"binary_missing\" };\n\t\tmkdirSync(this.ownedModelsDir(), { recursive: true });\n\t\treturn this._spawnAndPoll(binary, {\n\t\t\tOLLAMA_MODELS: this.ownedModelsDir(),\n\t\t\tOLLAMA_FLASH_ATTENTION: \"1\",\n\t\t\tOLLAMA_KV_CACHE_TYPE: \"q8_0\",\n\t\t\tOLLAMA_NUM_PARALLEL: \"1\",\n\t\t\tOLLAMA_MAX_LOADED_MODELS: \"3\",\n\t\t});\n\t}\n\n\t/**\n\t * Start a serve that REUSES the user's own existing models directory (no `OLLAMA_MODELS`\n\t * override — falls through to Ollama's own default, `~/.ollama`), for callers that must see the\n\t * user's already-pulled models rather than pi's isolated/owned storage. Same idempotency and\n\t * hardened perf env as {@link start}; the only difference is which storage the server sees.\n\t */\n\tasync startReuseExisting(): Promise<{ started: boolean; reason: string }> {\n\t\tif (await this._serverUp()) {\n\t\t\treturn { started: false, reason: this._child ? \"already_running_managed\" : \"already_running_system\" };\n\t\t}\n\t\tconst binary = this._findBinary();\n\t\tif (!binary) return { started: false, reason: \"binary_missing\" };\n\t\treturn this._spawnAndPoll(binary, {\n\t\t\tOLLAMA_FLASH_ATTENTION: \"1\",\n\t\t\tOLLAMA_KV_CACHE_TYPE: \"q8_0\",\n\t\t\tOLLAMA_NUM_PARALLEL: \"1\",\n\t\t\tOLLAMA_MAX_LOADED_MODELS: \"3\",\n\t\t});\n\t}\n\n\t/** Resource hygiene only: stops the pi-managed serve process; never deletes anything. */\n\tstop(): { stopped: boolean } {\n\t\tif (!this._child) return { stopped: false };\n\t\ttry {\n\t\t\tthis._child.kill(\"SIGTERM\");\n\t\t} catch {\n\t\t\t// already gone\n\t\t}\n\t\tthis._child = undefined;\n\t\treturn { stopped: true };\n\t}\n\n\tasync list(): Promise<InstalledLocalModel[]> {\n\t\tconst response = await this._fetch(`${this._baseUrl}/api/tags`, { signal: AbortSignal.timeout(10_000) });\n\t\tif (!response.ok) throw new Error(`ollama list failed: HTTP ${response.status}`);\n\t\tconst data = (await response.json()) as { models?: Array<{ name?: string; size?: number }> };\n\t\treturn (data.models ?? [])\n\t\t\t.filter((model): model is { name: string; size?: number } => typeof model.name === \"string\")\n\t\t\t.map((model) => ({ name: model.name, sizeBytes: model.size ?? 0 }));\n\t}\n\n\t/** Pull a model through the server API (weights land in the SERVER's models dir). */\n\tasync pull(ref: string, onProgress?: (status: string) => void): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/pull`, {\n\t\t\t\tmethod: \"POST\",\n\t\t\t\theaders: { \"content-type\": \"application/json\" },\n\t\t\t\tbody: JSON.stringify({ model: ref, stream: true }),\n\t\t\t});\n\t\t\tif (!response.ok || !response.body) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror: `pull failed: HTTP ${response.status} ${await response.text().catch(() => \"\")}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\tconst reader = response.body.getReader();\n\t\t\tconst decoder = new TextDecoder();\n\t\t\tlet buffer = \"\";\n\t\t\tlet lastStatus = \"\";\n\t\t\tlet errorMessage: string | undefined;\n\t\t\tconst handleLine = (line: string): void => {\n\t\t\t\tif (!line.trim()) return;\n\t\t\t\ttry {\n\t\t\t\t\tconst event = JSON.parse(line) as { status?: string; error?: string };\n\t\t\t\t\tif (event.error) errorMessage = event.error;\n\t\t\t\t\tif (event.status && event.status !== lastStatus) {\n\t\t\t\t\t\tlastStatus = event.status;\n\t\t\t\t\t\tonProgress?.