@caupulican/pi-adaptative 0.81.37 → 0.81.39

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

@@ -0,0 +1 @@
+{"version":3,"file":"llamacpp-runtime.d.ts","sourceRoot":"","sources":["../../../src/core/models/llamacpp-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,YAAY,EAAS,MAAM,oBAAoB,CAAC;AAc9D,OAAO,KAAK,EAAE,QAAQ,EAAY,MAAM,aAAa,CAAC;AAMtD;;;;;;;;;;;;;;GAcG;AAEH;;;;;GAKG;AACH,eAAO,MAAM,6BAA6B,wBAAwB,CAAC;AAEnE,eAAO,MAAM,gCAAgC,+DAA+D,CAAC;AAE7G,MAAM,WAAW,oBAAoB;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,MAAM,CAAC;CACpB;AAED,mFAAmF;AACnF,eAAO,MAAM,UAAU,EAAE,oBAKxB,CAAC;AAEF,MAAM,MAAM,YAAY,GAAG,KAAK,GAAG,MAAM,CAAC;AAC1C,MAAM,MAAM,cAAc,GAAG,QAAQ,GAAG,KAAK,CAAC;AAE9C,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,cAAc,CAAC;IACrB,OAAO,EAAE,YAAY,CAAC;CACtB;AAcD;;;;;;;GAOG;AACH,wBAAgB,sBAAsB,CACrC,IAAI,EAAE,MAAM,EACZ,YAAY,EAAE,MAAM,EACpB,YAAY,EAAE,OAAO,GACnB,eAAe,GAAG,SAAS,CAqB7B;AAED,MAAM,WAAW,iBAAiB;IACjC,gBAAgB,EAAE,OAAO,CAAC;IAC1B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,mBAAmB;IACnC,EAAE,EAAE,OAAO,CAAC;IACZ,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;CACf;AAED,MAAM,MAAM,gBAAgB,GAAG;IAAE,EAAE,EAAE,IAAI,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAAG;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAQ5F,KAAK,iBAAiB,GAAG;IAAE,QAAQ,CAAC,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,QAAQ,CAAC;IAAC,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAA;CAAE,CAAC;AAC3F,KAAK,YAAY,GAAG,CACnB,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE,iBAAiB,KACtB,IAAI,CAAC,YAAY,EAAE,KAAK,GAAG,MAAM,GAAG,OAAO,GAAG,IAAI,CAAC,CAAC;AACzD,KAAK,qBAAqB,GAAG,CAC5B,KAAK,EAAE,QAAQ,EACf,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,cAAc,KAChB,OAAO,CAAC;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,KAAK,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CAAC;AAE9C,MAAM,WAAW,iBAAiB;IACjC,OAAO,CAAC,EAAE,OAAO,KAAK,CAAC;IACvB,OAAO,CAAC,EAAE,YAAY,CAAC;IACvB,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC;IACrC,OAAO,CAAC,EAAE,CAAC,EAAE,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;IACxC,kGAAgG;IAChG,UAAU,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,OAAO,CAAC;IAC1C;8FAC0F;IAC1F,YAAY,CAAC,EAAE,MAAM,OAAO,CAAC;IAC7B,QAAQ,CAAC,EAAE,MAAM,MAAM,CAAC;IACxB,IAAI,CAAC,EAAE,MAAM,MAAM,CAAC;IACpB;;4CAEwC;IACxC,cAAc,CAAC,EAAE,qBAAqB,CAAC;IACvC,+FAA+F;IAC/F,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,oBAAoB,CAAC,EAAE,MAAM,CAAC;CAC9B;AA0BD,qBAAa,oBAAoB;IAChC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAS;IACnC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAe;IACtC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA4B;IACpD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAgC;IACvD,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA+B;IAC3D,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAgB;IAC9C,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAe;IACzC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IACrC,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAoC;IAC5E,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAS;IAC7C,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAAS;IAC/C,OAAO,CAAC,MAAM,CAAkE;IAEhF,YAAY,IAAI,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,iBAAiB,CAAA;KAAE,EAiB/D;IAED,UAAU,IAAI,MAAM,CAEnB;IAED,SAAS,IAAI,MAAM,CAElB;IAED,OAAO,CAAC,WAAW;IAInB,OAAO,CAAC,aAAa;IAIrB,OAAO,CAAC,aAAa;IAoBrB,OAAO,CAAC,cAAc;IAIhB,MAAM,IAAI,OAAO,CAAC,iBAAiB,CAAC,CAUzC;IAED;;;yDAGqD;IACrD,OAAO,CAAC,kBAAkB;IAqB1B;;;;;OAKG;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,CAiCpG;YAEa,eAAe;YAYf,aAAa;YA+Bb,WAAW;YA+BX,oBAAoB;IAYlC,OAAO,CAAC,cAAc;IAQtB,OAAO,CAAC,eAAe;IAQvB;;;;;;OAMG;IACH,OAAO,CAAC,oBAAoB;IAQ5B;;;;;OAKG;IACG,aAAa,CAClB,IAAI,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,EACpC,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,MAAM,KAAK,IAAI,GACnC,OAAO,CAAC,mBAAmB,CAAC,CA0C9B;YAEa,SAAS;IAWvB;;;;OAIG;IACG,KAAK,CAAC,IAAI,EAAE;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,IAAI,EAAE,MAAM,CAAC;QACb,MAAM,EAAE,MAAM,CAAC;KACf,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAsC5B;IAED,IAAI,IAAI;QAAE,OAAO,EAAE,OAAO,CAAA;KAAE,CAS3B;IAED,SAAS,IAAI,OAAO,CAEnB;CACD","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport {\n\tcreateWriteStream,\n\ttype Dirent,\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trmSync,\n\tstatSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { arch as osArch, platform as osPlatform } from \"node:os\";\nimport { dirname, join, relative } from \"node:path\";\nimport type { Readable, Writable } from \"node:stream\";\nimport { pipeline } from \"node:stream/promises\";\nimport { spawnProcess, spawnProcessSync, waitForChildProcessWithTermination } from \"../../utils/child-process.ts\";\nimport { killProcessTree, trackDetachedChildPid, untrackDetachedChildPid } from \"../../utils/shell.ts\";\nimport { modelsDir as agentModelsDir, runtimesDir as agentRuntimesDir } from \"../agent-paths.ts\";\n\n/**\n * Managed runtime for prism-ml 1-bit GGUF models (Bonsai-27B first). The provider requires their\n * llama.cpp fork (Q1_0_g128 hybrid-attention kernels) — stock llama.cpp/Ollama cannot serve these\n * weights — so this module owns runtime install (prebuilt release download+extract), GGUF\n * downloads, and llama-server lifecycle. Mirrors local-runtime.ts's OllamaRuntime: injectable\n * seams for fetch/spawn/exists, pi-owned directories under agentDir (runtimes/, models/),\n * detached+tracked child processes, onProgress as best-effort UI feedback, honest error taxonomies\n * instead of silent fallbacks.\n *\n * The fork publishes precompiled per-platform archives on GitHub Releases (no local build): pin a\n * release tag, download the matching asset, extract it, and locate the `llama-server` binary once\n * at install time (release archive layouts vary — bin/ vs build/bin/ at the top level — so this is\n * a deterministic one-time scan, not a fallback). The found relative path and backend (cpu/cuda)\n * are persisted to a manifest so later `detect()`/`serve()` calls never need to re-scan.\n */\n\n/**\n * Pinned Prism ML llama.cpp fork release — verified via\n * `GET https://api.github.com/repos/PrismML-Eng/llama.cpp/releases/tags/prism-b9594-38c66ad` on\n * 2026-07-18 (release exists, all assets referenced below are present on it). Bump here (and\n * re-verify with the same API call) when the provider ships new kernels.\n */\nexport const PRISM_LLAMACPP_PINNED_RELEASE = \"prism-b9594-38c66ad\";\n\nexport const PRISM_LLAMACPP_RELEASES_BASE_URL = \"https://github.com/PrismML-Eng/llama.cpp/releases/download\";\n\nexport interface PrismModelDescriptor {\n\trepo: string;\n\tfile: string;\n\tmmprojFile?: string;\n\tdisplayName: string;\n}\n\n/** Curated first-model descriptor; the `/models add` wiring task consumes this. */\nexport const BONSAI_27B: PrismModelDescriptor = {\n\trepo: \"prism-ml/Bonsai-27B-gguf\",\n\tfile: \"Bonsai-27B-Q1_0.gguf\",\n\tmmprojFile: \"Bonsai-27B-mmproj-Q8_0.gguf\",\n\tdisplayName: \"Bonsai-27B (1-bit Q1_0 + vision)\",\n};\n\nexport type PrismBackend = \"cpu\" | \"cuda\";\nexport type PrismAssetKind = \"tar-gz\" | \"zip\";\n\nexport interface PrismLlamaAsset {\n\tname: string;\n\tkind: PrismAssetKind;\n\tbackend: PrismBackend;\n}\n\n// Verbatim asset names from the pinned release (see PRISM_LLAMACPP_PINNED_RELEASE doc comment) —\n// hardcoded rather than templated from the tag because the provider's naming isn't fully\n// consistent across platforms (Windows CPU assets drop the tag/commit segment entirely; the\n// Windows CUDA asset uses a different build-number segment, \"b1\" not \"b9594\").\nconst LINUX_X64_CPU_ASSET = \"llama-prism-b9594-38c66ad-bin-ubuntu-x64.tar.gz\";\nconst LINUX_ARM64_CPU_ASSET = \"llama-prism-b9594-38c66ad-bin-ubuntu-arm64.tar.gz\";\nconst LINUX_X64_CUDA_ASSET = \"llama-prism-b9594-38c66ad-bin-linux-cuda-12.4-x64.tar.gz\";\nconst MACOS_ARM64_ASSET = \"llama-prism-b9594-38c66ad-bin-macos-arm64.tar.gz\";\nconst MACOS_X64_ASSET = \"llama-prism-b9594-38c66ad-bin-macos-x64.tar.gz\";\nconst WIN_X64_CPU_ASSET = \"llama-bin-win-cpu-x64.zip\";\nconst WIN_ARM64_CPU_ASSET = \"llama-bin-win-cpu-arm64.zip\";\n\n/**\n * Maps a platform/arch/GPU triple to the exact Prism llama.cpp release asset for\n * {@link PRISM_LLAMACPP_PINNED_RELEASE} — verified against the real GitHub release, not guessed.\n * CPU asset by default; the CUDA 12.4 variant is only offered for linux x64 with an NVIDIA GPU\n * present (a 12.8 variant also exists upstream but has no wired caller yet). Windows CUDA needs a\n * companion `cudart-*` archive and is out of scope for v1 — Windows always resolves to the CPU zip\n * regardless of `hasNvidiaGpu`. Pure and exported so it's independently testable.\n */\nexport function resolvePrismLlamaAsset(\n\tplat: string,\n\tarchitecture: string,\n\thasNvidiaGpu: boolean,\n): PrismLlamaAsset | undefined {\n\tif (plat === \"linux\") {\n\t\tif (architecture === \"x64\") {\n\t\t\treturn hasNvidiaGpu\n\t\t\t\t? { name: LINUX_X64_CUDA_ASSET, kind: \"tar-gz\", backend: \"cuda\" }\n\t\t\t\t: { name: LINUX_X64_CPU_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\t}\n\t\tif (architecture === \"arm64\") return { name: LINUX_ARM64_CPU_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\tif (plat === \"darwin\") {\n\t\tif (architecture === \"arm64\") return { name: MACOS_ARM64_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\tif (architecture === \"x64\") return { name: MACOS_X64_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\tif (plat === \"win32\") {\n\t\tif (architecture === \"x64\") return { name: WIN_X64_CPU_ASSET, kind: \"zip\", backend: \"cpu\" };\n\t\tif (architecture === \"arm64\") return { name: WIN_ARM64_CPU_ASSET, kind: \"zip\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\treturn undefined;\n}\n\nexport interface PrismDetectResult {\n\truntimeInstalled: boolean;\n\tbinaryPath?: string;\n\trelease?: string;\n}\n\nexport interface PrismDownloadResult {\n\tok: boolean;\n\tpath?: string;\n\tskipped?: boolean;\n\terror?: string;\n}\n\nexport type PrismServeResult = { ok: true; baseUrl: string } | { ok: false; error: string };\n\ninterface PrismInstallManifest {\n\trelease: string;\n\tbinaryRelPath: string;\n\tbackend: PrismBackend;\n}\n\ntype PrismSpawnOptions = { detached?: boolean; stdio?: \"ignore\"; env?: NodeJS.ProcessEnv };\ntype PrismSpawnFn = (\n\tcommand: string,\n\targs: string[],\n\toptions: PrismSpawnOptions,\n) => Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\">;\ntype PrismExtractArchiveFn = (\n\tinput: Readable,\n\tdestDir: string,\n\tkind: PrismAssetKind,\n) => Promise<{ ok: boolean; error?: string }>;\n\nexport interface PrismLlamaCppDeps {\n\tfetchFn?: typeof fetch;\n\tspawnFn?: PrismSpawnFn;\n\texistsFn?: (path: string) => boolean;\n\tsleepFn?: (ms: number) => Promise<void>;\n\t/** Whether a named command exists on PATH (nvidia-smi, tar — for zip extraction on Windows). */\n\thasCommand?: (command: string) => boolean;\n\t/** Decided once at install time and persisted as `backend` — serve() reads the persisted value\n\t * so `-ngl 99` reflects what was actually installed, not the host's current GPU state. */\n\thasNvidiaGpu?: () => boolean;\n\tplatform?: () => string;\n\tarch?: () => string;\n\t/** Runs the extraction step for a downloaded archive. Injectable so installManaged's\n\t * download->extract->scan orchestration is testable without a real tar/unzip pipeline; defaults\n\t * to the real spawn-based extractor. */\n\textractArchive?: PrismExtractArchiveFn;\n\t/** Health-poll bounds for serve(); default ~120s (240 * 500ms), both overridable for tests. */\n\thealthPollAttempts?: number;\n\thealthPollIntervalMs?: number;\n}\n\nconst EXTRACTION_TIMEOUT_MS = 10 * 60_000;\nconst COMMAND_KILL_GRACE_MS = 2_000;\nconst DEFAULT_HEALTH_POLL_ATTEMPTS = 240;\nconst DEFAULT_HEALTH_POLL_INTERVAL_MS = 500;\nconst HEALTH_CHECK_TIMEOUT_MS = 2_000;\nconst HEAD_REQUEST_TIMEOUT_MS = 10_000;\n\nfunction parseContentLength(header: string | null): number | undefined {\n\tif (!header) return undefined;\n\tconst value = Number(header);\n\treturn Number.isFinite(value) && value >= 0 ? value : undefined;\n}\n\n/**\n * `spawnProcess(..., { stdio: [\"pipe\", ...] })` always yields a non-null `stdin` at runtime, but\n * child-process.ts's typed overload only narrows `.stdin` for the capture-only stdio shape used\n * elsewhere. This makes the true invariant explicit and fails loudly instead of silently\n * 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\nexport class PrismLlamaCppRuntime {\n\tprivate readonly _agentDir: string;\n\tprivate readonly _fetch: typeof fetch;\n\tprivate readonly _spawn: PrismSpawnFn;\n\tprivate readonly _exists: (path: string) => boolean;\n\tprivate readonly _sleep: (ms: number) => Promise<void>;\n\tprivate readonly _hasCommand: (command: string) => boolean;\n\tprivate readonly _hasNvidiaGpu: () => boolean;\n\tprivate readonly _platform: () => string;\n\tprivate readonly _arch: () => string;\n\tprivate readonly _extractArchiveOverride: PrismExtractArchiveFn | undefined;\n\tprivate readonly _healthPollAttempts: number;\n\tprivate readonly _healthPollIntervalMs: number;\n\tprivate _child: Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\"> | undefined;\n\n\tconstructor(args: { agentDir: string; deps?: PrismLlamaCppDeps }) {\n\t\tthis._agentDir = args.agentDir;\n\t\tthis._fetch = args.deps?.fetchFn ?? fetch;\n\t\tthis._spawn =\n\t\t\targs.deps?.spawnFn ?? ((command, argv, options) => spawn(command, argv, { ...options, stdio: \"ignore\" }));\n\t\tthis._exists = args.deps?.existsFn ?? existsSync;\n\t\tthis._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));\n\t\tthis._hasCommand =\n\t\t\targs.deps?.hasCommand ??\n\t\t\t((command) =>\n\t\t\t\tspawnProcessSync(command, [\"--version\"], { encoding: \"utf8\", timeout: 5_000 }).error === undefined);\n\t\tthis._hasNvidiaGpu = args.deps?.hasNvidiaGpu ?? (() => this._hasCommand(\"nvidia-smi\"));\n\t\tthis._platform = args.deps?.platform ?? osPlatform;\n\t\tthis._arch = args.deps?.arch ?? osArch;\n\t\tthis._extractArchiveOverride = args.deps?.extractArchive;\n\t\tthis._healthPollAttempts = args.deps?.healthPollAttempts ?? DEFAULT_HEALTH_POLL_ATTEMPTS;\n\t\tthis._healthPollIntervalMs = args.deps?.healthPollIntervalMs ?? DEFAULT_HEALTH_POLL_INTERVAL_MS;\n\t}\n\n\truntimeDir(): string {\n\t\treturn agentRuntimesDir(\"prism-llamacpp\", this._agentDir);\n\t}\n\n\tmodelsDir(): string {\n\t\treturn agentModelsDir(\"llamacpp\", this._agentDir);\n\t}\n\n\tprivate _binaryName(): string {\n\t\treturn this._platform() === \"win32\" ? \"llama-server.exe\" : \"llama-server\";\n\t}\n\n\tprivate _manifestPath(): string {\n\t\treturn join(this.runtimeDir(), \"install.json\");\n\t}\n\n\tprivate _readManifest(): PrismInstallManifest | undefined {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(readFileSync(this._manifestPath(), \"utf8\")) as {\n\t\t\t\trelease?: unknown;\n\t\t\t\tbinaryRelPath?: unknown;\n\t\t\t\tbackend?: unknown;\n\t\t\t};\n\t\t\tif (\n\t\t\t\ttypeof parsed.release === \"string\" &&\n\t\t\t\ttypeof parsed.binaryRelPath === \"string\" &&\n\t\t\t\t(parsed.backend === \"cpu\" || parsed.backend === \"cuda\")\n\t\t\t) {\n\t\t\t\treturn { release: parsed.release, binaryRelPath: parsed.binaryRelPath, backend: parsed.backend };\n\t\t\t}\n\t\t\treturn undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _writeManifest(manifest: PrismInstallManifest): void {\n\t\twriteFileSync(this._manifestPath(), JSON.stringify(manifest, null, \"\\t\"));\n\t}\n\n\tasync detect(): Promise<PrismDetectResult> {\n\t\tconst manifest = this._readManifest();\n\t\tif (!manifest) return { runtimeInstalled: false };\n\t\tconst binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);\n\t\tconst runtimeInstalled = this._exists(binaryPath);\n\t\treturn {\n\t\t\truntimeInstalled,\n\t\t\tbinaryPath: runtimeInstalled ? binaryPath : undefined,\n\t\t\trelease: manifest.release,\n\t\t};\n\t}\n\n\t/** One-time recursive scan for the `llama-server` binary inside a freshly extracted release\n\t * archive — layouts vary (bin/ vs build/bin/ at the top level) across platforms/backends, so\n\t * this locates it deterministically instead of assuming a fixed path, then the result is\n\t * persisted so later calls never need to re-scan. */\n\tprivate _findBinaryRelPath(root: string): string | undefined {\n\t\tconst targetName = this._binaryName();\n\t\tconst stack: string[] = [root];\n\t\twhile (stack.length > 0) {\n\t\t\tconst dir = stack.pop();\n\t\t\tif (dir === undefined) continue;\n\t\t\tlet entries: Dirent[];\n\t\t\ttry {\n\t\t\t\tentries = readdirSync(dir, { withFileTypes: true });\n\t\t\t} catch {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tfor (const entry of entries) {\n\t\t\t\tconst full = join(dir, entry.name);\n\t\t\t\tif (entry.isDirectory()) stack.push(full);\n\t\t\t\telse if (entry.isFile() && entry.name === targetName) return relative(root, full);\n\t\t\t}\n\t\t}\n\t\treturn undefined;\n\t}\n\n\t/**\n\t * Download the pinned release asset for this host and extract it (consent-gated by the caller,\n\t * same contract as OllamaRuntime#installManaged — this method only does the mechanical\n\t * download+extract+locate and reports the outcome honestly). No compiler toolchain required:\n\t * the fork ships prebuilt binaries.\n\t */\n\tasync installManaged(onProgress?: (status: string) => void): Promise<{ ok: boolean; error?: string }> {\n\t\tconst asset = resolvePrismLlamaAsset(this._platform(), this._arch(), this._hasNvidiaGpu());\n\t\tif (!asset) return { ok: false, error: \"unsupported-platform\" };\n\n\t\tonProgress?.(`Downloading ${asset.name}…`);\n\t\tconst downloadUrl = `${PRISM_LLAMACPP_RELEASES_BASE_URL}/${PRISM_LLAMACPP_PINNED_RELEASE}/${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 = this.runtimeDir();\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 Readable, destDir, asset.kind);\n\t\tif (!extracted.ok) return extracted;\n\n\t\tonProgress?.(\"Locating llama-server binary…\");\n\t\tconst binaryRelPath = this._findBinaryRelPath(destDir);\n\t\tif (!binaryRelPath) {\n\t\t\treturn { ok: false, error: \"binary-missing: no llama-server binary found in the extracted archive\" };\n\t\t}\n\n\t\tthis._writeManifest({ release: PRISM_LLAMACPP_PINNED_RELEASE, binaryRelPath, backend: asset.backend });\n\t\tonProgress?.(\"Prism llama.cpp runtime installed.\");\n\t\treturn { ok: true };\n\t}\n\n\tprivate async _extractArchive(\n\t\tinput: Readable,\n\t\tdestDir: string,\n\t\tkind: PrismAssetKind,\n\t): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\treturn kind === \"zip\" ? await this._extractZip(input, destDir) : await this._extractTarGz(input, destDir);\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 _extractTarGz(input: Readable, destDir: string): Promise<{ ok: boolean; error?: string }> {\n\t\tconst tarProc = spawnProcess(\"tar\", [\"-xzf\", \"-\", \"-C\", destDir], {\n\t\t\tdetached: process.platform !== \"win32\",\n\t\t\tstdio: [\"pipe\", \"ignore\", \"ignore\"],\n\t\t});\n\t\tconst terminationController = new AbortController();\n\t\tconst processWait = waitForChildProcessWithTermination(tarProc, {\n\t\t\tsignal: terminationController.signal,\n\t\t\ttimeoutMs: EXTRACTION_TIMEOUT_MS,\n\t\t\tkillGraceMs: COMMAND_KILL_GRACE_MS,\n\t\t});\n\t\ttry {\n\t\t\tawait pipeline(input, requireStdin(tarProc, \"tar\"));\n\t\t} catch (error) {\n\t\t\tterminationController.abort();\n\t\t\tawait processWait.catch(() => {});\n\t\t\tthrow error;\n\t\t}\n\t\tconst terminal = await processWait;\n\t\tif (terminal.reason === \"timeout\" || terminal.code !== 0) {\n\t\t\treturn {\n\t\t\t\tok: false,\n\t\t\t\terror:\n\t\t\t\t\tterminal.reason === \"timeout\"\n\t\t\t\t\t\t? `extract-fail: tar extraction timed out after ${EXTRACTION_TIMEOUT_MS}ms`\n\t\t\t\t\t\t: `extract-fail: tar exited with code ${terminal.code ?? \"unknown\"}`,\n\t\t\t};\n\t\t}\n\t\treturn { ok: true };\n\t}\n\n\tprivate async _extractZip(input: Readable, destDir: string): Promise<{ ok: boolean; error?: string }> {\n\t\t// Zip's central directory needs seekable file access — buffer to a temp file first.\n\t\tconst zipPath = join(destDir, \"..\", `prism-llamacpp-download-${process.pid}-${Date.now()}.zip`);\n\t\tawait pipeline(input, 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, { detached: process.platform !== \"win32\", stdio: \"ignore\" });\n\t\t\tconst terminal = await waitForChildProcessWithTermination(proc, {\n\t\t\t\ttimeoutMs: EXTRACTION_TIMEOUT_MS,\n\t\t\t\tkillGraceMs: COMMAND_KILL_GRACE_MS,\n\t\t\t});\n\t\t\tif (terminal.reason === \"timeout\" || terminal.code !== 0) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror:\n\t\t\t\t\t\tterminal.reason === \"timeout\"\n\t\t\t\t\t\t\t? `extract-fail: ${extractCommand} timed out after ${EXTRACTION_TIMEOUT_MS}ms`\n\t\t\t\t\t\t\t: `extract-fail: ${extractCommand} exited with code ${terminal.code ?? \"unknown\"}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\treturn { ok: true };\n\t\t} finally {\n\t\t\ttry {\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\tprivate async _remoteContentLength(url: string): Promise<number | undefined> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(url, {\n\t\t\t\tmethod: \"HEAD\",\n\t\t\t\tsignal: AbortSignal.timeout(HEAD_REQUEST_TIMEOUT_MS),\n\t\t\t});\n\t\t\treturn response.ok ? parseContentLength(response.headers.get(\"content-length\")) : undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _fileSizeBytes(path: string): number | undefined {\n\t\ttry {\n\t\t\treturn statSync(path).size;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _cleanupPartial(path: string): void {\n\t\ttry {\n\t\t\trmSync(path, { force: true });\n\t\t} catch {\n\t\t\t// best-effort cleanup\n\t\t}\n\t}\n\n\t/**\n\t * Wait until a write stream's underlying fd is actually closed. `pipeline()` destroys the\n\t * stream on error but its promise can settle before the fd close completes (Windows in\n\t * particular keeps the file handle open a tick longer) — unlinking before that races with the\n\t * OS and can leave the \"partial\" file behind or fail the delete outright. Resolves immediately\n\t * if the stream already reports `closed`.