@caupulican/pi-adaptative 0.85.3 → 0.85.4

package/dist/core/models/model-ref.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-ref.js","sourceRoot":"","sources":["../../../src/core/models/model-ref.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAUnD,MAAM,UAAU,GAAG,6CAA6C,CAAC;AACjE,MAAM,MAAM,GAAG,gDAAgD,CAAC;AAChE,MAAM,MAAM,GAAG,wEAAwE,CAAC;AACxF,MAAM,eAAe,GAAG,iBAAiB,CAAC;AAC1C,MAAM,iCAAiC,GAAG,IAAI,GAAG,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC;AAC3E;;;;;;;GAOG;AACH,MAAM,wBAAwB,GAAG,IAAI,GAAG,CAAiB,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAC7G;;;;;;;GAOG;AACH,MAAM,eAAe,GAAG,uBAAuB,CAAC;AAEhD,SAAS,6BAA6B,CAAC,GAAW,EAAE,IAAY,EAAE,KAAc;IAC/E,MAAM,OAAO,GAAG,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC;IACjC,MAAM,YAAY,GAAG,wBAAwB,CAAC,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC;IACzE,IAAI,YAAY,EAAE,CAAC;QAClB,OAAO,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,YAAY,EAAE,GAAG,EAAE,SAAS,YAAY,EAAE,EAAE,CAAC;IACxF,CAAC;IACD,IAAI,CAAC,KAAK,IAAI,OAAO,CAAC,WAAW,EAAE,KAAK,eAAe,CAAC,WAAW,EAAE,EAAE,CAAC;QACvE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,SAAS,eAAe,EAAE,EAAE,CAAC;IAC5D,CAAC;IACD,IAAI,CAAC,KAAK,IAAI,iCAAiC,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,OAAO,EAAE,IAAI,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,OAAO,EAAE,EAAE,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC;AAClF,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,QAAgB;IACpD,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC9B,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC;IAC3E,IAAI,KAAK,CAAC,MAAM,GAAG,GAAG;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wCAAwC,EAAE,CAAC;IAEtG,yEAAyE;IACzE,MAAM,cAAc,GAAG,iCAAiC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACrE,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,QAAQ,GAAG,cAAc,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACjE,IAAI,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpC,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wDAAwD,EAAE,CAAC;QAC/F,CAAC;QACD,MAAM,KAAK,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAC7C,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;YAAE,OAAO,KAAK,CAAC;QACzC,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,8DAA8D,EAAE,CAAC;IACrG,CAAC;IAED,IAAI,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QACrD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wEAAwE,EAAE,CAAC;IAC/G,CAAC;IAED,+FAA+F;IAC/F,+FAA+F;IAC/F,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACjC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,6BAA6B,CAAC,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC;IAC5D,CAAC;IAED,yBAAyB;IACzB,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACjC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,6BAA6B,CAAC,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC3B,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,8DAA8D,EAAE,CAAC;IACrG,CAAC;IAED,sFAAsF;IACtF,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACzB,MAAM,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC7B,IAAI,QAAQ,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,QAAQ,IAAI,KAAK,EAAE,EAAE,CAAC;QACrD,CAAC;QACD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,mDAAmD,EAAE,CAAC;IAC1F,CAAC;IAED,gEAAgE;IAChE,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5B,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAC1C,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,kCAAkC,EAAE,CAAC;AACzE,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,0BAA0B,CAAC,GAAW,EAAE,aAAqB;IAC5E,IAAI,GAAG,KAAK,aAAa;QAAE,OAAO,IAAI,CAAC;IACvC,MAAM,MAAM,GAAG,CAAC,KAAa,EAAW,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,aAAa,KAAK,GAAG,GAAG,SAAS;QAAE,OAAO,IAAI,CAAC;IACnE,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,GAAG,KAAK,GAAG,aAAa,SAAS;QAAE,OAAO,IAAI,CAAC;IAC7E,OAAO,KAAK,CAAC;AACd,CAAC","sourcesContent":["/**\n * Model source normalizer (local-model-lifecycle-design.md, \"one panel, three types\"):\n * everything a user can paste — an ollama tag, an hf.co GGUF ref, a full HuggingFace URL, a\n * copied `ollama pull ...` install command, or an API `provider/model` name — normalizes to ONE\n * typed source. Pure string work: pasted install commands are PARSED for their reference and\n * NEVER executed as shell; unknown forms are rejected with the reason.\n */\n\nimport { BONSAI_27B } from \"./llamacpp-runtime.ts\";\n\nexport type ModelSource =\n\t| { type: \"api\"; ref: string }\n\t| { type: \"local\"; pullRef: string }\n\t| { type: \"transformers\"; modelId: string; ref: string }\n\t| { type: \"prism-llamacpp\"; modelId: string; ref: string }\n\t| { type: \"needle\"; ref: string }\n\t| { type: \"rejected\"; reason: string };\n\nconst OLLAMA_TAG = /^[a-z0-9][a-z0-9._-]*(?::[A-Za-z0-9._-]+)?$/;\nconst HF_REF = /^hf\\.co\\/([\\w.-]+)\\/([\\w.-]+)(?::([\\w.-]+))?$/i;\nconst HF_URL = /^https?:\\/\\/(?:www\\.)?huggingface\\.co\\/([\\w.-]+)\\/([\\w.-]+)(?:\\/.*)?$/i;\nconst SHELL_METACHARS = /[;&|`$<>(){}\\\\]/;\nconst PI_MANAGED_TRANSFORMERS_MODEL_IDS = new Set([\"openbmb/MiniCPM5-1B\"]);\n/**\n * Curated prism-ml models that require pi's managed prism llama.cpp fork (see llamacpp-runtime.ts's\n * module docstring: stock llama.cpp/Ollama cannot serve these weights). These GGUF refs route to the\n * \"prism-llamacpp\" source instead of Ollama's local pull path regardless of a supplied quant suffix\n * — the curated descriptor already pins the exact quant file to fetch, so a quant on the input ref is\n * accepted but not required. Keyed lowercase for case-insensitive matching; values are the\n * canonical-cased modelId, sourced from the descriptor itself so the two never drift apart.\n */\nconst PRISM_LLAMACPP_MODEL_IDS = new Map<string, string>([[BONSAI_27B.repo.toLowerCase(), BONSAI_27B.repo]]);\n/**\n * The single curated needle model id (see needle-runtime.ts) — a standalone function-call test\n * bench with no GGUF/quant concept at all (a pip-installed Python package + one pinned pickle\n * checkpoint), so unlike the prism route above, a quant suffix on the input is NOT accepted — it\n * falls through to the ordinary local/Ollama route instead, mirroring the transformers route's\n * `!quant` gate. Only this exact repo routes to \"needle\"; any other Cactus-Compute/* ref keeps the\n * ordinary local route.\n */\nconst NEEDLE_MODEL_ID = \"Cactus-Compute/needle\";\n\nfunction normalizeHuggingFaceReference(org: string, repo: string, quant?: string): ModelSource {\n\tconst modelId = `${org}/${repo}`;\n\tconst prismModelId = PRISM_LLAMACPP_MODEL_IDS.get(modelId.toLowerCase());\n\tif (prismModelId) {\n\t\treturn { type: \"prism-llamacpp\", modelId: prismModelId, ref: `hf.co/${prismModelId}` };\n\t}\n\tif (!quant && modelId.toLowerCase() === NEEDLE_MODEL_ID.toLowerCase()) {\n\t\treturn { type: \"needle\", ref: `hf.co/${NEEDLE_MODEL_ID}` };\n\t}\n\tif (!quant && PI_MANAGED_TRANSFORMERS_MODEL_IDS.has(modelId)) {\n\t\treturn { type: \"transformers\", modelId, ref: `hf.co/${modelId}` };\n\t}\n\treturn { type: \"local\", pullRef: `hf.co/${modelId}${quant ? `:${quant}` : \"\"}` };\n}\n\nexport function normalizeModelSource(rawInput: string): ModelSource {\n\tconst input = rawInput.trim();\n\tif (input.length === 0) return { type: \"rejected\", reason: \"empty input\" };\n\tif (input.length > 500) return { type: \"rejected\", reason: \"input too long to be a model reference\" };\n\n\t// Pasted install command: extract the reference, never execute anything.