(event.status);\n\t\t\t\t\t}\n\t\t\t\t} catch {\n\t\t\t\t\t// partial line noise\n\t\t\t\t}\n\t\t\t};\n\t\t\tfor (;;) {\n\t\t\t\tconst { done, value } = await reader.read();\n\t\t\t\tif (done) break;\n\t\t\t\tbuffer += decoder.decode(value, { stream: true });\n\t\t\t\tconst lines = buffer.split(\"\\n\");\n\t\t\t\tbuffer = lines.pop() ?? \"\";\n\t\t\t\tfor (const line of lines) handleLine(line);\n\t\t\t}\n\t\t\t// The stream's FINAL line has no trailing newline — it stays in the buffer and holds\n\t\t\t// the terminal \"success\"/error event; dropping it misreports every completed pull.\n\t\t\thandleLine(buffer);\n\t\t\tif (errorMessage) return { ok: false, error: errorMessage };\n\t\t\treturn lastStatus === \"success\"\n\t\t\t\t? { ok: true }\n\t\t\t\t: { ok: false, error: `pull ended with: ${lastStatus || \"unknown\"}` };\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: error instanceof Error ? error.message : String(error) };\n\t\t}\n\t}\n\n\t/** EXPLICIT user action only — callers must have shown what gets deleted and confirmed. */\n\tasync remove(ref: string): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${this._baseUrl}/api/delete`, {\n\t\t\t\tmethod: \"DELETE\",\n\t\t\t\theaders: { \"content-type\": \"application/json\" },\n\t\t\t\tbody: JSON.stringify({ model: ref }),\n\t\t\t});\n\t\t\tif (!response.ok) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror: `delete failed: HTTP ${response.status} ${await response.text().catch(() => \"\")}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\treturn { ok: true };\n\t\t} catch (error) {\n\t\t\treturn { ok: false, error: error instanceof Error ? error.message : String(error) };\n\t\t}\n\t}\n}\n"]}

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

@@ -1,7 +1,46 @@
 import { spawn } from "node:child_process";
 import { existsSync, mkdirSync } from "node:fs";
-import { homedir } from "node:os";
+import { homedir, arch as osArch, platform as osPlatform } from "node:os";
 import { delimiter, join } from "node:path";
+import * as nodeZlib from "node:zlib";
+import { spawnProcess, spawnProcessSync, waitForChildProcess } from "../../utils/child-process.js";
+/**
+ * `spawnProcess(..., { stdio: ["pipe", "pipe", "pipe"] })` always yields a non-null `stdin` at
+ * runtime, but child-process.ts's typed overload only narrows `.stdin` to non-null for the
+ * capture-only (`StdioNull, StdioPipe, StdioPipe`) shape, so the generic overload's `Writable |
+ * null` leaks through here. This makes the true invariant explicit and fails loudly (instead of
+ * silently misbehaving) if it's ever violated by a future stdio change.
+ */
+function requireStdin(proc, label) {
+    if (!proc.stdin)
+        throw new Error(`${label}: no stdin pipe`);
+    return proc.stdin;
+}
+/** Pinned ollama release for the managed installer below. Bump here when needed — one constant. */
+export const OLLAMA_PINNED_VERSION = "0.31.1";
+/**
+ * Maps a platform/arch pair (node's `os.platform()`/`os.arch()`) to the exact ollama release asset
+ * name for {@link OLLAMA_PINNED_VERSION} — verified against the real ollama/ollama GitHub release
+ * and its own install.sh, not guessed: darwin ships a CLI `.tgz` (the `.app`/`.dmg`/`Ollama-*.zip`
+ * assets are the separate GUI-app installer, not what we want), linux ships `.tar.zst` per arch,
+ * windows ships `.zip` per arch. Pure and exported so it's independently testable and reusable
+ * (e.g. by a future doctor-driven managed install) without touching OllamaRuntime.
+ */
+export function resolveOllamaAsset(plat, architecture) {
+    if (plat === "darwin") {
+        // Universal binary — one asset for both arm64 and x64.