\n\t */\n\tprivate _waitForStreamClosed(stream: Writable & { closed?: boolean }): Promise<void> {\n\t\tif (stream.closed) return Promise.resolve();\n\t\treturn new Promise((resolve) => {\n\t\t\tstream.once(\"close\", () => resolve());\n\t\t\tstream.destroy();\n\t\t});\n\t}\n\n\t/**\n\t * Stream a GGUF (or mmproj) file from Hugging Face into pi's owned models dir. Skips a re-download\n\t * when the local file already matches the remote size; verifies size when the response reports\n\t * `content-length` and deletes the partial file on any failure or mismatch — never leaves a\n\t * corrupt/truncated weight file behind for a later load to silently misread.\n\t */\n\tasync downloadModel(\n\t\targs: { repo: string; file: string },\n\t\tonProgress?: (status: string) => void,\n\t): Promise<PrismDownloadResult> {\n\t\tconst destPath = join(this.modelsDir(), args.repo, args.file);\n\t\tconst url = `https://huggingface.co/${args.repo}/resolve/main/${args.file}`;\n\t\tmkdirSync(dirname(destPath), { recursive: true });\n\n\t\tif (this._exists(destPath)) {\n\t\t\tconst remoteSize = await this._remoteContentLength(url);\n\t\t\tif (remoteSize !== undefined && remoteSize === this._fileSizeBytes(destPath)) {\n\t\t\t\tonProgress?.(`${args.file} already downloaded (${remoteSize} bytes) — skipping.`);\n\t\t\t\treturn { ok: true, path: destPath, skipped: true };\n\t\t\t}\n\t\t}\n\n\t\tonProgress?.(`Downloading ${args.file} from ${args.repo}…`);\n\t\tlet response: Response;\n\t\ttry {\n\t\t\tresponse = await this._fetch(url);\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 expectedBytes = parseContentLength(response.headers.get(\"content-length\"));\n\t\tconst writeStream = createWriteStream(destPath);\n\t\ttry {\n\t\t\tawait pipeline(response.body as unknown as Readable, writeStream);\n\t\t} catch (error) {\n\t\t\tawait this._waitForStreamClosed(writeStream);\n\t\t\tthis._cleanupPartial(destPath);\n\t\t\treturn { ok: false, error: `download-fail: ${error instanceof Error ? error.message : String(error)}` };\n\t\t}\n\n\t\tconst actualBytes = this._fileSizeBytes(destPath);\n\t\tif (expectedBytes !== undefined && actualBytes !== expectedBytes) {\n\t\t\tthis._cleanupPartial(destPath);\n\t\t\treturn { ok: false, error: `size-mismatch: expected ${expectedBytes} bytes, got ${actualBytes ?? 0}` };\n\t\t}\n\n\t\tonProgress?.(`${args.file} downloaded (${actualBytes ?? 0} bytes).`);\n\t\treturn { ok: true, path: destPath };\n\t}\n\n\tprivate async _healthUp(baseUrl: string): Promise<boolean> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${baseUrl}/health`, {\n\t\t\t\tsignal: AbortSignal.timeout(HEALTH_CHECK_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\t/**\n\t * Spawn `llama-server` detached+tracked (killed on parent shutdown even if pi crashes without\n\t * calling stop()) and poll `/health` until ready. `-ngl 99` is only added when the installed\n\t * asset's persisted `backend` is `\"cuda\"`, not from the host's current GPU state.\n\t */\n\tasync serve(args: {\n\t\tmodelPath: string;\n\t\tmmprojPath?: string;\n\t\tport: number;\n\t\tnumCtx: number;\n\t}): Promise<PrismServeResult> {\n\t\tconst manifest = this._readManifest();\n\t\tif (!manifest) return { ok: false, error: \"binary-missing\" };\n\t\tconst binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);\n\t\tif (!this._exists(binaryPath)) return { ok: false, error: \"binary-missing\" };\n\n\t\tconst baseUrl = `http://127.0.0.1:${args.port}`;\n\t\tconst argv = [\n\t\t\t\"-m\",\n\t\t\targs.modelPath,\n\t\t\t...(args.mmprojPath ? [\"--mmproj\", args.mmprojPath] : []),\n\t\t\t\"--host\",\n\t\t\t\"127.0.0.1\",\n\t\t\t\"--port\",\n\t\t\tString(args.port),\n\t\t\t\"-c\",\n\t\t\tString(args.numCtx),\n\t\t\t...(manifest.backend === \"cuda\" ? [\"-ngl\", \"99\"] : []),\n\t\t];\n\n\t\tconst child = this._spawn(binaryPath, argv, {\n\t\t\tdetached: process.platform !== \"win32\",\n\t\t\tstdio: \"ignore\",\n\t\t\tenv: process.env,\n\t\t});\n\t\tif (child.pid) trackDetachedChildPid(child.pid);\n\t\tchild.unref?.();\n\t\tchild.on(\"exit\", () => {\n\t\t\tthis._child = undefined;\n\t\t});\n\t\tthis._child = child;\n\n\t\tfor (let attempt = 0; attempt < this._healthPollAttempts; attempt++) {\n\t\t\tif (await this._healthUp(baseUrl)) return { ok: true, baseUrl };\n\t\t\tawait this._sleep(this._healthPollIntervalMs);\n\t\t}\n\t\tthis.stop();\n\t\treturn { ok: false, error: \"health-timeout\" };\n\t}\n\n\tstop(): { stopped: boolean } {\n\t\tconst child = this._child;\n\t\tif (!child) return { stopped: false };\n\t\tthis._child = undefined;\n\t\tif (child.pid) {\n\t\t\tuntrackDetachedChildPid(child.pid);\n\t\t\tkillProcessTree(child.pid);\n\t\t}\n\t\treturn { stopped: true };\n\t}\n\n\tisRunning(): boolean {\n\t\treturn this._child !== undefined;\n\t}\n}\n"]}

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

@@ -0,0 +1,475 @@
+import { spawn } from "node:child_process";
+import { createWriteStream, existsSync, mkdirSync, readdirSync, readFileSync, rmSync, statSync, writeFileSync, } from "node:fs";
+import { arch as osArch, platform as osPlatform } from "node:os";
+import { dirname, join, relative } from "node:path";
+import { pipeline } from "node:stream/promises";
+import { spawnProcess, spawnProcessSync, waitForChildProcessWithTermination } from "../../utils/child-process.js";
+import { killProcessTree, trackDetachedChildPid, untrackDetachedChildPid } from "../../utils/shell.js";
+import { modelsDir as agentModelsDir, runtimesDir as agentRuntimesDir } from "../agent-paths.js";
+/**
+ * Managed runtime for prism-ml 1-bit GGUF models (Bonsai-27B first). The provider requires their
+ * llama.cpp fork (Q1_0_g128 hybrid-attention kernels) — stock llama.cpp/Ollama cannot serve these
+ * weights — so this module owns runtime install (prebuilt release download+extract), GGUF
+ * downloads, and llama-server lifecycle. Mirrors local-runtime.ts's OllamaRuntime: injectable
+ * seams for fetch/spawn/exists, pi-owned directories under agentDir (runtimes/, models/),
+ * detached+tracked child processes, onProgress as best-effort UI feedback, honest error taxonomies
+ * instead of silent fallbacks.
+ *
+ * The fork publishes precompiled per-platform archives on GitHub Releases (no local build): pin a
+ * release tag, download the matching asset, extract it, and locate the `llama-server` binary once
+ * at install time (release archive layouts vary — bin/ vs build/bin/ at the top level — so this is
+ * a deterministic one-time scan, not a fallback). The found relative path and backend (cpu/cuda)
+ * are persisted to a manifest so later `detect()`/`serve()` calls never need to re-scan.
+ */
+/**
+ * Pinned Prism ML llama.cpp fork release — verified via
+ * `GET https://api.github.com/repos/PrismML-Eng/llama.cpp/releases/tags/prism-b9594-38c66ad` on
+ * 2026-07-18 (release exists, all assets referenced below are present on it). Bump here (and
+ * re-verify with the same API call) when the provider ships new kernels.
+ */
+export const PRISM_LLAMACPP_PINNED_RELEASE = "prism-b9594-38c66ad";
+export const PRISM_LLAMACPP_RELEASES_BASE_URL = "https://github.com/PrismML-Eng/llama.cpp/releases/download";
+/** Curated first-model descriptor; the `/models add` wiring task consumes this. */
+export const BONSAI_27B = {
+    repo: "prism-ml/Bonsai-27B-gguf",
+    file: "Bonsai-27B-Q1_0.gguf",
+    mmprojFile: "Bonsai-27B-mmproj-Q8_0.gguf",
+    displayName: "Bonsai-27B (1-bit Q1_0 + vision)",
+};
+// Verbatim asset names from the pinned release (see PRISM_LLAMACPP_PINNED_RELEASE doc comment) —
+// hardcoded rather than templated from the tag because the provider's naming isn't fully
+// consistent across platforms (Windows CPU assets drop the tag/commit segment entirely; the
+// Windows CUDA asset uses a different build-number segment, "b1" not "b9594").
+const LINUX_X64_CPU_ASSET = "llama-prism-b9594-38c66ad-bin-ubuntu-x64.tar.gz";
+const LINUX_ARM64_CPU_ASSET = "llama-prism-b9594-38c66ad-bin-ubuntu-arm64.tar.gz";
+const LINUX_X64_CUDA_ASSET = "llama-prism-b9594-38c66ad-bin-linux-cuda-12.4-x64.tar.gz";
+const MACOS_ARM64_ASSET = "llama-prism-b9594-38c66ad-bin-macos-arm64.tar.gz";
+const MACOS_X64_ASSET = "llama-prism-b9594-38c66ad-bin-macos-x64.tar.gz";
+const WIN_X64_CPU_ASSET = "llama-bin-win-cpu-x64.zip";
+const WIN_ARM64_CPU_ASSET = "llama-bin-win-cpu-arm64.zip";
+/**
+ * Maps a platform/arch/GPU triple to the exact Prism llama.cpp release asset for
+ * {@link PRISM_LLAMACPP_PINNED_RELEASE} — verified against the real GitHub release, not guessed.
+ * CPU asset by default; the CUDA 12.4 variant is only offered for linux x64 with an NVIDIA GPU
+ * present (a 12.8 variant also exists upstream but has no wired caller yet). Windows CUDA needs a
+ * companion `cudart-*` archive and is out of scope for v1 — Windows always resolves to the CPU zip
+ * regardless of `hasNvidiaGpu`. Pure and exported so it's independently testable.
+ */
+export function resolvePrismLlamaAsset(plat, architecture, hasNvidiaGpu) {
+    if (plat === "linux") {
+        if (architecture === "x64") {
+            return hasNvidiaGpu
+                ? { name: LINUX_X64_CUDA_ASSET, kind: "tar-gz", backend: "cuda" }
+                : { name: LINUX_X64_CPU_ASSET, kind: "tar-gz", backend: "cpu" };
+        }
+        if (architecture === "arm64")
+            return { name: LINUX_ARM64_CPU_ASSET, kind: "tar-gz", backend: "cpu" };
+        return undefined;
+    }
+    if (plat === "darwin") {
+        if (architecture === "arm64")
+            return { name: MACOS_ARM64_ASSET, kind: "tar-gz", backend: "cpu" };
+        if (architecture === "x64")
+            return { name: MACOS_X64_ASSET, kind: "tar-gz", backend: "cpu" };
+        return undefined;
+    }
+    if (plat === "win32") {
+        if (architecture === "x64")
+            return { name: WIN_X64_CPU_ASSET, kind: "zip", backend: "cpu" };
+        if (architecture === "arm64")
+            return { name: WIN_ARM64_CPU_ASSET, kind: "zip", backend: "cpu" };
+        return undefined;
+    }
+    return undefined;
+}
+const EXTRACTION_TIMEOUT_MS = 10 * 60_000;
+const COMMAND_KILL_GRACE_MS = 2_000;
+const DEFAULT_HEALTH_POLL_ATTEMPTS = 240;
+const DEFAULT_HEALTH_POLL_INTERVAL_MS = 500;
+const HEALTH_CHECK_TIMEOUT_MS = 2_000;
+const HEAD_REQUEST_TIMEOUT_MS = 10_000;
+function parseContentLength(header) {
+    if (!header)
+        return undefined;
+    const value = Number(header);
+    return Number.isFinite(value) && value >= 0 ? value : undefined;
+}
+/**
+ * `spawnProcess(..., { stdio: ["pipe", ...] })` always yields a non-null `stdin` at runtime, but
+ * child-process.ts's typed overload only narrows `.stdin` for the capture-only stdio shape used
+ * elsewhere. 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;
+}
+export class PrismLlamaCppRuntime {
+    _agentDir;
+    _fetch;
+    _spawn;
+    _exists;
+    _sleep;
+    _hasCommand;
+    _hasNvidiaGpu;
+    _platform;
+    _arch;
+    _extractArchiveOverride;
+    _healthPollAttempts;
+    _healthPollIntervalMs;
+    _child;
+    constructor(args) {
+        this._agentDir = args.agentDir;
+        this._fetch = args.deps?.fetchFn ?? fetch;
+        this._spawn =
+            args.deps?.spawnFn ?? ((command, argv, options) => spawn(command, argv, { ...options, stdio: "ignore" }));
+        this._exists = args.deps?.existsFn ?? existsSync;
+        this._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));
+        this._hasCommand =
+            args.deps?.hasCommand ??