\n\tconst installCommand = /^ollama\\s+(?:pull|run)\\s+(.+)$/i.exec(input);\n\tif (installCommand) {\n\t\tconst argument = installCommand[1]!.trim().split(/\\s+/)[0] ?? \"\";\n\t\tif (SHELL_METACHARS.test(argument)) {\n\t\t\treturn { type: \"rejected\", reason: \"install command argument contains shell metacharacters\" };\n\t\t}\n\t\tconst inner = normalizeModelSource(argument);\n\t\tif (inner.type === \"local\") return inner;\n\t\treturn { type: \"rejected\", reason: `could not extract a model reference from the install command` };\n\t}\n\n\tif (SHELL_METACHARS.test(input) || /\\s/.test(input)) {\n\t\treturn { type: \"rejected\", reason: \"not a recognized model reference (contains spaces or shell characters)\" };\n\t}\n\n\t// Full HuggingFace URL -> either a pi-managed Transformers model (curated full-base ids) or an\n\t// Ollama hf.co pull ref (GGUF/quant refs still belong to Ollama and keep their :quant suffix).\n\tconst hfUrl = HF_URL.exec(input);\n\tif (hfUrl) {\n\t\treturn normalizeHuggingFaceReference(hfUrl[1]!, hfUrl[2]!);\n\t}\n\n\t// hf.co/org/repo[:quant]\n\tconst hfRef = HF_REF.exec(input);\n\tif (hfRef) {\n\t\treturn normalizeHuggingFaceReference(hfRef[1]!, hfRef[2]!, hfRef[3]);\n\t}\n\n\tif (input.includes(\"://\")) {\n\t\treturn { type: \"rejected\", reason: \"only huggingface.co URLs are recognized as local model links\" };\n\t}\n\n\t// provider/model -> API-registered model (nothing to install; auth + selection only).\n\tif (input.includes(\"/\")) {\n\t\tconst [provider, ...rest] = input.split(\"/\");\n\t\tconst model = rest.join(\"/\");\n\t\tif (provider && model && !model.includes(\"/\")) {\n\t\t\treturn { type: \"api\", ref: `${provider}/${model}` };\n\t\t}\n\t\treturn { type: \"rejected\", reason: \"expected provider/model or hf.co/org/repo[:quant]\" };\n\t}\n\n\t// Bare ollama tag (\"qwen3:1.7b\", \"pi-lifter:latest\", \"llama3\").\n\tif (OLLAMA_TAG.test(input)) {\n\t\treturn { type: \"local\", pullRef: input };\n\t}\n\n\treturn { type: \"rejected\", reason: \"not a recognized model reference\" };\n}\n\n/**\n * Whether an installed-model listing entry (`installedName`, as `OllamaRuntime.list()` reports it)\n * is the same model as `ref` (a user-typed reference, or a `models.json`-registered id). Ollama\n * always reports installed models with an explicit tag — a pull with no tag is normalized to\n * `:latest` server-side — but a `ref` is frequently bare (\"llama3\"), so an exact string compare\n * alone misses that pairing. Exact match always counts; otherwise treat a bare ref against its\n * `:latest`-tagged listing as the same model, and the reverse (a `ref` explicitly pinned to\n * `:latest` against a bare listing), since the local-runtime interface is kept runtime-agnostic\n * (see local-runtime.ts's module docstring) and a future/alternate runtime could list bare names.\n * Shared by every ref-vs-installed-list comparison — see `AgentSession._warnIfManualModelChoiceIsRisky`\n * and `removeLocalModel` in interactive-mode.ts, which had the identical bug independently.\n */\nexport function matchesInstalledLocalModel(ref: string, installedName: string): boolean {\n\tif (ref === installedName) return true;\n\tconst hasTag = (value: string): boolean => value.includes(\":\");\n\tif (!hasTag(ref) && installedName === `${ref}:latest`) return true;\n\tif (!hasTag(installedName) && ref === `${installedName}:latest`) return true;\n\treturn false;\n}\n"]}
+{"version":3,"file":"model-ref.js","sourceRoot":"","sources":["../../../src/core/models/model-ref.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AAEH,OAAO,EAAE,gCAAgC,EAAE,MAAM,8BAA8B,CAAC;AAUhF,MAAM,UAAU,GAAG,6CAA6C,CAAC;AACjE,MAAM,MAAM,GAAG,gDAAgD,CAAC;AAChE,MAAM,MAAM,GAAG,wEAAwE,CAAC;AACxF,MAAM,eAAe,GAAG,iBAAiB,CAAC;AAC1C,MAAM,iCAAiC,GAAG,IAAI,GAAG,CAAC,CAAC,qBAAqB,CAAC,CAAC,CAAC;AAC3E;;;;GAIG;AACH;;;;;;;GAOG;AACH,MAAM,eAAe,GAAG,uBAAuB,CAAC;AAEhD,SAAS,6BAA6B,CAAC,GAAW,EAAE,IAAY,EAAE,KAAc;IAC/E,MAAM,OAAO,GAAG,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC;IACjC,MAAM,eAAe,GAAG,gCAAgC,CAAC,OAAO,CAAC,CAAC;IAClE,IAAI,eAAe,EAAE,CAAC;QACrB,OAAO;YACN,IAAI,EAAE,gBAAgB;YACtB,OAAO,EAAE,eAAe,CAAC,IAAI;YAC7B,GAAG,EAAE,SAAS,eAAe,CAAC,IAAI,EAAE;SACpC,CAAC;IACH,CAAC;IACD,IAAI,CAAC,KAAK,IAAI,OAAO,CAAC,WAAW,EAAE,KAAK,eAAe,CAAC,WAAW,EAAE,EAAE,CAAC;QACvE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,SAAS,eAAe,EAAE,EAAE,CAAC;IAC5D,CAAC;IACD,IAAI,CAAC,KAAK,IAAI,iCAAiC,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,OAAO,EAAE,IAAI,EAAE,cAAc,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,OAAO,EAAE,EAAE,CAAC;IACnE,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,OAAO,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC;AAClF,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,QAAgB;IACpD,MAAM,KAAK,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;IAC9B,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC;IAC3E,IAAI,KAAK,CAAC,MAAM,GAAG,GAAG;QAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wCAAwC,EAAE,CAAC;IAEtG,yEAAyE;IACzE,MAAM,cAAc,GAAG,iCAAiC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACrE,IAAI,cAAc,EAAE,CAAC;QACpB,MAAM,QAAQ,GAAG,cAAc,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACjE,IAAI,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YACpC,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wDAAwD,EAAE,CAAC;QAC/F,CAAC;QACD,MAAM,KAAK,GAAG,oBAAoB,CAAC,QAAQ,CAAC,CAAC;QAC7C,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;YAAE,OAAO,KAAK,CAAC;QACzC,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,8DAA8D,EAAE,CAAC;IACrG,CAAC;IAED,IAAI,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QACrD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,wEAAwE,EAAE,CAAC;IAC/G,CAAC;IAED,+FAA+F;IAC/F,+FAA+F;IAC/F,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACjC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,6BAA6B,CAAC,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAE,CAAC,CAAC;IAC5D,CAAC;IAED,yBAAyB;IACzB,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;IACjC,IAAI,KAAK,EAAE,CAAC;QACX,OAAO,6BAA6B,CAAC,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAE,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IACtE,CAAC;IAED,IAAI,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;QAC3B,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,8DAA8D,EAAE,CAAC;IACrG,CAAC;IAED,sFAAsF;IACtF,IAAI,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;QACzB,MAAM,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QAC7C,MAAM,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC7B,IAAI,QAAQ,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YAC/C,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,QAAQ,IAAI,KAAK,EAAE,EAAE,CAAC;QACrD,CAAC;QACD,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,mDAAmD,EAAE,CAAC;IAC1F,CAAC;IAED,gEAAgE;IAChE,IAAI,UAAU,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5B,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;IAC1C,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,EAAE,kCAAkC,EAAE,CAAC;AACzE,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,0BAA0B,CAAC,GAAW,EAAE,aAAqB;IAC5E,IAAI,GAAG,KAAK,aAAa;QAAE,OAAO,IAAI,CAAC;IACvC,MAAM,MAAM,GAAG,CAAC,KAAa,EAAW,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;IAC/D,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,IAAI,aAAa,KAAK,GAAG,GAAG,SAAS;QAAE,OAAO,IAAI,CAAC;IACnE,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,IAAI,GAAG,KAAK,GAAG,aAAa,SAAS;QAAE,OAAO,IAAI,CAAC;IAC7E,OAAO,KAAK,CAAC;AACd,CAAC","sourcesContent":["/**\n * Model source normalizer (local-model-lifecycle-design.md, \"one panel, three types\"):\n * everything a user can paste — an ollama tag, an hf.co GGUF ref, a full HuggingFace URL, a\n * copied `ollama pull ...