+        return { name: "ollama-darwin.tgz", kind: "tar-gz" };
+    }
+    if (plat === "linux") {
+        const archName = architecture === "arm64" ? "arm64" : "amd64";
+        return { name: `ollama-linux-${archName}.tar.zst`, kind: "tar-zst" };
+    }
+    if (plat === "win32") {
+        const archName = architecture === "arm64" ? "arm64" : "amd64";
+        return { name: `ollama-windows-${archName}.zip`, kind: "zip" };
+    }
+    return undefined;
+}
 const DEFAULT_BASE_URL = "http://127.0.0.1:11434";
 const HEALTH_TIMEOUT_MS = 2_000;
 const START_POLL_ATTEMPTS = 20;
@@ -15,6 +54,11 @@ export class OllamaRuntime {
     _envPath;
     _homeDir;
     _sleep;
+    _platform;
+    _arch;
+    _createZstdDecompress;
+    _hasCommand;
+    _extractArchiveOverride;
     _child;
     constructor(args) {
         this._agentDir = args.agentDir;
@@ -25,6 +69,23 @@ export class OllamaRuntime {
         this._envPath = args.deps?.envPath ?? process.env.PATH ?? "";
         this._homeDir = args.deps?.homeDir ?? homedir();
         this._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));
+        this._platform = args.deps?.platform ?? osPlatform;
+        this._arch = args.deps?.arch ?? osArch;
+        // Feature-detect (not version-sniff): recent Node added zstd natively to node:zlib. Verified
+        // present on this repo's minimum supported Node (>=22.19.0); still detected at runtime rather
+        // than assumed, so an older/different Node correctly falls through to the system-zstd path.
+        // `"createZstdDecompress" in deps` (rather than `??`) so a test can force the fallback path by
+        // passing the key as explicitly undefined, distinct from omitting it (real auto-detection).
+        this._createZstdDecompress =
+            args.deps && "createZstdDecompress" in args.deps
+                ? args.deps.createZstdDecompress
+                : typeof nodeZlib.createZstdDecompress === "function"
+                    ? () => nodeZlib.createZstdDecompress()
+                    : undefined;
+        this._hasCommand =
+            args.deps?.hasCommand ??
+                ((command) => spawnProcessSync(command, ["--version"], { encoding: "utf8" }).error === undefined);
+        this._extractArchiveOverride = args.deps?.extractArchive;
     }
     get baseUrl() {
         return this._baseUrl;
@@ -71,10 +132,15 @@ export class OllamaRuntime {
             ownedModelsDir: this.ownedModelsDir(),
         };
     }
-    /** GUIDE MODE: exact manual steps, printed, never executed (no `curl | sh`, no sudo). */
+    /**
+     * Manual-steps fallback text — shown when the managed install below wasn't offered (headless),
+     * was declined, or itself failed. #31 reversed the stance for ollama specifically: pi CAN install
+     * it for you now (consent-gated, a controlled download into pi's own runtimes dir, never
+     * `curl | sh`) — these steps are the alternative for when that path isn't taken.
+     */
     installGuide() {
         return [
-            "Ollama is not installed. Pi never runs installers itself — manual steps (user-level, no sudo):",
+            "Ollama is not installed. Pi can install it for you now (asks first, never curl|sh) — or follow these manual steps yourself:",
             "  1. Download the pinned release archive for your platform:",
             "     https://github.com/ollama/ollama/releases (asset: ollama-linux-amd64.tar.zst or your platform)",
             `  2. Extract it to ${join(this._homeDir, ".local", "share", "ollama-dist")}`,
@@ -82,29 +148,134 @@ export class OllamaRuntime {
             "Alternatively install system-wide from https://ollama.com/download and pi will use the system server.",
         ];
     }
+    /** Which decompressor to use for a `.tar.zst` asset — native (this Node's own zlib) is always
+     * preferred when available, even if a system `zstd` ALSO exists, since it needs no external
+     * dependency at all. Verified per-runtime by feature detection (see the constructor), not by
+     * guessing from a Node version number. */
+    _chooseZstdStrategy() {
+        if (this._createZstdDecompress)
+            return { kind: "native" };
+        if (this._hasCommand("zstd"))
+            return { kind: "system" };
+        return { kind: "unavailable" };
+    }
     /**
-     * Start a pi-managed serve with OWNED storage and the hardened env verified on this class of
-     * hardware. No-op (reported) when a server already responds — pi never double-serves.
+     * Managed install (#31): a consent-gated, controlled download of the pinned ollama release into
+     * pi's OWN runtimes dir (`<agentDir>/runtimes/ollama/`) — never `curl | sh`, never a second
+     * server instance, and distinct from {@link start}'s owned MODEL storage (this is the binary
+     * itself). Callers (the router's consent flow, and later a doctor-driven install) own the
+     * consent/UI; this method only does the mechanical download+extract+verify and reports the
+     * outcome honestly. `onProgress` is best-effort UI feedback, not a completion signal.