+                ((command) => spawnProcessSync(command, ["--version"], { encoding: "utf8", timeout: 5_000 }).error === undefined);
+        this._hasNvidiaGpu = args.deps?.hasNvidiaGpu ?? (() => this._hasCommand("nvidia-smi"));
+        this._platform = args.deps?.platform ?? osPlatform;
+        this._arch = args.deps?.arch ?? osArch;
+        this._extractArchiveOverride = args.deps?.extractArchive;
+        this._healthPollAttempts = args.deps?.healthPollAttempts ?? DEFAULT_HEALTH_POLL_ATTEMPTS;
+        this._healthPollIntervalMs = args.deps?.healthPollIntervalMs ?? DEFAULT_HEALTH_POLL_INTERVAL_MS;
+    }
+    runtimeDir() {
+        return agentRuntimesDir("prism-llamacpp", this._agentDir);
+    }
+    modelsDir() {
+        return agentModelsDir("llamacpp", this._agentDir);
+    }
+    _binaryName() {
+        return this._platform() === "win32" ? "llama-server.exe" : "llama-server";
+    }
+    _manifestPath() {
+        return join(this.runtimeDir(), "install.json");
+    }
+    _readManifest() {
+        try {
+            const parsed = JSON.parse(readFileSync(this._manifestPath(), "utf8"));
+            if (typeof parsed.release === "string" &&
+                typeof parsed.binaryRelPath === "string" &&
+                (parsed.backend === "cpu" || parsed.backend === "cuda")) {
+                return { release: parsed.release, binaryRelPath: parsed.binaryRelPath, backend: parsed.backend };
+            }
+            return undefined;
+        }
+        catch {
+            return undefined;
+        }
+    }
+    _writeManifest(manifest) {
+        writeFileSync(this._manifestPath(), JSON.stringify(manifest, null, "\t"));
+    }
+    async detect() {
+        const manifest = this._readManifest();
+        if (!manifest)
+            return { runtimeInstalled: false };
+        const binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);
+        const runtimeInstalled = this._exists(binaryPath);
+        return {
+            runtimeInstalled,
+            binaryPath: runtimeInstalled ? binaryPath : undefined,
+            release: manifest.release,
+        };
+    }
+    /** One-time recursive scan for the `llama-server` binary inside a freshly extracted release
+     * archive — layouts vary (bin/ vs build/bin/ at the top level) across platforms/backends, so
+     * this locates it deterministically instead of assuming a fixed path, then the result is
+     * persisted so later calls never need to re-scan. */
+    _findBinaryRelPath(root) {
+        const targetName = this._binaryName();
+        const stack = [root];
+        while (stack.length > 0) {
+            const dir = stack.pop();
+            if (dir === undefined)
+                continue;
+            let entries;
+            try {
+                entries = readdirSync(dir, { withFileTypes: true });
+            }
+            catch {
+                continue;
+            }
+            for (const entry of entries) {
+                const full = join(dir, entry.name);
+                if (entry.isDirectory())
+                    stack.push(full);
+                else if (entry.isFile() && entry.name === targetName)
+                    return relative(root, full);
+            }
+        }
+        return undefined;
+    }
+    /**
+     * Download the pinned release asset for this host and extract it (consent-gated by the caller,
+     * same contract as OllamaRuntime#installManaged — this method only does the mechanical
+     * download+extract+locate and reports the outcome honestly). No compiler toolchain required:
+     * the fork ships prebuilt binaries.
+     */
+    async installManaged(onProgress) {
+        const asset = resolvePrismLlamaAsset(this._platform(), this._arch(), this._hasNvidiaGpu());
+        if (!asset)
+            return { ok: false, error: "unsupported-platform" };
+        onProgress?.(`Downloading ${asset.name}…`);
+        const downloadUrl = `${PRISM_LLAMACPP_RELEASES_BASE_URL}/${PRISM_LLAMACPP_PINNED_RELEASE}/${asset.name}`;
+        let response;
+        try {
+            response = await this._fetch(downloadUrl);
+        }
+        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 = this.runtimeDir();
+        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;
+        onProgress?.("Locating llama-server binary…");
+        const binaryRelPath = this._findBinaryRelPath(destDir);
+        if (!binaryRelPath) {
+            return { ok: false, error: "binary-missing: no llama-server binary found in the extracted archive" };
+        }
+        this._writeManifest({ release: PRISM_LLAMACPP_PINNED_RELEASE, binaryRelPath, backend: asset.backend });
+        onProgress?.("Prism llama.cpp runtime installed.");
+        return { ok: true };
+    }
+    async _extractArchive(input, destDir, kind) {
+        try {
+            return kind === "zip" ? await this._extractZip(input, destDir) : await this._extractTarGz(input, destDir);
+        }
+        catch (error) {
+            return { ok: false, error: `extract-fail: ${error instanceof Error ? error.message : String(error)}` };
+        }
+    }
+    async _extractTarGz(input, destDir) {
+        const tarProc = spawnProcess("tar", ["-xzf", "-", "-C", destDir], {
+            detached: process.platform !== "win32",
+            stdio: ["pipe", "ignore", "ignore"],
+        });
+        const terminationController = new AbortController();
+        const processWait = waitForChildProcessWithTermination(tarProc, {
+            signal: terminationController.signal,
+            timeoutMs: EXTRACTION_TIMEOUT_MS,
+            killGraceMs: COMMAND_KILL_GRACE_MS,
+        });
+        try {
+            await pipeline(input, requireStdin(tarProc, "tar"));
+        }
+        catch (error) {
+            terminationController.abort();
+            await processWait.catch(() => { });
+            throw error;
+        }
+        const terminal = await processWait;
+        if (terminal.reason === "timeout" || terminal.code !== 0) {
+            return {
+                ok: false,
+                error: terminal.reason === "timeout"
+                    ? `extract-fail: tar extraction timed out after ${EXTRACTION_TIMEOUT_MS}ms`
+                    : `extract-fail: tar exited with code ${terminal.code ?? "unknown"}`,
+            };
+        }
+        return { ok: true };
+    }
+    async _extractZip(input, destDir) {
+        // Zip's central directory needs seekable file access — buffer to a temp file first.
+        const zipPath = join(destDir, "..", `prism-llamacpp-download-${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, { detached: process.platform !== "win32", stdio: "ignore" });
+            const terminal = await waitForChildProcessWithTermination(proc, {
+                timeoutMs: EXTRACTION_TIMEOUT_MS,
+                killGraceMs: COMMAND_KILL_GRACE_MS,
+            });
+            if (terminal.reason === "timeout" || terminal.code !== 0) {
+                return {
+                    ok: false,
+                    error: terminal.reason === "timeout"
+                        ? `extract-fail: ${extractCommand} timed out after ${EXTRACTION_TIMEOUT_MS}ms`
+                        : `extract-fail: ${extractCommand} exited with code ${terminal.code ?? "unknown"}`,
+                };
+            }
+            return { ok: true };
+        }
+        finally {
+            try {
+                rmSync(zipPath, { force: true });
+            }
+            catch {
+                // best-effort cleanup
+            }
+        }
+    }
+    async _remoteContentLength(url) {
+        try {
+            const response = await this._fetch(url, {
+                method: "HEAD",
+                signal: AbortSignal.timeout(HEAD_REQUEST_TIMEOUT_MS),
+            });
+            return response.ok ? parseContentLength(response.headers.get("content-length")) : undefined;
+        }
+        catch {
+            return undefined;
+        }
+    }
+    _fileSizeBytes(path) {
+        try {
+            return statSync(path).size;
+        }
+        catch {
+            return undefined;
+        }
+    }
+    _cleanupPartial(path) {
+        try {
+            rmSync(path, { force: true });
+        }
+        catch {
+            // best-effort cleanup
+        }
+    }
+    /**
+     * Wait until a write stream's underlying fd is actually closed. `pipeline()` destroys the
+     * stream on error but its promise can settle before the fd close completes (Windows in
+     * particular keeps the file handle open a tick longer) — unlinking before that races with the
+     * OS and can leave the "partial" file behind or fail the delete outright. Resolves immediately
+     * if the stream already reports `closed`.
+     */
+    _waitForStreamClosed(stream) {
+        if (stream.closed)
+            return Promise.resolve();
+        return new Promise((resolve) => {
+            stream.once("close", () => resolve());
+            stream.destroy();
+        });
+    }
+    /**
+     * Stream a GGUF (or mmproj) file from Hugging Face into pi's owned models dir. Skips a re-download
+     * when the local file already matches the remote size; verifies size when the response reports
+     * `content-length` and deletes the partial file on any failure or mismatch — never leaves a
+     * corrupt/truncated weight file behind for a later load to silently misread.
+     */
+    async downloadModel(args, onProgress) {
+        const destPath = join(this.modelsDir(), args.repo, args.file);
+        const url = `https://huggingface.co/${args.repo}/resolve/main/${args.file}`;
+        mkdirSync(dirname(destPath), { recursive: true });
+        if (this._exists(destPath)) {
+            const remoteSize = await this._remoteContentLength(url);
+            if (remoteSize !== undefined && remoteSize === this._fileSizeBytes(destPath)) {
+                onProgress?.(`${args.file} already downloaded (${remoteSize} bytes) — skipping.`);
+                return { ok: true, path: destPath, skipped: true };
+            }
+        }
+        onProgress?.(`Downloading ${args.file} from ${args.repo}…`);
+        let response;
+        try {
+            response = await this._fetch(url);
+        }
+        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 expectedBytes = parseContentLength(response.headers.get("content-length"));
+        const writeStream = createWriteStream(destPath);
+        try {
+            await pipeline(response.body, writeStream);
+        }
+        catch (error) {
+            await this._waitForStreamClosed(writeStream);
+            this._cleanupPartial(destPath);
+            return { ok: false, error: `download-fail: ${error instanceof Error ? error.message : String(error)}` };
+        }
+        const actualBytes = this._fileSizeBytes(destPath);
+        if (expectedBytes !== undefined && actualBytes !== expectedBytes) {
+            this._cleanupPartial(destPath);
+            return { ok: false, error: `size-mismatch: expected ${expectedBytes} bytes, got ${actualBytes ?? 0}` };
+        }
+        onProgress?.(`${args.file} downloaded (${actualBytes ?? 0} bytes).`);
+        return { ok: true, path: destPath };
+    }
+    async _healthUp(baseUrl) {
+        try {
+            const response = await this._fetch(`${baseUrl}/health`, {
+                signal: AbortSignal.timeout(HEALTH_CHECK_TIMEOUT_MS),
+            });
+            return response.ok;
+        }
+        catch {
+            return false;
+        }
+    }
+    /**
+     * Spawn `llama-server` detached+tracked (killed on parent shutdown even if pi crashes without
+     * calling stop()) and poll `/health` until ready. `-ngl 99` is only added when the installed
+     * asset's persisted `backend` is `"cuda"`, not from the host's current GPU state.