` install command, or an API `provider/model` name — normalizes to ONE\n * typed source. Pure string work: pasted install commands are PARSED for their reference and\n * NEVER executed as shell; unknown forms are rejected with the reason.\n */\n\nimport { resolvePrismLocalModelDescriptor } from \"./local-execution-planner.ts\";\n\nexport type ModelSource =\n\t| { type: \"api\"; ref: string }\n\t| { type: \"local\"; pullRef: string }\n\t| { type: \"transformers\"; modelId: string; ref: string }\n\t| { type: \"prism-llamacpp\"; modelId: string; ref: string }\n\t| { type: \"needle\"; ref: string }\n\t| { type: \"rejected\"; reason: string };\n\nconst OLLAMA_TAG = /^[a-z0-9][a-z0-9._-]*(?::[A-Za-z0-9._-]+)?$/;\nconst HF_REF = /^hf\\.co\\/([\\w.-]+)\\/([\\w.-]+)(?::([\\w.-]+))?$/i;\nconst HF_URL = /^https?:\\/\\/(?:www\\.)?huggingface\\.co\\/([\\w.-]+)\\/([\\w.-]+)(?:\\/.*)?$/i;\nconst SHELL_METACHARS = /[;&|`$<>(){}\\\\]/;\nconst PI_MANAGED_TRANSFORMERS_MODEL_IDS = new Set([\"openbmb/MiniCPM5-1B\"]);\n/**\n * Curated prism-ml models route through the managed Prism llama.cpp adapter instead of Ollama's\n * generic pull path. The execution planner owns the exact artifact and case-insensitive identity;\n * a user-provided quant suffix is accepted as input but cannot override that validated artifact.\n */\n/**\n * The single curated needle model id (see needle-runtime.ts) — a standalone function-call test\n * bench with no GGUF/quant concept at all (a pip-installed Python package + one pinned pickle\n * checkpoint), so unlike the prism route above, a quant suffix on the input is NOT accepted — it\n * falls through to the ordinary local/Ollama route instead, mirroring the transformers route's\n * `!quant` gate. Only this exact repo routes to \"needle\"; any other Cactus-Compute/* ref keeps the\n * ordinary local route.\n */\nconst NEEDLE_MODEL_ID = \"Cactus-Compute/needle\";\n\nfunction normalizeHuggingFaceReference(org: string, repo: string, quant?: string): ModelSource {\n\tconst modelId = `${org}/${repo}`;\n\tconst prismDescriptor = resolvePrismLocalModelDescriptor(modelId);\n\tif (prismDescriptor) {\n\t\treturn {\n\t\t\ttype: \"prism-llamacpp\",\n\t\t\tmodelId: prismDescriptor.repo,\n\t\t\tref: `hf.co/${prismDescriptor.repo}`,\n\t\t};\n\t}\n\tif (!quant && modelId.toLowerCase() === NEEDLE_MODEL_ID.toLowerCase()) {\n\t\treturn { type: \"needle\", ref: `hf.co/${NEEDLE_MODEL_ID}` };\n\t}\n\tif (!quant && PI_MANAGED_TRANSFORMERS_MODEL_IDS.has(modelId)) {\n\t\treturn { type: \"transformers\", modelId, ref: `hf.co/${modelId}` };\n\t}\n\treturn { type: \"local\", pullRef: `hf.co/${modelId}${quant ? `:${quant}` : \"\"}` };\n}\n\nexport function normalizeModelSource(rawInput: string): ModelSource {\n\tconst input = rawInput.trim();\n\tif (input.length === 0) return { type: \"rejected\", reason: \"empty input\" };\n\tif (input.length > 500) return { type: \"rejected\", reason: \"input too long to be a model reference\" };\n\n\t// Pasted install command: extract the reference, never execute anything.\n\tconst installCommand = /^ollama\\s+(?:pull|run)\\s+(.+)$/i.exec(input);\n\tif (installCommand) {\n\t\tconst argument = installCommand[1]!.trim().split(/\\s+/)[0] ?? \"\";\n\t\tif (SHELL_METACHARS.test(argument)) {\n\t\t\treturn { type: \"rejected\", reason: \"install command argument contains shell metacharacters\" };\n\t\t}\n\t\tconst inner = normalizeModelSource(argument);\n\t\tif (inner.type === \"local\") return inner;\n\t\treturn { type: \"rejected\", reason: `could not extract a model reference from the install command` };\n\t}\n\n\tif (SHELL_METACHARS.test(input) || /\\s/.test(input)) {\n\t\treturn { type: \"rejected\", reason: \"not a recognized model reference (contains spaces or shell characters)\" };\n\t}\n\n\t// Full HuggingFace URL -> either a pi-managed Transformers model (curated full-base ids) or an\n\t// Ollama hf.co pull ref (GGUF/quant refs still belong to Ollama and keep their :quant suffix).\n\tconst hfUrl = HF_URL.exec(input);\n\tif (hfUrl) {\n\t\treturn normalizeHuggingFaceReference(hfUrl[1]!, hfUrl[2]!);\n\t}\n\n\t// hf.co/org/repo[:quant]\n\tconst hfRef = HF_REF.exec(input);\n\tif (hfRef) {\n\t\treturn normalizeHuggingFaceReference(hfRef[1]!, hfRef[2]!, hfRef[3]);\n\t}\n\n\tif (input.includes(\"://\")) {\n\t\treturn { type: \"rejected\", reason: \"only huggingface.co URLs are recognized as local model links\" };\n\t}\n\n\t// provider/model -> API-registered model (nothing to install; auth + selection only).\n\tif (input.includes(\"/\")) {\n\t\tconst [provider, ...rest] = input.split(\"/\");\n\t\tconst model = rest.join(\"/\");\n\t\tif (provider && model && !model.includes(\"/\")) {\n\t\t\treturn { type: \"api\", ref: `${provider}/${model}` };\n\t\t}\n\t\treturn { type: \"rejected\", reason: \"expected provider/model or hf.co/org/repo[:quant]\" };\n\t}\n\n\t// Bare ollama tag (\"qwen3:1.7b\", \"pi-lifter:latest\", \"llama3\").\n\tif (OLLAMA_TAG.test(input)) {\n\t\treturn { type: \"local\", pullRef: input };\n\t}\n\n\treturn { type: \"rejected\", reason: \"not a recognized model reference\" };\n}\n\n/**\n * Whether an installed-model listing entry (`installedName`, as `OllamaRuntime.list()` reports it)\n * is the same model as `ref` (a user-typed reference, or a `models.json`-registered id). Ollama\n * always reports installed models with an explicit tag — a pull with no tag is normalized to\n * `:latest` server-side — but a `ref` is frequently bare (\"llama3\"), so an exact string compare\n * alone misses that pairing. Exact match always counts; otherwise treat a bare ref against its\n * `:latest`-tagged listing as the same model, and the reverse (a `ref` explicitly pinned to\n * `:latest` against a bare listing), since the local-runtime interface is kept runtime-agnostic\n * (see local-runtime.ts's module docstring) and a future/alternate runtime could list bare names.\n * Shared by every ref-vs-installed-list comparison — see `AgentSession._warnIfManualModelChoiceIsRisky`\n * and `removeLocalModel` in interactive-mode.ts, which had the identical bug independently.\n */\nexport function matchesInstalledLocalModel(ref: string, installedName: string): boolean {\n\tif (ref === installedName) return true;\n\tconst hasTag = (value: string): boolean => value.includes(\":\");\n\tif (!hasTag(ref) && installedName === `${ref}:latest`) return true;\n\tif (!hasTag(installedName) && ref === `${installedName}:latest`) return true;\n\treturn false;\n}\n"]}

package/dist/core/models/prism-llamacpp-lifecycle.d.ts

@@ -1,5 +1,5 @@
 /**
- * Core lifecycle logic for pi-managed prism llama.cpp models (Bonsai-27B first), shared between the
+ * Core lifecycle logic for Pi-managed Prism llama.cpp models, shared between the
  * `/models add` install pipeline (modes/interactive/local-model-commands.ts) and the router's
  * readiness gate (local-runtime-controller.ts). ONE place owns "ensure both GGUF files exist, then
  * serve" so a turn routed to a cold Bonsai-27B and a fresh `/models add` both go through the
@@ -9,18 +9,19 @@
  * sessions and isolated/background lanes alike — never depends on the interactive-only UI layer;
  * mirrors the existing local-runtime.ts (mechanics) / local-runtime-controller.ts (gate) split.