      */
-    async start() {
-        if (await this._serverUp()) {
-            return { started: false, reason: this._child ? "already_running_managed" : "already_running_system" };
+    async installManaged(onProgress) {
+        const asset = resolveOllamaAsset(this._platform(), this._arch());
+        if (!asset)
+            return { ok: false, error: "unsupported-platform" };
+        onProgress?.(`Downloading ${asset.name}…`);
+        const downloadUrl = `https://github.com/ollama/ollama/releases/download/v${OLLAMA_PINNED_VERSION}/${asset.name}`;
+        let response;
+        try {
+            response = await this._fetch(downloadUrl);
         }
-        const binary = this._findBinary();
-        if (!binary)
-            return { started: false, reason: "binary_missing" };
-        mkdirSync(this.ownedModelsDir(), { recursive: true });
+        catch (error) {
+            return { ok: false, error: `download-fail: ${error instanceof Error ? error.message : String(error)}` };
+        }
+        if (!response.ok || !response.body) {
+            return { ok: false, error: `download-fail: HTTP ${response.status}` };
+        }
+        const destDir = join(this._agentDir, "runtimes", "ollama");
+        mkdirSync(destDir, { recursive: true });
+        onProgress?.(`Extracting ${asset.name}…`);
+        const extract = this._extractArchiveOverride ?? ((input, dest, kind) => this._extractArchive(input, dest, kind));
+        const extracted = await extract(response.body, destDir, asset.kind);
+        if (!extracted.ok)
+            return extracted;
+        const binaryName = this._platform() === "win32" ? "ollama.exe" : "ollama";
+        const binaryPath = join(destDir, "bin", binaryName);
+        if (!this._exists(binaryPath)) {
+            return { ok: false, error: "extract-fail: binary not found after extraction" };
+        }
+        onProgress?.("Ollama installed.");
+        return { ok: true };
+    }
+    /** Real extraction: `tar-gz` lets `tar` itself gunzip; `tar-zst` decompresses first (native
+     * node:zlib preferred, system `zstd` as fallback, an honest error if neither exists — see
+     * {@link _chooseZstdStrategy}); `zip` needs a seekable file, so it's buffered to disk first. */
+    async _extractArchive(input, destDir, kind) {
+        try {
+            if (kind === "zip") {
+                return await this._extractZip(input, destDir);
+            }
+            return await this._extractTar(input, destDir, kind);
+        }
+        catch (error) {
+            return { ok: false, error: `extract-fail: ${error instanceof Error ? error.message : String(error)}` };
+        }
+    }
+    async _extractTar(input, destDir, kind) {
+        let decompressed = input;
+        const tarArgs = ["-xf", "-", "-C", destDir];
+        if (kind === "tar-zst") {
+            const strategy = this._chooseZstdStrategy();
+            if (strategy.kind === "unavailable") {
+                return {
+                    ok: false,
+                    error: "zstd-missing: this archive needs zstd to decompress and neither Node's built-in support nor a system `zstd` binary was found. Install zstd (e.g. `apt-get install zstd`, `dnf install zstd`, `pacman -S zstd`) and try again.",
+                };
+            }
+            if (strategy.kind === "native") {
+                decompressed = decompressed.pipe(this._createZstdDecompress?.());
+            }
+            else {
+                const zstdProc = spawnProcess("zstd", ["-d"], { stdio: ["pipe", "pipe", "pipe"] });
+                decompressed.pipe(requireStdin(zstdProc, "zstd"));
+                decompressed = zstdProc.stdout;
+            }
+        }
+        else {
+            // tar itself handles gzip via -z — no separate decompression step needed.
+            tarArgs[0] = "-xzf";
+        }
+        const tarProc = spawnProcess("tar", tarArgs, { stdio: ["pipe", "pipe", "pipe"] });
+        decompressed.pipe(requireStdin(tarProc, "tar"));
+        const code = await waitForChildProcess(tarProc);
+        if (code !== 0) {
+            return { ok: false, error: `extract-fail: tar exited with code ${code}` };
+        }
+        return { ok: true };
+    }
+    async _extractZip(input, destDir) {
+        // Zip's central directory needs seekable file access — buffer to a temp file first.