+     */
+    async serve(args) {
+        const manifest = this._readManifest();
+        if (!manifest)
+            return { ok: false, error: "binary-missing" };
+        const binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);
+        if (!this._exists(binaryPath))
+            return { ok: false, error: "binary-missing" };
+        const baseUrl = `http://127.0.0.1:${args.port}`;
+        const argv = [
+            "-m",
+            args.modelPath,
+            ...(args.mmprojPath ? ["--mmproj", args.mmprojPath] : []),
+            "--host",
+            "127.0.0.1",
+            "--port",
+            String(args.port),
+            "-c",
+            String(args.numCtx),
+            ...(manifest.backend === "cuda" ? ["-ngl", "99"] : []),
+        ];
+        const child = this._spawn(binaryPath, argv, {
+            detached: process.platform !== "win32",
+            stdio: "ignore",
+            env: process.env,
+        });
+        if (child.pid)
+            trackDetachedChildPid(child.pid);
+        child.unref?.();
+        child.on("exit", () => {
+            this._child = undefined;
+        });
+        this._child = child;
+        for (let attempt = 0; attempt < this._healthPollAttempts; attempt++) {
+            if (await this._healthUp(baseUrl))
+                return { ok: true, baseUrl };
+            await this._sleep(this._healthPollIntervalMs);
+        }
+        this.stop();
+        return { ok: false, error: "health-timeout" };
+    }
+    stop() {
+        const child = this._child;
+        if (!child)
+            return { stopped: false };
+        this._child = undefined;
+        if (child.pid) {
+            untrackDetachedChildPid(child.pid);
+            killProcessTree(child.pid);
+        }
+        return { stopped: true };
+    }
+    isRunning() {
+        return this._child !== undefined;
+    }
+}
+//# sourceMappingURL=llamacpp-runtime.js.map

package/dist/core/models/llamacpp-runtime.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llamacpp-runtime.js","sourceRoot":"","sources":["../../../src/core/models/llamacpp-runtime.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EACN,iBAAiB,EAEjB,UAAU,EACV,SAAS,EACT,WAAW,EACX,YAAY,EACZ,MAAM,EACN,QAAQ,EACR,aAAa,GACb,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,IAAI,IAAI,MAAM,EAAE,QAAQ,IAAI,UAAU,EAAE,MAAM,SAAS,CAAC;AACjE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,WAAW,CAAC;AAEpD,OAAO,EAAE,QAAQ,EAAE,MAAM,sBAAsB,CAAC;AAChD,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,kCAAkC,EAAE,MAAM,8BAA8B,CAAC;AAClH,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAC;AACvG,OAAO,EAAE,SAAS,IAAI,cAAc,EAAE,WAAW,IAAI,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAEjG;;;;;;;;;;;;;;GAcG;AAEH;;;;;GAKG;AACH,MAAM,CAAC,MAAM,6BAA6B,GAAG,qBAAqB,CAAC;AAEnE,MAAM,CAAC,MAAM,gCAAgC,GAAG,4DAA4D,CAAC;AAS7G,mFAAmF;AACnF,MAAM,CAAC,MAAM,UAAU,GAAyB;IAC/C,IAAI,EAAE,0BAA0B;IAChC,IAAI,EAAE,sBAAsB;IAC5B,UAAU,EAAE,6BAA6B;IACzC,WAAW,EAAE,kCAAkC;CAC/C,CAAC;AAWF,mGAAiG;AACjG,yFAAyF;AACzF,4FAA4F;AAC5F,+EAA+E;AAC/E,MAAM,mBAAmB,GAAG,iDAAiD,CAAC;AAC9E,MAAM,qBAAqB,GAAG,mDAAmD,CAAC;AAClF,MAAM,oBAAoB,GAAG,0DAA0D,CAAC;AACxF,MAAM,iBAAiB,GAAG,kDAAkD,CAAC;AAC7E,MAAM,eAAe,GAAG,gDAAgD,CAAC;AACzE,MAAM,iBAAiB,GAAG,2BAA2B,CAAC;AACtD,MAAM,mBAAmB,GAAG,6BAA6B,CAAC;AAE1D;;;;;;;GAOG;AACH,MAAM,UAAU,sBAAsB,CACrC,IAAY,EACZ,YAAoB,EACpB,YAAqB,EACS;IAC9B,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;QACtB,IAAI,YAAY,KAAK,KAAK,EAAE,CAAC;YAC5B,OAAO,YAAY;gBAClB,CAAC,CAAC,EAAE,IAAI,EAAE,oBAAoB,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE;gBACjE,CAAC,CAAC,EAAE,IAAI,EAAE,mBAAmB,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAClE,CAAC;QACD,IAAI,YAAY,KAAK,OAAO;YAAE,OAAO,EAAE,IAAI,EAAE,qBAAqB,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QACrG,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;QACvB,IAAI,YAAY,KAAK,OAAO;YAAE,OAAO,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QACjG,IAAI,YAAY,KAAK,KAAK;YAAE,OAAO,EAAE,IAAI,EAAE,eAAe,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAC7F,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;QACtB,IAAI,YAAY,KAAK,KAAK;YAAE,OAAO,EAAE,IAAI,EAAE,iBAAiB,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAC5F,IAAI,YAAY,KAAK,OAAO;YAAE,OAAO,EAAE,IAAI,EAAE,mBAAmB,EAAE,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAChG,OAAO,SAAS,CAAC;IAClB,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAwDD,MAAM,qBAAqB,GAAG,EAAE,GAAG,MAAM,CAAC;AAC1C,MAAM,qBAAqB,GAAG,KAAK,CAAC;AACpC,MAAM,4BAA4B,GAAG,GAAG,CAAC;AACzC,MAAM,+BAA+B,GAAG,GAAG,CAAC;AAC5C,MAAM,uBAAuB,GAAG,KAAK,CAAC;AACtC,MAAM,uBAAuB,GAAG,MAAM,CAAC;AAEvC,SAAS,kBAAkB,CAAC,MAAqB,EAAsB;IACtE,IAAI,CAAC,MAAM;QAAE,OAAO,SAAS,CAAC;IAC9B,MAAM,KAAK,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC;IAC7B,OAAO,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CAChE;AAED;;;;;GAKG;AACH,SAAS,YAAY,CAAC,IAAkB,EAAE,KAAa,EAAY;IAClE,IAAI,CAAC,IAAI,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,GAAG,KAAK,iBAAiB,CAAC,CAAC;IAC5D,OAAO,IAAI,CAAC,KAAK,CAAC;AAAA,CAClB;AAED,MAAM,OAAO,oBAAoB;IACf,SAAS,CAAS;IAClB,MAAM,CAAe;IACrB,MAAM,CAAe;IACrB,OAAO,CAA4B;IACnC,MAAM,CAAgC;IACtC,WAAW,CAA+B;IAC1C,aAAa,CAAgB;IAC7B,SAAS,CAAe;IACxB,KAAK,CAAe;IACpB,uBAAuB,CAAoC;IAC3D,mBAAmB,CAAS;IAC5B,qBAAqB,CAAS;IACvC,MAAM,CAAkE;IAEhF,YAAY,IAAoD,EAAE;QACjE,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC/B,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,IAAI,EAAE,OAAO,IAAI,KAAK,CAAC;QAC1C,IAAI,CAAC,MAAM;YACV,IAAI,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,CAAC,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,EAAE,GAAG,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC;QAC3G,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,QAAQ,IAAI,UAAU,CAAC;QACjD,IAAI,CAAC,MAAM,GAAG,IAAI,CAAC,IAAI,EAAE,OAAO,IAAI,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;QAChG,IAAI,CAAC,WAAW;YACf,IAAI,CAAC,IAAI,EAAE,UAAU;gBACrB,CAAC,CAAC,OAAO,EAAE,EAAE,CACZ,gBAAgB,CAAC,OAAO,EAAE,CAAC,WAAW,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;QACtG,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC,IAAI,EAAE,YAAY,IAAI,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,YAAY,CAAC,CAAC,CAAC;QACvF,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,IAAI,EAAE,QAAQ,IAAI,UAAU,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,IAAI,EAAE,IAAI,IAAI,MAAM,CAAC;QACvC,IAAI,CAAC,uBAAuB,GAAG,IAAI,CAAC,IAAI,EAAE,cAAc,CAAC;QACzD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,IAAI,EAAE,kBAAkB,IAAI,4BAA4B,CAAC;QACzF,IAAI,CAAC,qBAAqB,GAAG,IAAI,CAAC,IAAI,EAAE,oBAAoB,IAAI,+BAA+B,CAAC;IAAA,CAChG;IAED,UAAU,GAAW;QACpB,OAAO,gBAAgB,CAAC,gBAAgB,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IAAA,CAC1D;IAED,SAAS,GAAW;QACnB,OAAO,cAAc,CAAC,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IAAA,CAClD;IAEO,WAAW,GAAW;QAC7B,OAAO,IAAI,CAAC,SAAS,EAAE,KAAK,OAAO,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,cAAc,CAAC;IAAA,CAC1E;IAEO,aAAa,GAAW;QAC/B,OAAO,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,cAAc,CAAC,CAAC;IAAA,CAC/C;IAEO,aAAa,GAAqC;QACzD,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,aAAa,EAAE,EAAE,MAAM,CAAC,CAInE,CAAC;YACF,IACC,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ;gBAClC,OAAO,MAAM,CAAC,aAAa,KAAK,QAAQ;gBACxC,CAAC,MAAM,CAAC,OAAO,KAAK,KAAK,IAAI,MAAM,CAAC,OAAO,KAAK,MAAM,CAAC,EACtD,CAAC;gBACF,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,aAAa,EAAE,MAAM,CAAC,aAAa,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;YAClG,CAAC;YACD,OAAO,SAAS,CAAC;QAClB,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,SAAS,CAAC;QAClB,CAAC;IAAA,CACD;IAEO,cAAc,CAAC,QAA8B,EAAQ;QAC5D,aAAa,CAAC,IAAI,CAAC,aAAa,EAAE,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;IAAA,CAC1E;IAED,KAAK,CAAC,MAAM,GAA+B;QAC1C,MAAM,QAAQ,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QACtC,IAAI,CAAC,QAAQ;YAAE,OAAO,EAAE,gBAAgB,EAAE,KAAK,EAAE,CAAC;QAClD,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,QAAQ,CAAC,aAAa,CAAC,CAAC;QACnE,MAAM,gBAAgB,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QAClD,OAAO;YACN,gBAAgB;YAChB,UAAU,EAAE,gBAAgB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,SAAS;YACrD,OAAO,EAAE,QAAQ,CAAC,OAAO;SACzB,CAAC;IAAA,CACF;IAED;;;yDAGqD;IAC7C,kBAAkB,CAAC,IAAY,EAAsB;QAC5D,MAAM,UAAU,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;QACtC,MAAM,KAAK,GAAa,CAAC,IAAI,CAAC,CAAC;QAC/B,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzB,MAAM,GAAG,GAAG,KAAK,CAAC,GAAG,EAAE,CAAC;YACxB,IAAI,GAAG,KAAK,SAAS;gBAAE,SAAS;YAChC,IAAI,OAAiB,CAAC;YACtB,IAAI,CAAC;gBACJ,OAAO,GAAG,WAAW,CAAC,GAAG,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;YACrD,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS;YACV,CAAC;YACD,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;gBAC7B,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;gBACnC,IAAI,KAAK,CAAC,WAAW,EAAE;oBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;qBACrC,IAAI,KAAK,CAAC,MAAM,EAAE,IAAI,KAAK,CAAC,IAAI,KAAK,UAAU;oBAAE,OAAO,QAAQ,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;YACnF,CAAC;QACF,CAAC;QACD,OAAO,SAAS,CAAC;IAAA,CACjB;IAED;;;;;OAKG;IACH,KAAK,CAAC,cAAc,CAAC,UAAqC,EAA4C;QACrG,MAAM,KAAK,GAAG,sBAAsB,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,IAAI,CAAC,KAAK,EAAE,EAAE,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC;QAC3F,IAAI,CAAC,KAAK;YAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,sBAAsB,EAAE,CAAC;QAEhE,UAAU,EAAE,CAAC,eAAe,KAAK,CAAC,IAAI,KAAG,CAAC,CAAC;QAC3C,MAAM,WAAW,GAAG,GAAG,gCAAgC,IAAI,6BAA6B,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC;QACzG,IAAI,QAAkB,CAAC;QACvB,IAAI,CAAC;YACJ,QAAQ,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAC3C,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,kBAAkB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACzG,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACpC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,uBAAuB,QAAQ,CAAC,MAAM,EAAE,EAAE,CAAC;QACvE,CAAC;QAED,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC;QAClC,SAAS,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAExC,UAAU,EAAE,CAAC,cAAc,KAAK,CAAC,IAAI,KAAG,CAAC,CAAC;QAC1C,MAAM,OAAO,GAAG,IAAI,CAAC,uBAAuB,IAAI,CAAC,CAAC,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,KAAK,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;QACjH,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,IAA2B,EAAE,OAAO,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;QAC3F,IAAI,CAAC,SAAS,CAAC,EAAE;YAAE,OAAO,SAAS,CAAC;QAEpC,UAAU,EAAE,CAAC,iCAA+B,CAAC,CAAC;QAC9C,MAAM,aAAa,GAAG,IAAI,CAAC,kBAAkB,CAAC,OAAO,CAAC,CAAC;QACvD,IAAI,CAAC,aAAa,EAAE,CAAC;YACpB