  */
-import { type PrismLlamaCppRuntime, type PrismModelDescriptor } from "./llamacpp-runtime.ts";
+import type { PrismLlamaCppRuntime, PrismModelDescriptor } from "./llamacpp-runtime.ts";
+export { PRISM_LLAMACPP_DESCRIPTORS } from "./local-execution-planner.ts";
+export { probePrismLlamaCppServer } from "./prism-llamacpp-server-probe.ts";
 /**
- * Curated prism llama.cpp descriptors, keyed by repo id — matches the modelId model-ref.ts's
- * curated `PRISM_LLAMACPP_MODEL_IDS` route returns, and the id `registerPrismLlamaCppModel` writes
- * as the model's id in models.json. Add future curated prism-ml models to both places together.
+ * Curated Prism llama.cpp descriptors, keyed by repo id — the same catalog owns model-ref routing
+ * and the id `registerPrismLlamaCppModel` writes to models.json. Future curated Prism models enter
+ * through that one catalog rather than parallel identifier lists.
  *
  * Also the pi-managed-vs-user-owned discriminator (see {@link isPiManagedPrismLlamaCppModel}): a
  * `llama-cpp` provider model is gated/self-healed by pi ONLY when its id is a key here — a user's
  * own hand-configured entry (e.g. the built-in `llama-cpp/local` catalog model on port 8080,
  * pointing at a server the user runs themselves) is never touched.
  */
-export declare const PRISM_LLAMACPP_DESCRIPTORS: Record<string, PrismModelDescriptor>;
 /**
  * True only for a `llama-cpp` provider model pi itself registered (id present in
  * {@link PRISM_LLAMACPP_DESCRIPTORS}) — never true for a user's own hand-configured llama-cpp
@@ -63,27 +64,16 @@ export type EnsurePrismModelServedResult = {
  * The SOLE path allowed to call `runtime.serve()` for a prism llama.cpp model — no caller may spawn
  * llama-server without first passing through here (both `addPrismLlamaCppModel`'s install pipeline
  * and the readiness gate's self-heal path call this, never `runtime.serve()` directly).
- * Unconditionally re-verifies BOTH GGUF files via `runtime.downloadModel()` immediately before every
+ * Unconditionally re-verifies every required GGUF file via `runtime.downloadModel()` immediately before every
  * serve — that call is idempotent (skips a re-download when the local file already matches the
  * remote size, so the happy path costs one stat) and self-heals a file that went missing from disk
  * (deleted, a partial download from a prior crash, host cleanup, etc.) by re-fetching it instead of
  * ever starting llama-server without it. The vision projector is mandatory whenever the descriptor
- * declares one (Bonsai-27B always does — see BONSAI_27B.mmprojFile): a download failure on EITHER
- * file means llama-server is never spawned. There is no text-only fallback — Bonsai-27B never
- * silently degrades to a vision-less serve because the projector went missing.
+ * declares one (the 27B descriptors do): a download failure on any required file means
+ * llama-server is never spawned. There is no text-only fallback for a vision descriptor.
  */
 export declare function ensurePrismModelFilesThenServe(runtime: PrismLlamaCppRuntime, descriptor: PrismModelDescriptor, args: {
     port: number;
     numCtx: number;
 }, onProgress: (message: string) => void): Promise<EnsurePrismModelServedResult>;
-/**
- * Direct HTTP reachability probe against a prism llama.cpp server root (NOT the `/v1` OpenAI-compat
- * suffix models.json registers — same root/health split as llama-server's own health endpoint used
- * internally by PrismLlamaCppRuntime#serve). PrismLlamaCppRuntime has no public method for this (its
- * own equivalent, `_healthUp`, is private and scoped to a `serve()` call in progress), so the
- * readiness gate needs its own — this is deliberately independent of any single runtime instance:
- * it answers "is a server already alive at this URL", true even for a server this session never
- * spawned (started earlier, or by a different process).