+        const { createWriteStream } = await import("node:fs");
+        const { pipeline } = await import("node:stream/promises");
+        const zipPath = join(destDir, "..", `ollama-download-${process.pid}-${Date.now()}.zip`);
+        await pipeline(input, createWriteStream(zipPath));
+        try {
+            const extractCommand = this._platform() === "win32" && this._hasCommand("tar") ? "tar" : "unzip";
+            const args = extractCommand === "tar" ? ["-xf", zipPath, "-C", destDir] : ["-q", zipPath, "-d", destDir];
+            const proc = spawnProcess(extractCommand, args, { stdio: ["ignore", "pipe", "pipe"] });
+            const code = await waitForChildProcess(proc);
+            if (code !== 0) {
+                return { ok: false, error: `extract-fail: ${extractCommand} exited with code ${code}` };
+            }
+            return { ok: true };
+        }
+        finally {
+            try {
+                const { rmSync } = await import("node:fs");
+                rmSync(zipPath, { force: true });
+            }
+            catch {
+                // best-effort cleanup
+            }
+        }
+    }
+    /** Shared spawn-then-health-poll for both start modes below; `extraEnv` is the only thing that
+     * differs between them (owned storage vs reusing the user's own). */
+    async _spawnAndPoll(binary, extraEnv) {
         const host = this._baseUrl.replace(/^https?:\/\//, "");
         this._child = this._spawn(binary.path, ["serve"], {
-            env: {
-                ...process.env,
-                OLLAMA_HOST: host,
-                OLLAMA_MODELS: this.ownedModelsDir(),
-                OLLAMA_FLASH_ATTENTION: "1",
-                OLLAMA_KV_CACHE_TYPE: "q8_0",
-                OLLAMA_NUM_PARALLEL: "1",
-                OLLAMA_MAX_LOADED_MODELS: "3",
-            },
+            env: { ...process.env, OLLAMA_HOST: host, ...extraEnv },
         });
         this._child.unref?.();
         for (let attempt = 0; attempt < START_POLL_ATTEMPTS; attempt++) {
@@ -115,6 +286,47 @@ export class OllamaRuntime {
         this.stop();
         return { started: false, reason: "health_check_timeout" };
     }
+    /**
+     * Start a pi-managed serve with OWNED storage and the hardened env verified on this class of
+     * hardware. No-op (reported) when a server already responds — pi never double-serves. Used by
+     * `/models add` et al, where isolated per-model-pull storage is the point.
+     */
+    async start() {
+        if (await this._serverUp()) {
+            return { started: false, reason: this._child ? "already_running_managed" : "already_running_system" };
+        }
+        const binary = this._findBinary();
+        if (!binary)
+            return { started: false, reason: "binary_missing" };
+        mkdirSync(this.ownedModelsDir(), { recursive: true });
+        return this._spawnAndPoll(binary, {
+            OLLAMA_MODELS: this.ownedModelsDir(),
+            OLLAMA_FLASH_ATTENTION: "1",
+            OLLAMA_KV_CACHE_TYPE: "q8_0",
+            OLLAMA_NUM_PARALLEL: "1",
+            OLLAMA_MAX_LOADED_MODELS: "3",
+        });
+    }
+    /**
+     * Start a serve that REUSES the user's own existing models directory (no `OLLAMA_MODELS`
+     * override — falls through to Ollama's own default, `~/.ollama`), for callers that must see the
+     * user's already-pulled models rather than pi's isolated/owned storage. Same idempotency and
+     * hardened perf env as {@link start}; the only difference is which storage the server sees.
+     */
+    async startReuseExisting() {
+        if (await this._serverUp()) {
+            return { started: false, reason: this._child ? "already_running_managed" : "already_running_system" };
+        }
+        const binary = this._findBinary();
+        if (!binary)
+            return { started: false, reason: "binary_missing" };
+        return this._spawnAndPoll(binary, {
+            OLLAMA_FLASH_ATTENTION: "1",
+            OLLAMA_KV_CACHE_TYPE: "q8_0",
+            OLLAMA_NUM_PARALLEL: "1",
+            OLLAMA_MAX_LOADED_MODELS: "3",
+        });
+    }
     /** Resource hygiene only: stops the pi-managed serve process; never deletes anything. */
     stop() {
         if (!this._child)