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,uEAAuE,EAAE,CAAC;QACtG,CAAC;QAED,IAAI,CAAC,cAAc,CAAC,EAAE,OAAO,EAAE,6BAA6B,EAAE,aAAa,EAAE,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;QACvG,UAAU,EAAE,CAAC,oCAAoC,CAAC,CAAC;QACnD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;IAAA,CACpB;IAEO,KAAK,CAAC,eAAe,CAC5B,KAAe,EACf,OAAe,EACf,IAAoB,EACuB;QAC3C,IAAI,CAAC;YACJ,OAAO,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC;QAC3G,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,iBAAiB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACxG,CAAC;IAAA,CACD;IAEO,KAAK,CAAC,aAAa,CAAC,KAAe,EAAE,OAAe,EAA4C;QACvG,MAAM,OAAO,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,EAAE;YACjE,QAAQ,EAAE,OAAO,CAAC,QAAQ,KAAK,OAAO;YACtC,KAAK,EAAE,CAAC,MAAM,EAAE,QAAQ,EAAE,QAAQ,CAAC;SACnC,CAAC,CAAC;QACH,MAAM,qBAAqB,GAAG,IAAI,eAAe,EAAE,CAAC;QACpD,MAAM,WAAW,GAAG,kCAAkC,CAAC,OAAO,EAAE;YAC/D,MAAM,EAAE,qBAAqB,CAAC,MAAM;YACpC,SAAS,EAAE,qBAAqB;YAChC,WAAW,EAAE,qBAAqB;SAClC,CAAC,CAAC;QACH,IAAI,CAAC;YACJ,MAAM,QAAQ,CAAC,KAAK,EAAE,YAAY,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC;QACrD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,qBAAqB,CAAC,KAAK,EAAE,CAAC;YAC9B,MAAM,WAAW,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAC,CAAC,CAAC,CAAC;YAClC,MAAM,KAAK,CAAC;QACb,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,WAAW,CAAC;QACnC,IAAI,QAAQ,CAAC,MAAM,KAAK,SAAS,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;YAC1D,OAAO;gBACN,EAAE,EAAE,KAAK;gBACT,KAAK,EACJ,QAAQ,CAAC,MAAM,KAAK,SAAS;oBAC5B,CAAC,CAAC,gDAAgD,qBAAqB,IAAI;oBAC3E,CAAC,CAAC,sCAAsC,QAAQ,CAAC,IAAI,IAAI,SAAS,EAAE;aACtE,CAAC;QACH,CAAC;QACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;IAAA,CACpB;IAEO,KAAK,CAAC,WAAW,CAAC,KAAe,EAAE,OAAe,EAA4C;QACrG,sFAAoF;QACpF,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,EAAE,IAAI,EAAE,2BAA2B,OAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,CAAC;QAChG,MAAM,QAAQ,CAAC,KAAK,EAAE,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAAC;QAClD,IAAI,CAAC;YACJ,MAAM,cAAc,GAAG,IAAI,CAAC,SAAS,EAAE,KAAK,OAAO,IAAI,IAAI,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC;YACjG,MAAM,IAAI,GAAG,cAAc,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;YACzG,MAAM,IAAI,GAAG,YAAY,CAAC,cAAc,EAAE,IAAI,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,QAAQ,KAAK,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,CAAC,CAAC;YAC7G,MAAM,QAAQ,GAAG,MAAM,kCAAkC,CAAC,IAAI,EAAE;gBAC/D,SAAS,EAAE,qBAAqB;gBAChC,WAAW,EAAE,qBAAqB;aAClC,CAAC,CAAC;YACH,IAAI,QAAQ,CAAC,MAAM,KAAK,SAAS,IAAI,QAAQ,CAAC,IAAI,KAAK,CAAC,EAAE,CAAC;gBAC1D,OAAO;oBACN,EAAE,EAAE,KAAK;oBACT,KAAK,EACJ,QAAQ,CAAC,MAAM,KAAK,SAAS;wBAC5B,CAAC,CAAC,iBAAiB,cAAc,oBAAoB,qBAAqB,IAAI;wBAC9E,CAAC,CAAC,iBAAiB,cAAc,qBAAqB,QAAQ,CAAC,IAAI,IAAI,SAAS,EAAE;iBACpF,CAAC;YACH,CAAC;YACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;QACrB,CAAC;gBAAS,CAAC;YACV,IAAI,CAAC;gBACJ,MAAM,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;YAClC,CAAC;YAAC,MAAM,CAAC;gBACR,sBAAsB;YACvB,CAAC;QACF,CAAC;IAAA,CACD;IAEO,KAAK,CAAC,oBAAoB,CAAC,GAAW,EAA+B;QAC5E,IAAI,CAAC;YACJ,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,GAAG,EAAE;gBACvC,MAAM,EAAE,MAAM;gBACd,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,uBAAuB,CAAC;aACpD,CAAC,CAAC;YACH,OAAO,QAAQ,CAAC,EAAE,CAAC,CAAC,CAAC,kBAAkB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAC7F,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,SAAS,CAAC;QAClB,CAAC;IAAA,CACD;IAEO,cAAc,CAAC,IAAY,EAAsB;QACxD,IAAI,CAAC;YACJ,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC;QAC5B,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,SAAS,CAAC;QAClB,CAAC;IAAA,CACD;IAEO,eAAe,CAAC,IAAY,EAAQ;QAC3C,IAAI,CAAC;YACJ,MAAM,CAAC,IAAI,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAC/B,CAAC;QAAC,MAAM,CAAC;YACR,sBAAsB;QACvB,CAAC;IAAA,CACD;IAED;;;;;;OAMG;IACK,oBAAoB,CAAC,MAAuC,EAAiB;QACpF,IAAI,MAAM,CAAC,MAAM;YAAE,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC5C,OAAO,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC;YAC/B,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC;YACtC,MAAM,CAAC,OAAO,EAAE,CAAC;QAAA,CACjB,CAAC,CAAC;IAAA,CACH;IAED;;;;;OAKG;IACH,KAAK,CAAC,aAAa,CAClB,IAAoC,EACpC,UAAqC,EACN;QAC/B,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;QAC9D,MAAM,GAAG,GAAG,0BAA0B,IAAI,CAAC,IAAI,iBAAiB,IAAI,CAAC,IAAI,EAAE,CAAC;QAC5E,SAAS,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAElD,IAAI,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC5B,MAAM,UAAU,GAAG,MAAM,IAAI,CAAC,oBAAoB,CAAC,GAAG,CAAC,CAAC;YACxD,IAAI,UAAU,KAAK,SAAS,IAAI,UAAU,KAAK,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC9E,UAAU,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,wBAAwB,UAAU,uBAAqB,CAAC,CAAC;gBAClF,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;YACpD,CAAC;QACF,CAAC;QAED,UAAU,EAAE,CAAC,eAAe,IAAI,CAAC,IAAI,SAAS,IAAI,CAAC,IAAI,KAAG,CAAC,CAAC;QAC5D,IAAI,QAAkB,CAAC;QACvB,IAAI,CAAC;YACJ,QAAQ,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACnC,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,kBAAkB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACzG,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;YACpC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,uBAAuB,QAAQ,CAAC,MAAM,EAAE,EAAE,CAAC;QACvE,CAAC;QAED,MAAM,aAAa,GAAG,kBAAkB,CAAC,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC;QACjF,MAAM,WAAW,GAAG,iBAAiB,CAAC,QAAQ,CAAC,CAAC;QAChD,IAAI,CAAC;YACJ,MAAM,QAAQ,CAAC,QAAQ,CAAC,IAA2B,EAAE,WAAW,CAAC,CAAC;QACnE,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YAChB,MAAM,IAAI,CAAC,oBAAoB,CAAC,WAAW,CAAC,CAAC;YAC7C,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;YAC/B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,kBAAkB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACzG,CAAC;QAED,MAAM,WAAW,GAAG,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,CAAC;QAClD,IAAI,aAAa,KAAK,SAAS,IAAI,WAAW,KAAK,aAAa,EAAE,CAAC;YAClE,IAAI,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;YAC/B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,2BAA2B,aAAa,eAAe,WAAW,IAAI,CAAC,EAAE,EAAE,CAAC;QACxG,CAAC;QAED,UAAU,EAAE,CAAC,GAAG,IAAI,CAAC,IAAI,gBAAgB,WAAW,IAAI,CAAC,UAAU,CAAC,CAAC;QACrE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;IAAA,CACpC;IAEO,KAAK,CAAC,SAAS,CAAC,OAAe,EAAoB;QAC1D,IAAI,CAAC;YACJ,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,GAAG,OAAO,SAAS,EAAE;gBACvD,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,uBAAuB,CAAC;aACpD,CAAC,CAAC;YACH,OAAO,QAAQ,CAAC,EAAE,CAAC;QACpB,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,KAAK,CAAC;QACd,CAAC;IAAA,CACD;IAED;;;;OAIG;IACH,KAAK,CAAC,KAAK,CAAC,IAKX,EAA6B;QAC7B,MAAM,QAAQ,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QACtC,IAAI,CAAC,QAAQ;YAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,EAAE,CAAC;QAC7D,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,QAAQ,CAAC,aAAa,CAAC,CAAC;QACnE,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC;YAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,EAAE,CAAC;QAE7E,MAAM,OAAO,GAAG,oBAAoB,IAAI,CAAC,IAAI,EAAE,CAAC;QAChD,MAAM,IAAI,GAAG;YACZ,IAAI;YACJ,IAAI,CAAC,SAAS;YACd,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,UAAU,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YACzD,QAAQ;YACR,WAAW;YACX,QAAQ;YACR,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC;YACjB,IAAI;YACJ,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC;YACnB,GAAG,CAAC,QAAQ,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;SACtD,CAAC;QAEF,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,IAAI,EAAE;YAC3C,QAAQ,EAAE,OAAO,CAAC,QAAQ,KAAK,OAAO;YACtC,KAAK,EAAE,QAAQ;YACf,GAAG,EAAE,OAAO,CAAC,GAAG;SAChB,CAAC,CAAC;QACH,IAAI,KAAK,CAAC,GAAG;YAAE,qBAAqB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAChD,KAAK,CAAC,KAAK,EAAE,EAAE,CAAC;QAChB,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE,CAAC;YACtB,IAAI,CAAC,MAAM,GAAG,SAAS,CAAC;QAAA,CACxB,CAAC,CAAC;QACH,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC;QAEpB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,IAAI,CAAC,mBAAmB,EAAE,OAAO,EAAE,EAAE,CAAC;YACrE,IAAI,MAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC;gBAAE,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;YAChE,MAAM,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;QAC/C,CAAC;QACD,IAAI,CAAC,IAAI,EAAE,CAAC;QACZ,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,EAAE,CAAC;IAAA,CAC9C;IAED,IAAI,GAAyB;QAC5B,MAAM,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC;QAC1B,IAAI,CAAC,KAAK;YAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QACtC,IAAI,CAAC,MAAM,GAAG,SAAS,CAAC;QACxB,IAAI,KAAK,CAAC,GAAG,EAAE,CAAC;YACf,uBAAuB,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;YACnC,eAAe,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC5B,CAAC;QACD,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;IAAA,CACzB;IAED,SAAS,GAAY;QACpB,OAAO,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC;IAAA,CACjC;CACD","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport {\n\tcreateWriteStream,\n\ttype Dirent,\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trmSync,\n\tstatSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { arch as osArch, platform as osPlatform } from \"node:os\";\nimport { dirname, join, relative } from \"node:path\";\nimport type { Readable, Writable } from \"node:stream\";\nimport { pipeline } from \"node:stream/promises\";\nimport { spawnProcess, spawnProcessSync, waitForChildProcessWithTermination } from \"../../utils/child-process.ts\";\nimport { killProcessTree, trackDetachedChildPid, untrackDetachedChildPid } from \"../../utils/shell.ts\";\nimport { modelsDir as agentModelsDir, runtimesDir as agentRuntimesDir } from \"../agent-paths.ts\";\n\n/**\n * Managed runtime for prism-ml 1-bit GGUF models (Bonsai-27B first). The provider requires their\n * llama.cpp fork (Q1_0_g128 hybrid-attention kernels) — stock llama.cpp/Ollama cannot serve these\n * weights — so this module owns runtime install (prebuilt release download+extract), GGUF\n * downloads, and llama-server lifecycle. Mirrors local-runtime.ts's OllamaRuntime: injectable\n * seams for fetch/spawn/exists, pi-owned directories under agentDir (runtimes/, models/),\n * detached+tracked child processes, onProgress as best-effort UI feedback, honest error taxonomies\n * instead of silent fallbacks.\n *\n * The fork publishes precompiled per-platform archives on GitHub Releases (no local build): pin a\n * release tag, download the matching asset, extract it, and locate the `llama-server` binary once\n * at install time (release archive layouts vary — bin/ vs build/bin/ at the top level — so this is\n * a deterministic one-time scan, not a fallback). The found relative path and backend (cpu/cuda)\n * are persisted to a manifest so later `detect()`/`serve()` calls never need to re-scan.