- */
-export declare function isPrismLlamaCppServerHealthy(serverUrl: string, fetchFn?: typeof fetch, timeoutMs?: number): Promise<boolean>;
 //# sourceMappingURL=prism-llamacpp-lifecycle.d.ts.map

package/dist/core/models/prism-llamacpp-lifecycle.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"prism-llamacpp-lifecycle.d.ts","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-lifecycle.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAc,KAAK,oBAAoB,EAAE,KAAK,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAGzG;;;;;;;;;GASG;AACH,eAAO,MAAM,0BAA0B,EAAE,MAAM,CAAC,MAAM,EAAE,oBAAoB,CAE3E,CAAC;AAEF;;;;;GAKG;AACH,wBAAgB,6BAA6B,CAAC,KAAK,EAAE;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAE9F;AAED;;;;;GAKG;AACH,eAAO,MAAM,yBAAyB,OAAO,CAAC;AAE9C;;;;;;;;;;;GAWG;AACH,wBAAgB,yBAAyB,CAAC,aAAa,EAAE,MAAM,GAAG,MAAM,CAMvE;AAED,MAAM,MAAM,4BAA4B,GACrC;IAAE,EAAE,EAAE,IAAI,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC7B;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,gBAAgB,GAAG,iBAAiB,GAAG,OAAO,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvF;;;;;;;;;;;;GAYG;AACH,wBAAsB,8BAA8B,CACnD,OAAO,EAAE,oBAAoB,EAC7B,UAAU,EAAE,oBAAoB,EAChC,IAAI,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EACtC,UAAU,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,GACnC,OAAO,CAAC,4BAA4B,CAAC,CA6BvC;AAID;;;;;;;;GAQG;AACH,wBAAsB,4BAA4B,CACjD,SAAS,EAAE,MAAM,EACjB,OAAO,GAAE,OAAO,KAAa,EAC7B,SAAS,GAAE,MAAwC,GACjD,OAAO,CAAC,OAAO,CAAC,CAOlB"}
+{"version":3,"file":"prism-llamacpp-lifecycle.d.ts","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-lifecycle.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,KAAK,EAAE,oBAAoB,EAAE,oBAAoB,EAAE,MAAM,uBAAuB,CAAC;AAIxF,OAAO,EAAE,0BAA0B,EAAE,MAAM,8BAA8B,CAAC;AAC1E,OAAO,EAAE,wBAAwB,EAAE,MAAM,kCAAkC,CAAC;AAE5E;;;;;;;;;GASG;AACH;;;;;GAKG;AACH,wBAAgB,6BAA6B,CAAC,KAAK,EAAE;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAE9F;AAED;;;;;GAKG;AACH,eAAO,MAAM,yBAAyB,OAAO,CAAC;AAE9C;;;;;;;;;;;GAWG;AACH,wBAAgB,yBAAyB,CAAC,aAAa,EAAE,MAAM,GAAG,MAAM,CAMvE;AAED,MAAM,MAAM,4BAA4B,GACrC;IAAE,EAAE,EAAE,IAAI,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,GAC7B;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,gBAAgB,GAAG,iBAAiB,GAAG,OAAO,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvF;;;;;;;;;;;GAWG;AACH,wBAAsB,8BAA8B,CACnD,OAAO,EAAE,oBAAoB,EAC7B,UAAU,EAAE,oBAAoB,EAChC,IAAI,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,EACtC,UAAU,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,GACnC,OAAO,CAAC,4BAA4B,CAAC,CA8BvC"}

package/dist/core/models/prism-llamacpp-lifecycle.js

@@ -1,5 +1,5 @@
 /**
- * Core lifecycle logic for pi-managed prism llama.cpp models (Bonsai-27B first), shared between the
+ * Core lifecycle logic for Pi-managed Prism llama.cpp models, shared between the
  * `/models add` install pipeline (modes/interactive/local-model-commands.ts) and the router's
  * readiness gate (local-runtime-controller.ts). ONE place owns "ensure both GGUF files exist, then
  * serve" so a turn routed to a cold Bonsai-27B and a fresh `/models add` both go through the
@@ -9,21 +9,20 @@
  * sessions and isolated/background lanes alike — never depends on the interactive-only UI layer;
  * mirrors the existing local-runtime.ts (mechanics) / local-runtime-controller.ts (gate) split.
  */
-import { BONSAI_27B } from "./llamacpp-runtime.js";
+import { PRISM_LLAMACPP_DESCRIPTORS } from "./local-execution-planner.js";
 import { PRISM_LLAMACPP_PROVIDER } from "./local-registration.js";
+export { PRISM_LLAMACPP_DESCRIPTORS } from "./local-execution-planner.js";
+export { probePrismLlamaCppServer } from "./prism-llamacpp-server-probe.js";
 /**
- * Curated prism llama.cpp descriptors, keyed by repo id — matches the modelId model-ref.ts's
- * curated `PRISM_LLAMACPP_MODEL_IDS` route returns, and the id `registerPrismLlamaCppModel` writes
- * as the model's id in models.json. Add future curated prism-ml models to both places together.
+ * Curated Prism llama.cpp descriptors, keyed by repo id — the same catalog owns model-ref routing
+ * and the id `registerPrismLlamaCppModel` writes to models.json. Future curated Prism models enter
+ * through that one catalog rather than parallel identifier lists.
  *
  * Also the pi-managed-vs-user-owned discriminator (see {@link isPiManagedPrismLlamaCppModel}): a
  * `llama-cpp` provider model is gated/self-healed by pi ONLY when its id is a key here — a user's
  * own hand-configured entry (e.g. the built-in `llama-cpp/local` catalog model on port 8080,
  * pointing at a server the user runs themselves) is never touched.
  */
-export const PRISM_LLAMACPP_DESCRIPTORS = {
-    [BONSAI_27B.repo]: BONSAI_27B,
-};
 /**
  * True only for a `llama-cpp` provider model pi itself registered (id present in
  * {@link PRISM_LLAMACPP_DESCRIPTORS}) — never true for a user's own hand-configured llama-cpp
@@ -66,14 +65,13 @@ export function derivePrismLlamaCppNumCtx(totalMemBytes) {
  * The SOLE path allowed to call `runtime.serve()` for a prism llama.cpp model — no caller may spawn
  * llama-server without first passing through here (both `addPrismLlamaCppModel`'s install pipeline
  * and the readiness gate's self-heal path call this, never `runtime.serve()` directly).
- * Unconditionally re-verifies BOTH GGUF files via `runtime.downloadModel()` immediately before every
+ * Unconditionally re-verifies every required GGUF file via `runtime.downloadModel()` immediately before every
  * serve — that call is idempotent (skips a re-download when the local file already matches the
  * remote size, so the happy path costs one stat) and self-heals a file that went missing from disk
  * (deleted, a partial download from a prior crash, host cleanup, etc.) by re-fetching it instead of
  * ever starting llama-server without it. The vision projector is mandatory whenever the descriptor
- * declares one (Bonsai-27B always does — see BONSAI_27B.mmprojFile): a download failure on EITHER
- * file means llama-server is never spawned. There is no text-only fallback — Bonsai-27B never
- * silently degrades to a vision-less serve because the projector went missing.
+ * declares one (the 27B descriptors do): a download failure on any required file means
+ * llama-server is never spawned. There is no text-only fallback for a vision descriptor.
  */
 export async function ensurePrismModelFilesThenServe(runtime, descriptor, args, onProgress) {
     const downloadedModel = await runtime.downloadModel({ repo: descriptor.repo, file: descriptor.file }, (progress) => onProgress(`  ${descriptor.file}: ${progress}`));
@@ -91,6 +89,7 @@ export async function ensurePrismModelFilesThenServe(runtime, descriptor, args,
     onProgress(`Starting ${descriptor.displayName} on 127.0.0.1:${args.port} (context ${args.numCtx})…`);
     const served = await runtime.serve({
         modelPath: downloadedModel.path,
+        modelAlias: descriptor.repo,
         mmprojPath,
         port: args.port,
         numCtx: args.numCtx,
@@ -99,23 +98,4 @@ export async function ensurePrismModelFilesThenServe(runtime, descriptor, args,
         return { ok: false, stage: "serve", error: served.error };
     return { ok: true, baseUrl: served.baseUrl };
 }
-const DEFAULT_HEALTH_CHECK_TIMEOUT_MS = 2_000;
-/**
- * Direct HTTP reachability probe against a prism llama.cpp server root (NOT the `/v1` OpenAI-compat
- * suffix models.json registers — same root/health split as llama-server's own health endpoint used
- * internally by PrismLlamaCppRuntime#serve). PrismLlamaCppRuntime has no public method for this (its
- * own equivalent, `_healthUp`, is private and scoped to a `serve()` call in progress), so the
- * readiness gate needs its own — this is deliberately independent of any single runtime instance:
- * it answers "is a server already alive at this URL", true even for a server this session never
- * spawned (started earlier, or by a different process).
- */
-export async function isPrismLlamaCppServerHealthy(serverUrl, fetchFn = fetch, timeoutMs = DEFAULT_HEALTH_CHECK_TIMEOUT_MS) {
-    try {
-        const response = await fetchFn(`${serverUrl}/health`, { signal: AbortSignal.timeout(timeoutMs) });
-        return response.ok;
-    }
-    catch {
-        return false;
-    }
-}
 //# sourceMappingURL=prism-llamacpp-lifecycle.js.map

package/dist/core/models/prism-llamacpp-lifecycle.js.map

@@ -1 +1 @@
-{"version":3,"file":"prism-llamacpp-lifecycle.js","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-lifecycle.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,UAAU,EAAwD,MAAM,uBAAuB,CAAC;AACzG,OAAO,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AAElE;;;;;;;;;GASG;AACH,MAAM,CAAC,MAAM,0BAA0B,GAAyC;IAC/E,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,UAAU;CAC7B,CAAC;AAEF;;;;;GAKG;AACH,MAAM,UAAU,6BAA6B,CAAC,KAAuC;IACpF,OAAO,KAAK,CAAC,QAAQ,KAAK,uBAAuB,IAAI,MAAM,CAAC,MAAM,CAAC,0BAA0B,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;AAC1G,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC;AAE9C;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,yBAAyB,CAAC,aAAqB;IAC9D,MAAM,OAAO,GAAG,aAAa,GAAG,GAAG,CAAC;IACpC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC;IACjC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC;IACjC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,KAAK,CAAC;IAChC,OAAO,KAAK,CAAC;AACd,CAAC;AAMD;;;;;;;;;;;;GAYG;AACH,MAAM,CAAC,KAAK,UAAU,8BAA8B,CACnD,OAA6B,EAC7B,UAAgC,EAChC,IAAsC,EACtC,UAAqC;IAErC,MAAM,eAAe,GAAG,MAAM,OAAO,CAAC,aAAa,CAAC,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,EAAE,EAAE,CAClH,UAAU,CAAC,KAAK,UAAU,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC,CAC/C,CAAC;IACF,IAAI,CAAC,eAAe,CAAC,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QAClD,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,EAAE,KAAK,EAAE,eAAe,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC;IAChG,CAAC;IAED,IAAI,UAA8B,CAAC;IACnC,IAAI,UAAU,CAAC,UAAU,EAAE,CAAC;QAC3B,MAAM,gBAAgB,GAAG,MAAM,OAAO,CAAC,aAAa,CACnD,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,IAAI,EAAE,UAAU,CAAC,UAAU,EAAE,EACtD,CAAC,QAAQ,EAAE,EAAE,CAAC,UAAU,CAAC,KAAK,UAAU,CAAC,UAAU,KAAK,QAAQ,EAAE,CAAC,CACnE,CAAC;QACF,IAAI,CAAC,gBAAgB,CAAC,EAAE,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,CAAC;YACpD,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,iBAAiB,EAAE,KAAK,EAAE,gBAAgB,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC;QAClG,CAAC;QACD,UAAU,GAAG,gBAAgB,CAAC,IAAI,CAAC;IACpC,CAAC;IAED,UAAU,CAAC,YAAY,UAAU,CAAC,WAAW,iBAAiB,IAAI,CAAC,IAAI,aAAa,IAAI,CAAC,MAAM,IAAI,CAAC,CAAC;IACrG,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,KAAK,CAAC;QAClC,SAAS,EAAE,eAAe,CAAC,IAAI;QAC/B,UAAU;QACV,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,MAAM,EAAE,IAAI,CAAC,MAAM;KACnB,CAAC,CAAC;IACH,IAAI,CAAC,MAAM,CAAC,EAAE;QAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC;IAC1E,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;AAC9C,CAAC;AAED,MAAM,+BAA+B,GAAG,KAAK,CAAC;AAE9C;;;;;;;;GAQG;AACH,MAAM,CAAC,KAAK,UAAU,4BAA4B,CACjD,SAAiB,EACjB,OAAO,GAAiB,KAAK,EAC7B,SAAS,GAAW,+BAA+B;IAEnD,IAAI,CAAC;QACJ,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,SAAS,SAAS,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;QAClG,OAAO,QAAQ,CAAC,EAAE,CAAC;IACpB,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AACF,CAAC","sourcesContent":["/**\n * Core lifecycle logic for pi-managed prism llama.cpp models (Bonsai-27B first), shared between the\n * `/models add` install pipeline (modes/interactive/local-model-commands.ts) and the router's\n * readiness gate (local-runtime-controller.ts). ONE place owns \"ensure both GGUF files exist, then\n * serve\" so a turn routed to a cold Bonsai-27B and a fresh `/models add` both go through the\n * identical self-healing path instead of two parallel, potentially-diverging implementations.\n *\n * Lives in core/models/ (not modes/) so the headless readiness gate — used by print/RPC/interactive\n * sessions and isolated/background lanes alike — never depends on the interactive-only UI layer;\n * mirrors the existing local-runtime.ts (mechanics) / local-runtime-controller.ts (gate) split.\n */\n\nimport { BONSAI_27B, type PrismLlamaCppRuntime, type PrismModelDescriptor } from \"./llamacpp-runtime.ts\";\nimport { PRISM_LLAMACPP_PROVIDER } from \"./local-registration.ts\";\n\n/**\n * Curated prism llama.cpp descriptors, keyed by repo id — matches the modelId model-ref.ts's\n * curated `PRISM_LLAMACPP_MODEL_IDS` route returns, and the id `registerPrismLlamaCppModel` writes\n * as the model's id in models.json. Add future curated prism-ml models to both places together.\n *\n * Also the pi-managed-vs-user-owned discriminator (see {@link isPiManagedPrismLlamaCppModel}): a\n * `llama-cpp` provider model is gated/self-healed by pi ONLY when its id is a key here — a user's\n * own hand-configured entry (e.g. the built-in `llama-cpp/local` catalog model on port 8080,\n * pointing at a server the user runs themselves) is never touched.\n */\nexport const PRISM_LLAMACPP_DESCRIPTORS: Record<string, PrismModelDescriptor> = {\n\t[BONSAI_27B.repo]: BONSAI_27B,\n};\n\n/**\n * True only for a `llama-cpp` provider model pi itself registered (id present in\n * {@link PRISM_LLAMACPP_DESCRIPTORS}) — never true for a user's own hand-configured llama-cpp\n * model, including the built-in `llama-cpp/local` catalog entry. This is the gate that must run\n * before ANY self-heal/serve action so pi never touches a server it doesn't own.\n */\nexport function isPiManagedPrismLlamaCppModel(model: { provider: string; id: string }): boolean {\n\treturn model.provider === PRISM_LLAMACPP_PROVIDER && Object.hasOwn(PRISM_LLAMACPP_DESCRIPTORS, model.id);\n}\n\n/**\n * Fixed local port for pi's own managed prism llama.cpp server. Deliberately NOT the generic\n * built-in `llama-cpp/local` catalog model's conventional port 8080 (models.generated.ts) — that\n * port is the convention for a user's OWN manually started llama-server pi merely points at; pi's\n * managed instance must never collide with it.\n */\nexport const PRISM_LLAMACPP_SERVE_PORT = 8090;\n\n/**\n * Conservative RAM-only context cap for the pi-managed prism llama.cpp server, used at FIRST\n * install time only (`/models add`) — a later self-heal re-serve reuses the model's already\n * registered `contextWindow` instead (see the readiness gate in local-runtime-controller.ts), never\n * re-deriving from current RAM, so a served context size can't drift from what pi already told the\n * rest of the session (compaction, etc.) to expect. PrismLlamaCppRuntime has no `/api/show`-\n * equivalent GGUF metadata endpoint (unlike Ollama, which context-sizing.ts's\n * deriveLocalContextSizing depends on), so this is a coarse rung table mirroring\n * context-sizing.ts's CONTEXT_RUNGS, hard-capped at 32768 regardless of headroom: Bonsai-27B\n * advertises a 262K-class context, but that is not a realistic KV budget for a CPU-served model on\n * consumer hardware, and there is no larger validated rung to grow into yet.