\n */\n\n/**\n * Pinned Prism ML llama.cpp fork release — verified via\n * `GET https://api.github.com/repos/PrismML-Eng/llama.cpp/releases/tags/prism-b9594-38c66ad` on\n * 2026-07-18 (release exists, all assets referenced below are present on it). Bump here (and\n * re-verify with the same API call) when the provider ships new kernels.\n */\nexport const PRISM_LLAMACPP_PINNED_RELEASE = \"prism-b9594-38c66ad\";\n\nexport const PRISM_LLAMACPP_RELEASES_BASE_URL = \"https://github.com/PrismML-Eng/llama.cpp/releases/download\";\n\nexport interface PrismModelDescriptor {\n\trepo: string;\n\tfile: string;\n\tmmprojFile?: string;\n\tdisplayName: string;\n}\n\n/** Curated first-model descriptor; the `/models add` wiring task consumes this. */\nexport const BONSAI_27B: PrismModelDescriptor = {\n\trepo: \"prism-ml/Bonsai-27B-gguf\",\n\tfile: \"Bonsai-27B-Q1_0.gguf\",\n\tmmprojFile: \"Bonsai-27B-mmproj-Q8_0.gguf\",\n\tdisplayName: \"Bonsai-27B (1-bit Q1_0 + vision)\",\n};\n\nexport type PrismBackend = \"cpu\" | \"cuda\";\nexport type PrismAssetKind = \"tar-gz\" | \"zip\";\n\nexport interface PrismLlamaAsset {\n\tname: string;\n\tkind: PrismAssetKind;\n\tbackend: PrismBackend;\n}\n\n// Verbatim asset names from the pinned release (see PRISM_LLAMACPP_PINNED_RELEASE doc comment) —\n// hardcoded rather than templated from the tag because the provider's naming isn't fully\n// consistent across platforms (Windows CPU assets drop the tag/commit segment entirely; the\n// Windows CUDA asset uses a different build-number segment, \"b1\" not \"b9594\").\nconst LINUX_X64_CPU_ASSET = \"llama-prism-b9594-38c66ad-bin-ubuntu-x64.tar.gz\";\nconst LINUX_ARM64_CPU_ASSET = \"llama-prism-b9594-38c66ad-bin-ubuntu-arm64.tar.gz\";\nconst LINUX_X64_CUDA_ASSET = \"llama-prism-b9594-38c66ad-bin-linux-cuda-12.4-x64.tar.gz\";\nconst MACOS_ARM64_ASSET = \"llama-prism-b9594-38c66ad-bin-macos-arm64.tar.gz\";\nconst MACOS_X64_ASSET = \"llama-prism-b9594-38c66ad-bin-macos-x64.tar.gz\";\nconst WIN_X64_CPU_ASSET = \"llama-bin-win-cpu-x64.zip\";\nconst WIN_ARM64_CPU_ASSET = \"llama-bin-win-cpu-arm64.zip\";\n\n/**\n * Maps a platform/arch/GPU triple to the exact Prism llama.cpp release asset for\n * {@link PRISM_LLAMACPP_PINNED_RELEASE} — verified against the real GitHub release, not guessed.\n * CPU asset by default; the CUDA 12.4 variant is only offered for linux x64 with an NVIDIA GPU\n * present (a 12.8 variant also exists upstream but has no wired caller yet). Windows CUDA needs a\n * companion `cudart-*` archive and is out of scope for v1 — Windows always resolves to the CPU zip\n * regardless of `hasNvidiaGpu`. Pure and exported so it's independently testable.\n */\nexport function resolvePrismLlamaAsset(\n\tplat: string,\n\tarchitecture: string,\n\thasNvidiaGpu: boolean,\n): PrismLlamaAsset | undefined {\n\tif (plat === \"linux\") {\n\t\tif (architecture === \"x64\") {\n\t\t\treturn hasNvidiaGpu\n\t\t\t\t? { name: LINUX_X64_CUDA_ASSET, kind: \"tar-gz\", backend: \"cuda\" }\n\t\t\t\t: { name: LINUX_X64_CPU_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\t}\n\t\tif (architecture === \"arm64\") return { name: LINUX_ARM64_CPU_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\tif (plat === \"darwin\") {\n\t\tif (architecture === \"arm64\") return { name: MACOS_ARM64_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\tif (architecture === \"x64\") return { name: MACOS_X64_ASSET, kind: \"tar-gz\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\tif (plat === \"win32\") {\n\t\tif (architecture === \"x64\") return { name: WIN_X64_CPU_ASSET, kind: \"zip\", backend: \"cpu\" };\n\t\tif (architecture === \"arm64\") return { name: WIN_ARM64_CPU_ASSET, kind: \"zip\", backend: \"cpu\" };\n\t\treturn undefined;\n\t}\n\treturn undefined;\n}\n\nexport interface PrismDetectResult {\n\truntimeInstalled: boolean;\n\tbinaryPath?: string;\n\trelease?: string;\n}\n\nexport interface PrismDownloadResult {\n\tok: boolean;\n\tpath?: string;\n\tskipped?: boolean;\n\terror?: string;\n}\n\nexport type PrismServeResult = { ok: true; baseUrl: string } | { ok: false; error: string };\n\ninterface PrismInstallManifest {\n\trelease: string;\n\tbinaryRelPath: string;\n\tbackend: PrismBackend;\n}\n\ntype PrismSpawnOptions = { detached?: boolean; stdio?: \"ignore\"; env?: NodeJS.ProcessEnv };\ntype PrismSpawnFn = (\n\tcommand: string,\n\targs: string[],\n\toptions: PrismSpawnOptions,\n) => Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\">;\ntype PrismExtractArchiveFn = (\n\tinput: Readable,\n\tdestDir: string,\n\tkind: PrismAssetKind,\n) => Promise<{ ok: boolean; error?: string }>;\n\nexport interface PrismLlamaCppDeps {\n\tfetchFn?: typeof fetch;\n\tspawnFn?: PrismSpawnFn;\n\texistsFn?: (path: string) => boolean;\n\tsleepFn?: (ms: number) => Promise<void>;\n\t/** Whether a named command exists on PATH (nvidia-smi, tar — for zip extraction on Windows). */\n\thasCommand?: (command: string) => boolean;\n\t/** Decided once at install time and persisted as `backend` — serve() reads the persisted value\n\t * so `-ngl 99` reflects what was actually installed, not the host's current GPU state. */\n\thasNvidiaGpu?: () => boolean;\n\tplatform?: () => string;\n\tarch?: () => string;\n\t/** Runs the extraction step for a downloaded archive. Injectable so installManaged's\n\t * download->extract->scan orchestration is testable without a real tar/unzip pipeline; defaults\n\t * to the real spawn-based extractor. */\n\textractArchive?: PrismExtractArchiveFn;\n\t/** Health-poll bounds for serve(); default ~120s (240 * 500ms), both overridable for tests. */\n\thealthPollAttempts?: number;\n\thealthPollIntervalMs?: number;\n}\n\nconst EXTRACTION_TIMEOUT_MS = 10 * 60_000;\nconst COMMAND_KILL_GRACE_MS = 2_000;\nconst DEFAULT_HEALTH_POLL_ATTEMPTS = 240;\nconst DEFAULT_HEALTH_POLL_INTERVAL_MS = 500;\nconst HEALTH_CHECK_TIMEOUT_MS = 2_000;\nconst HEAD_REQUEST_TIMEOUT_MS = 10_000;\n\nfunction parseContentLength(header: string | null): number | undefined {\n\tif (!header) return undefined;\n\tconst value = Number(header);\n\treturn Number.isFinite(value) && value >= 0 ? value : undefined;\n}\n\n/**\n * `spawnProcess(..., { stdio: [\"pipe\", ...] })` always yields a non-null `stdin` at runtime, but\n * child-process.ts's typed overload only narrows `.stdin` for the capture-only stdio shape used\n * elsewhere. This makes the true invariant explicit and fails loudly instead of silently\n * 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\nexport class PrismLlamaCppRuntime {\n\tprivate readonly _agentDir: string;\n\tprivate readonly _fetch: typeof fetch;\n\tprivate readonly _spawn: PrismSpawnFn;\n\tprivate readonly _exists: (path: string) => boolean;\n\tprivate readonly _sleep: (ms: number) => Promise<void>;\n\tprivate readonly _hasCommand: (command: string) => boolean;\n\tprivate readonly _hasNvidiaGpu: () => boolean;\n\tprivate readonly _platform: () => string;\n\tprivate readonly _arch: () => string;\n\tprivate readonly _extractArchiveOverride: PrismExtractArchiveFn | undefined;\n\tprivate readonly _healthPollAttempts: number;\n\tprivate readonly _healthPollIntervalMs: number;\n\tprivate _child: Pick<ChildProcess, \"pid\" | \"kill\" | \"unref\" | \"on\"> | undefined;\n\n\tconstructor(args: { agentDir: string; deps?: PrismLlamaCppDeps }) {\n\t\tthis._agentDir = args.agentDir;\n\t\tthis._fetch = args.deps?.fetchFn ?? fetch;\n\t\tthis._spawn =\n\t\t\targs.deps?.spawnFn ?? ((command, argv, options) => spawn(command, argv, { ...options, stdio: \"ignore\" }));\n\t\tthis._exists = args.deps?.existsFn ?? existsSync;\n\t\tthis._sleep = args.deps?.sleepFn ?? ((ms) => new Promise((resolve) => setTimeout(resolve, ms)));\n\t\tthis._hasCommand =\n\t\t\targs.deps?.hasCommand ??\n\t\t\t((command) =>\n\t\t\t\tspawnProcessSync(command, [\"--version\"], { encoding: \"utf8\", timeout: 5_000 }).error === undefined);\n\t\tthis._hasNvidiaGpu = args.deps?.hasNvidiaGpu ?? (() => this._hasCommand(\"nvidia-smi\"));\n\t\tthis._platform = args.deps?.platform ?? osPlatform;\n\t\tthis._arch = args.deps?.arch ?? osArch;\n\t\tthis._extractArchiveOverride = args.deps?.extractArchive;\n\t\tthis._healthPollAttempts = args.deps?.healthPollAttempts ?? DEFAULT_HEALTH_POLL_ATTEMPTS;\n\t\tthis._healthPollIntervalMs = args.deps?.healthPollIntervalMs ?? DEFAULT_HEALTH_POLL_INTERVAL_MS;\n\t}\n\n\truntimeDir(): string {\n\t\treturn agentRuntimesDir(\"prism-llamacpp\", this._agentDir);\n\t}\n\n\tmodelsDir(): string {\n\t\treturn agentModelsDir(\"llamacpp\", this._agentDir);\n\t}\n\n\tprivate _binaryName(): string {\n\t\treturn this._platform() === \"win32\" ? \"llama-server.exe\" : \"llama-server\";\n\t}\n\n\tprivate _manifestPath(): string {\n\t\treturn join(this.runtimeDir(), \"install.json\");\n\t}\n\n\tprivate _readManifest(): PrismInstallManifest | undefined {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(readFileSync(this._manifestPath(), \"utf8\")) as {\n\t\t\t\trelease?: unknown;\n\t\t\t\tbinaryRelPath?: unknown;\n\t\t\t\tbackend?: unknown;\n\t\t\t};\n\t\t\tif (\n\t\t\t\ttypeof parsed.release === \"string\" &&\n\t\t\t\ttypeof parsed.binaryRelPath === \"string\" &&\n\t\t\t\t(parsed.backend === \"cpu\" || parsed.backend === \"cuda\")\n\t\t\t) {\n\t\t\t\treturn { release: parsed.release, binaryRelPath: parsed.binaryRelPath, backend: parsed.backend };\n\t\t\t}\n\t\t\treturn undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _writeManifest(manifest: PrismInstallManifest): void {\n\t\twriteFileSync(this._manifestPath(), JSON.stringify(manifest, null, \"\\t\"));\n\t}\n\n\tasync detect(): Promise<PrismDetectResult> {\n\t\tconst manifest = this._readManifest();\n\t\tif (!manifest) return { runtimeInstalled: false };\n\t\tconst binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);\n\t\tconst runtimeInstalled = this._exists(binaryPath);\n\t\treturn {\n\t\t\truntimeInstalled,\n\t\t\tbinaryPath: runtimeInstalled ? binaryPath : undefined,\n\t\t\trelease: manifest.release,\n\t\t};\n\t}\n\n\t/** One-time recursive scan for the `llama-server` binary inside a freshly extracted release\n\t * archive — layouts vary (bin/ vs build/bin/ at the top level) across platforms/backends, so\n\t * this locates it deterministically instead of assuming a fixed path, then the result is\n\t * persisted so later calls never need to re-scan. */\n\tprivate _findBinaryRelPath(root: string): string | undefined {\n\t\tconst targetName = this._binaryName();\n\t\tconst stack: string[] = [root];\n\t\twhile (stack.length > 0) {\n\t\t\tconst dir = stack.pop();\n\t\t\tif (dir === undefined) continue;\n\t\t\tlet entries: Dirent[];\n\t\t\ttry {\n\t\t\t\tentries = readdirSync(dir, { withFileTypes: true });\n\t\t\t} catch {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tfor (const entry of entries) {\n\t\t\t\tconst full = join(dir, entry.name);\n\t\t\t\tif (entry.isDirectory()) stack.push(full);\n\t\t\t\telse if (entry.isFile() && entry.name === targetName) return relative(root, full);\n\t\t\t}\n\t\t}\n\t\treturn undefined;\n\t}\n\n\t/**\n\t * Download the pinned release asset for this host and extract it (consent-gated by the caller,\n\t * same contract as OllamaRuntime#installManaged — this method only does the mechanical\n\t * download+extract+locate and reports the outcome honestly). No compiler toolchain required:\n\t * the fork ships prebuilt binaries.