\n */\nexport function derivePrismLlamaCppNumCtx(totalMemBytes: number): number {\n\tconst totalGb = totalMemBytes / 1e9;\n\tif (totalGb >= 64) return 32_768;\n\tif (totalGb >= 32) return 16_384;\n\tif (totalGb >= 16) return 8_192;\n\treturn 4_096;\n}\n\nexport type EnsurePrismModelServedResult =\n\t| { ok: true; baseUrl: string }\n\t| { ok: false; stage: \"model-download\" | \"mmproj-download\" | \"serve\"; error: string };\n\n/**\n * The SOLE path allowed to call `runtime.serve()` for a prism llama.cpp model — no caller may spawn\n * llama-server without first passing through here (both `addPrismLlamaCppModel`'s install pipeline\n * and the readiness gate's self-heal path call this, never `runtime.serve()` directly).\n * Unconditionally re-verifies BOTH GGUF files via `runtime.downloadModel()` immediately before every\n * serve — that call is idempotent (skips a re-download when the local file already matches the\n * remote size, so the happy path costs one stat) and self-heals a file that went missing from disk\n * (deleted, a partial download from a prior crash, host cleanup, etc.) by re-fetching it instead of\n * ever starting llama-server without it. The vision projector is mandatory whenever the descriptor\n * declares one (Bonsai-27B always does — see BONSAI_27B.mmprojFile): a download failure on EITHER\n * file means llama-server is never spawned. There is no text-only fallback — Bonsai-27B never\n * silently degrades to a vision-less serve because the projector went missing.\n */\nexport async function ensurePrismModelFilesThenServe(\n\truntime: PrismLlamaCppRuntime,\n\tdescriptor: PrismModelDescriptor,\n\targs: { port: number; numCtx: number },\n\tonProgress: (message: string) => void,\n): Promise<EnsurePrismModelServedResult> {\n\tconst downloadedModel = await runtime.downloadModel({ repo: descriptor.repo, file: descriptor.file }, (progress) =>\n\t\tonProgress(`  ${descriptor.file}: ${progress}`),\n\t);\n\tif (!downloadedModel.ok || !downloadedModel.path) {\n\t\treturn { ok: false, stage: \"model-download\", error: downloadedModel.error ?? \"unknown error\" };\n\t}\n\n\tlet mmprojPath: string | undefined;\n\tif (descriptor.mmprojFile) {\n\t\tconst downloadedMmproj = await runtime.downloadModel(\n\t\t\t{ repo: descriptor.repo, file: descriptor.mmprojFile },\n\t\t\t(progress) => onProgress(`  ${descriptor.mmprojFile}: ${progress}`),\n\t\t);\n\t\tif (!downloadedMmproj.ok || !downloadedMmproj.path) {\n\t\t\treturn { ok: false, stage: \"mmproj-download\", error: downloadedMmproj.error ?? \"unknown error\" };\n\t\t}\n\t\tmmprojPath = downloadedMmproj.path;\n\t}\n\n\tonProgress(`Starting ${descriptor.displayName} on 127.0.0.1:${args.port} (context ${args.numCtx})…`);\n\tconst served = await runtime.serve({\n\t\tmodelPath: downloadedModel.path,\n\t\tmmprojPath,\n\t\tport: args.port,\n\t\tnumCtx: args.numCtx,\n\t});\n\tif (!served.ok) return { ok: false, stage: \"serve\", error: served.error };\n\treturn { ok: true, baseUrl: served.baseUrl };\n}\n\nconst DEFAULT_HEALTH_CHECK_TIMEOUT_MS = 2_000;\n\n/**\n * Direct HTTP reachability probe against a prism llama.cpp server root (NOT the `/v1` OpenAI-compat\n * suffix models.json registers — same root/health split as llama-server's own health endpoint used\n * internally by PrismLlamaCppRuntime#serve). PrismLlamaCppRuntime has no public method for this (its\n * own equivalent, `_healthUp`, is private and scoped to a `serve()` call in progress), so the\n * readiness gate needs its own — this is deliberately independent of any single runtime instance:\n * it answers \"is a server already alive at this URL\", true even for a server this session never\n * spawned (started earlier, or by a different process).\n */\nexport async function isPrismLlamaCppServerHealthy(\n\tserverUrl: string,\n\tfetchFn: typeof fetch = fetch,\n\ttimeoutMs: number = DEFAULT_HEALTH_CHECK_TIMEOUT_MS,\n): Promise<boolean> {\n\ttry {\n\t\tconst response = await fetchFn(`${serverUrl}/health`, { signal: AbortSignal.timeout(timeoutMs) });\n\t\treturn response.ok;\n\t} catch {\n\t\treturn false;\n\t}\n}\n"]}
+{"version":3,"file":"prism-llamacpp-lifecycle.js","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-lifecycle.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,EAAE,0BAA0B,EAAE,MAAM,8BAA8B,CAAC;AAC1E,OAAO,EAAE,uBAAuB,EAAE,MAAM,yBAAyB,CAAC;AAElE,OAAO,EAAE,0BAA0B,EAAE,MAAM,8BAA8B,CAAC;AAC1E,OAAO,EAAE,wBAAwB,EAAE,MAAM,kCAAkC,CAAC;AAE5E;;;;;;;;;GASG;AACH;;;;;GAKG;AACH,MAAM,UAAU,6BAA6B,CAAC,KAAuC;IACpF,OAAO,KAAK,CAAC,QAAQ,KAAK,uBAAuB,IAAI,MAAM,CAAC,MAAM,CAAC,0BAA0B,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;AAC1G,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,MAAM,yBAAyB,GAAG,IAAI,CAAC;AAE9C;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,yBAAyB,CAAC,aAAqB;IAC9D,MAAM,OAAO,GAAG,aAAa,GAAG,GAAG,CAAC;IACpC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC;IACjC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC;IACjC,IAAI,OAAO,IAAI,EAAE;QAAE,OAAO,KAAK,CAAC;IAChC,OAAO,KAAK,CAAC;AACd,CAAC;AAMD;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,8BAA8B,CACnD,OAA6B,EAC7B,UAAgC,EAChC,IAAsC,EACtC,UAAqC;IAErC,MAAM,eAAe,GAAG,MAAM,OAAO,CAAC,aAAa,CAAC,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,QAAQ,EAAE,EAAE,CAClH,UAAU,CAAC,KAAK,UAAU,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC,CAC/C,CAAC;IACF,IAAI,CAAC,eAAe,CAAC,EAAE,IAAI,CAAC,eAAe,CAAC,IAAI,EAAE,CAAC;QAClD,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,gBAAgB,EAAE,KAAK,EAAE,eAAe,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC;IAChG,CAAC;IAED,IAAI,UAA8B,CAAC;IACnC,IAAI,UAAU,CAAC,UAAU,EAAE,CAAC;QAC3B,MAAM,gBAAgB,GAAG,MAAM,OAAO,CAAC,aAAa,CACnD,EAAE,IAAI,EAAE,UAAU,CAAC,IAAI,EAAE,IAAI,EAAE,UAAU,CAAC,UAAU,EAAE,EACtD,CAAC,QAAQ,EAAE,EAAE,CAAC,UAAU,CAAC,KAAK,UAAU,CAAC,UAAU,KAAK,QAAQ,EAAE,CAAC,CACnE,CAAC;QACF,IAAI,CAAC,gBAAgB,CAAC,EAAE,IAAI,CAAC,gBAAgB,CAAC,IAAI,EAAE,CAAC;YACpD,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,iBAAiB,EAAE,KAAK,EAAE,gBAAgB,CAAC,KAAK,IAAI,eAAe,EAAE,CAAC;QAClG,CAAC;QACD,UAAU,GAAG,gBAAgB,CAAC,IAAI,CAAC;IACpC,CAAC;IAED,UAAU,CAAC,YAAY,UAAU,CAAC,WAAW,iBAAiB,IAAI,CAAC,IAAI,aAAa,IAAI,CAAC,MAAM,IAAI,CAAC,CAAC;IACrG,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,KAAK,CAAC;QAClC,SAAS,EAAE,eAAe,CAAC,IAAI;QAC/B,UAAU,EAAE,UAAU,CAAC,IAAI;QAC3B,UAAU;QACV,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,MAAM,EAAE,IAAI,CAAC,MAAM;KACnB,CAAC,CAAC;IACH,IAAI,CAAC,MAAM,CAAC,EAAE;QAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,CAAC,KAAK,EAAE,CAAC;IAC1E,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC;AAC9C,CAAC","sourcesContent":["/**\n * Core lifecycle logic for Pi-managed Prism llama.cpp models, shared between the\n * `/models add` install pipeline (modes/interactive/local-model-commands.ts) and the router's\n * readiness gate (local-runtime-controller.ts). ONE place owns \"ensure both GGUF files exist, then\n * serve\" so a turn routed to a cold Bonsai-27B and a fresh `/models add` both go through the\n * identical self-healing path instead of two parallel, potentially-diverging implementations.