\n\t */\n\tasync installManaged(onProgress?: (status: string) => void): Promise<{ ok: boolean; error?: string }> {\n\t\tconst asset = resolvePrismLlamaAsset(this._platform(), this._arch(), this._hasNvidiaGpu());\n\t\tif (!asset) return { ok: false, error: \"unsupported-platform\" };\n\n\t\tonProgress?.(`Downloading ${asset.name}…`);\n\t\tconst downloadUrl = `${PRISM_LLAMACPP_RELEASES_BASE_URL}/${PRISM_LLAMACPP_PINNED_RELEASE}/${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 = this.runtimeDir();\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 Readable, destDir, asset.kind);\n\t\tif (!extracted.ok) return extracted;\n\n\t\tonProgress?.(\"Locating llama-server binary…\");\n\t\tconst binaryRelPath = this._findBinaryRelPath(destDir);\n\t\tif (!binaryRelPath) {\n\t\t\treturn { ok: false, error: \"binary-missing: no llama-server binary found in the extracted archive\" };\n\t\t}\n\n\t\tthis._writeManifest({ release: PRISM_LLAMACPP_PINNED_RELEASE, binaryRelPath, backend: asset.backend });\n\t\tonProgress?.(\"Prism llama.cpp runtime installed.\");\n\t\treturn { ok: true };\n\t}\n\n\tprivate async _extractArchive(\n\t\tinput: Readable,\n\t\tdestDir: string,\n\t\tkind: PrismAssetKind,\n\t): Promise<{ ok: boolean; error?: string }> {\n\t\ttry {\n\t\t\treturn kind === \"zip\" ? await this._extractZip(input, destDir) : await this._extractTarGz(input, destDir);\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 _extractTarGz(input: Readable, destDir: string): Promise<{ ok: boolean; error?: string }> {\n\t\tconst tarProc = spawnProcess(\"tar\", [\"-xzf\", \"-\", \"-C\", destDir], {\n\t\t\tdetached: process.platform !== \"win32\",\n\t\t\tstdio: [\"pipe\", \"ignore\", \"ignore\"],\n\t\t});\n\t\tconst terminationController = new AbortController();\n\t\tconst processWait = waitForChildProcessWithTermination(tarProc, {\n\t\t\tsignal: terminationController.signal,\n\t\t\ttimeoutMs: EXTRACTION_TIMEOUT_MS,\n\t\t\tkillGraceMs: COMMAND_KILL_GRACE_MS,\n\t\t});\n\t\ttry {\n\t\t\tawait pipeline(input, requireStdin(tarProc, \"tar\"));\n\t\t} catch (error) {\n\t\t\tterminationController.abort();\n\t\t\tawait processWait.catch(() => {});\n\t\t\tthrow error;\n\t\t}\n\t\tconst terminal = await processWait;\n\t\tif (terminal.reason === \"timeout\" || terminal.code !== 0) {\n\t\t\treturn {\n\t\t\t\tok: false,\n\t\t\t\terror:\n\t\t\t\t\tterminal.reason === \"timeout\"\n\t\t\t\t\t\t? `extract-fail: tar extraction timed out after ${EXTRACTION_TIMEOUT_MS}ms`\n\t\t\t\t\t\t: `extract-fail: tar exited with code ${terminal.code ?? \"unknown\"}`,\n\t\t\t};\n\t\t}\n\t\treturn { ok: true };\n\t}\n\n\tprivate async _extractZip(input: Readable, destDir: string): Promise<{ ok: boolean; error?: string }> {\n\t\t// Zip's central directory needs seekable file access — buffer to a temp file first.\n\t\tconst zipPath = join(destDir, \"..\", `prism-llamacpp-download-${process.pid}-${Date.now()}.zip`);\n\t\tawait pipeline(input, 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, { detached: process.platform !== \"win32\", stdio: \"ignore\" });\n\t\t\tconst terminal = await waitForChildProcessWithTermination(proc, {\n\t\t\t\ttimeoutMs: EXTRACTION_TIMEOUT_MS,\n\t\t\t\tkillGraceMs: COMMAND_KILL_GRACE_MS,\n\t\t\t});\n\t\t\tif (terminal.reason === \"timeout\" || terminal.code !== 0) {\n\t\t\t\treturn {\n\t\t\t\t\tok: false,\n\t\t\t\t\terror:\n\t\t\t\t\t\tterminal.reason === \"timeout\"\n\t\t\t\t\t\t\t? `extract-fail: ${extractCommand} timed out after ${EXTRACTION_TIMEOUT_MS}ms`\n\t\t\t\t\t\t\t: `extract-fail: ${extractCommand} exited with code ${terminal.code ?? \"unknown\"}`,\n\t\t\t\t};\n\t\t\t}\n\t\t\treturn { ok: true };\n\t\t} finally {\n\t\t\ttry {\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\tprivate async _remoteContentLength(url: string): Promise<number | undefined> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(url, {\n\t\t\t\tmethod: \"HEAD\",\n\t\t\t\tsignal: AbortSignal.timeout(HEAD_REQUEST_TIMEOUT_MS),\n\t\t\t});\n\t\t\treturn response.ok ? parseContentLength(response.headers.get(\"content-length\")) : undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _fileSizeBytes(path: string): number | undefined {\n\t\ttry {\n\t\t\treturn statSync(path).size;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate _cleanupPartial(path: string): void {\n\t\ttry {\n\t\t\trmSync(path, { force: true });\n\t\t} catch {\n\t\t\t// best-effort cleanup\n\t\t}\n\t}\n\n\t/**\n\t * Wait until a write stream's underlying fd is actually closed. `pipeline()` destroys the\n\t * stream on error but its promise can settle before the fd close completes (Windows in\n\t * particular keeps the file handle open a tick longer) — unlinking before that races with the\n\t * OS and can leave the \"partial\" file behind or fail the delete outright. Resolves immediately\n\t * if the stream already reports `closed`.\n\t */\n\tprivate _waitForStreamClosed(stream: Writable & { closed?: boolean }): Promise<void> {\n\t\tif (stream.closed) return Promise.resolve();\n\t\treturn new Promise((resolve) => {\n\t\t\tstream.once(\"close\", () => resolve());\n\t\t\tstream.destroy();\n\t\t});\n\t}\n\n\t/**\n\t * Stream a GGUF (or mmproj) file from Hugging Face into pi's owned models dir. Skips a re-download\n\t * when the local file already matches the remote size; verifies size when the response reports\n\t * `content-length` and deletes the partial file on any failure or mismatch — never leaves a\n\t * corrupt/truncated weight file behind for a later load to silently misread.\n\t */\n\tasync downloadModel(\n\t\targs: { repo: string; file: string },\n\t\tonProgress?: (status: string) => void,\n\t): Promise<PrismDownloadResult> {\n\t\tconst destPath = join(this.modelsDir(), args.repo, args.file);\n\t\tconst url = `https://huggingface.co/${args.repo}/resolve/main/${args.file}`;\n\t\tmkdirSync(dirname(destPath), { recursive: true });\n\n\t\tif (this._exists(destPath)) {\n\t\t\tconst remoteSize = await this._remoteContentLength(url);\n\t\t\tif (remoteSize !== undefined && remoteSize === this._fileSizeBytes(destPath)) {\n\t\t\t\tonProgress?.(`${args.file} already downloaded (${remoteSize} bytes) — skipping.`);\n\t\t\t\treturn { ok: true, path: destPath, skipped: true };\n\t\t\t}\n\t\t}\n\n\t\tonProgress?.(`Downloading ${args.file} from ${args.repo}…`);\n\t\tlet response: Response;\n\t\ttry {\n\t\t\tresponse = await this._fetch(url);\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 expectedBytes = parseContentLength(response.headers.get(\"content-length\"));\n\t\tconst writeStream = createWriteStream(destPath);\n\t\ttry {\n\t\t\tawait pipeline(response.body as unknown as Readable, writeStream);\n\t\t} catch (error) {\n\t\t\tawait this._waitForStreamClosed(writeStream);\n\t\t\tthis._cleanupPartial(destPath);\n\t\t\treturn { ok: false, error: `download-fail: ${error instanceof Error ? error.message : String(error)}` };\n\t\t}\n\n\t\tconst actualBytes = this._fileSizeBytes(destPath);\n\t\tif (expectedBytes !== undefined && actualBytes !== expectedBytes) {\n\t\t\tthis._cleanupPartial(destPath);\n\t\t\treturn { ok: false, error: `size-mismatch: expected ${expectedBytes} bytes, got ${actualBytes ?? 0}` };\n\t\t}\n\n\t\tonProgress?.(`${args.file} downloaded (${actualBytes ?? 0} bytes).`);\n\t\treturn { ok: true, path: destPath };\n\t}\n\n\tprivate async _healthUp(baseUrl: string): Promise<boolean> {\n\t\ttry {\n\t\t\tconst response = await this._fetch(`${baseUrl}/health`, {\n\t\t\t\tsignal: AbortSignal.timeout(HEALTH_CHECK_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\t/**\n\t * Spawn `llama-server` detached+tracked (killed on parent shutdown even if pi crashes without\n\t * calling stop()) and poll `/health` until ready. `-ngl 99` is only added when the installed\n\t * asset's persisted `backend` is `\"cuda\"`, not from the host's current GPU state.\n\t */\n\tasync serve(args: {\n\t\tmodelPath: string;\n\t\tmmprojPath?: string;\n\t\tport: number;\n\t\tnumCtx: number;\n\t}): Promise<PrismServeResult> {\n\t\tconst manifest = this._readManifest();\n\t\tif (!manifest) return { ok: false, error: \"binary-missing\" };\n\t\tconst binaryPath = join(this.runtimeDir(), manifest.binaryRelPath);\n\t\tif (!this._exists(binaryPath)) return { ok: false, error: \"binary-missing\" };\n\n\t\tconst baseUrl = `http://127.0.0.1:${args.port}`;\n\t\tconst argv = [\n\t\t\t\"-m\",\n\t\t\targs.modelPath,\n\t\t\t...(args.mmprojPath ? [\"--mmproj\", args.mmprojPath] : []),\n\t\t\t\"--host\",\n\t\t\t\"127.0.0.1\",\n\t\t\t\"--port\",\n\t\t\tString(args.port),\n\t\t\t\"-c\",\n\t\t\tString(args.numCtx),\n\t\t\t...(manifest.backend === \"cuda\" ? [\"-ngl\", \"99\"] : []),\n\t\t];\n\n\t\tconst child = this._spawn(binaryPath, argv, {\n\t\t\tdetached: process.platform !== \"win32\",\n\t\t\tstdio: \"ignore\",\n\t\t\tenv: process.env,\n\t\t});\n\t\tif (child.pid) trackDetachedChildPid(child.pid);\n\t\tchild.unref?.();\n\t\tchild.on(\"exit\", () => {\n\t\t\tthis._child = undefined;\n\t\t});\n\t\tthis._child = child;\n\n\t\tfor (let attempt = 0; attempt < this._healthPollAttempts; attempt++) {\n\t\t\tif (await this._healthUp(baseUrl)) return { ok: true, baseUrl };\n\t\t\tawait this._sleep(this._healthPollIntervalMs);\n\t\t}\n\t\tthis.stop();\n\t\treturn { ok: false, error: \"health-timeout\" };\n\t}\n\n\tstop(): { stopped: boolean } {\n\t\tconst child = this._child;\n\t\tif (!child) return { stopped: false };\n\t\tthis._child = undefined;\n\t\tif (child.pid) {\n\t\t\tuntrackDetachedChildPid(child.pid);\n\t\t\tkillProcessTree(child.pid);\n\t\t}\n\t\treturn { stopped: true };\n\t}\n\n\tisRunning(): boolean {\n\t\treturn this._child !== undefined;\n\t}\n}\n"]}