\n *\n * Lives in core/models/ (not modes/) so the headless readiness gate — used by print/RPC/interactive\n * sessions and isolated/background lanes alike — never depends on the interactive-only UI layer;\n * mirrors the existing local-runtime.ts (mechanics) / local-runtime-controller.ts (gate) split.\n */\n\nimport type { PrismLlamaCppRuntime, PrismModelDescriptor } from \"./llamacpp-runtime.ts\";\nimport { PRISM_LLAMACPP_DESCRIPTORS } from \"./local-execution-planner.ts\";\nimport { PRISM_LLAMACPP_PROVIDER } from \"./local-registration.ts\";\n\nexport { PRISM_LLAMACPP_DESCRIPTORS } from \"./local-execution-planner.ts\";\nexport { probePrismLlamaCppServer } from \"./prism-llamacpp-server-probe.ts\";\n\n/**\n * Curated Prism llama.cpp descriptors, keyed by repo id — the same catalog owns model-ref routing\n * and the id `registerPrismLlamaCppModel` writes to models.json. Future curated Prism models enter\n * through that one catalog rather than parallel identifier lists.\n *\n * Also the pi-managed-vs-user-owned discriminator (see {@link isPiManagedPrismLlamaCppModel}): a\n * `llama-cpp` provider model is gated/self-healed by pi ONLY when its id is a key here — a user's\n * own hand-configured entry (e.g. the built-in `llama-cpp/local` catalog model on port 8080,\n * pointing at a server the user runs themselves) is never touched.\n */\n/**\n * True only for a `llama-cpp` provider model pi itself registered (id present in\n * {@link PRISM_LLAMACPP_DESCRIPTORS}) — never true for a user's own hand-configured llama-cpp\n * model, including the built-in `llama-cpp/local` catalog entry. This is the gate that must run\n * before ANY self-heal/serve action so pi never touches a server it doesn't own.\n */\nexport function isPiManagedPrismLlamaCppModel(model: { provider: string; id: string }): boolean {\n\treturn model.provider === PRISM_LLAMACPP_PROVIDER && Object.hasOwn(PRISM_LLAMACPP_DESCRIPTORS, model.id);\n}\n\n/**\n * Fixed local port for pi's own managed prism llama.cpp server. Deliberately NOT the generic\n * built-in `llama-cpp/local` catalog model's conventional port 8080 (models.generated.ts) — that\n * port is the convention for a user's OWN manually started llama-server pi merely points at; pi's\n * managed instance must never collide with it.\n */\nexport const PRISM_LLAMACPP_SERVE_PORT = 8090;\n\n/**\n * Conservative RAM-only context cap for the pi-managed prism llama.cpp server, used at FIRST\n * install time only (`/models add`) — a later self-heal re-serve reuses the model's already\n * registered `contextWindow` instead (see the readiness gate in local-runtime-controller.ts), never\n * re-deriving from current RAM, so a served context size can't drift from what pi already told the\n * rest of the session (compaction, etc.) to expect. PrismLlamaCppRuntime has no `/api/show`-\n * equivalent GGUF metadata endpoint (unlike Ollama, which context-sizing.ts's\n * deriveLocalContextSizing depends on), so this is a coarse rung table mirroring\n * context-sizing.ts's CONTEXT_RUNGS, hard-capped at 32768 regardless of headroom: Bonsai-27B\n * advertises a 262K-class context, but that is not a realistic KV budget for a CPU-served model on\n * consumer hardware, and there is no larger validated rung to grow into yet.\n */\nexport function derivePrismLlamaCppNumCtx(totalMemBytes: number): number {\n\tconst totalGb = totalMemBytes / 1e9;\n\tif (totalGb >= 64) return 32_768;\n\tif (totalGb >= 32) return 16_384;\n\tif (totalGb >= 16) return 8_192;\n\treturn 4_096;\n}\n\nexport type EnsurePrismModelServedResult =\n\t| { ok: true; baseUrl: string }\n\t| { ok: false; stage: \"model-download\" | \"mmproj-download\" | \"serve\"; error: string };\n\n/**\n * The SOLE path allowed to call `runtime.serve()` for a prism llama.cpp model — no caller may spawn\n * llama-server without first passing through here (both `addPrismLlamaCppModel`'s install pipeline\n * and the readiness gate's self-heal path call this, never `runtime.serve()` directly).\n * Unconditionally re-verifies every required GGUF file via `runtime.downloadModel()` immediately before every\n * serve — that call is idempotent (skips a re-download when the local file already matches the\n * remote size, so the happy path costs one stat) and self-heals a file that went missing from disk\n * (deleted, a partial download from a prior crash, host cleanup, etc.) by re-fetching it instead of\n * ever starting llama-server without it. The vision projector is mandatory whenever the descriptor\n * declares one (the 27B descriptors do): a download failure on any required file means\n * llama-server is never spawned. There is no text-only fallback for a vision descriptor.\n */\nexport async function ensurePrismModelFilesThenServe(\n\truntime: PrismLlamaCppRuntime,\n\tdescriptor: PrismModelDescriptor,\n\targs: { port: number; numCtx: number },\n\tonProgress: (message: string) => void,\n): Promise<EnsurePrismModelServedResult> {\n\tconst downloadedModel = await runtime.downloadModel({ repo: descriptor.repo, file: descriptor.file }, (progress) =>\n\t\tonProgress(`  ${descriptor.file}: ${progress}`),\n\t);\n\tif (!downloadedModel.ok || !downloadedModel.path) {\n\t\treturn { ok: false, stage: \"model-download\", error: downloadedModel.error ?? \"unknown error\" };\n\t}\n\n\tlet mmprojPath: string | undefined;\n\tif (descriptor.mmprojFile) {\n\t\tconst downloadedMmproj = await runtime.downloadModel(\n\t\t\t{ repo: descriptor.repo, file: descriptor.mmprojFile },\n\t\t\t(progress) => onProgress(`  ${descriptor.mmprojFile}: ${progress}`),\n\t\t);\n\t\tif (!downloadedMmproj.ok || !downloadedMmproj.path) {\n\t\t\treturn { ok: false, stage: \"mmproj-download\", error: downloadedMmproj.error ?? \"unknown error\" };\n\t\t}\n\t\tmmprojPath = downloadedMmproj.path;\n\t}\n\n\tonProgress(`Starting ${descriptor.displayName} on 127.0.0.1:${args.port} (context ${args.numCtx})…`);\n\tconst served = await runtime.serve({\n\t\tmodelPath: downloadedModel.path,\n\t\tmodelAlias: descriptor.repo,\n\t\tmmprojPath,\n\t\tport: args.port,\n\t\tnumCtx: args.numCtx,\n\t});\n\tif (!served.ok) return { ok: false, stage: \"serve\", error: served.error };\n\treturn { ok: true, baseUrl: served.baseUrl };\n}\n"]}

package/dist/core/models/prism-llamacpp-server-probe.d.ts

@@ -0,0 +1,10 @@
+export type PrismLlamaCppServerProbe = {
+    status: "down";
+    servedModelIds: [];
+} | {
+    status: "matching" | "conflict";
+    servedModelIds: string[];
+};
+/** Health alone is insufficient on a shared port: readiness requires the exact requested alias. */
+export declare function probePrismLlamaCppServer(serverUrl: string, expectedModelId: string, fetchFn?: typeof fetch, timeoutMs?: number): Promise<PrismLlamaCppServerProbe>;
+//# sourceMappingURL=prism-llamacpp-server-probe.d.ts.map

package/dist/core/models/prism-llamacpp-server-probe.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"prism-llamacpp-server-probe.d.ts","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-server-probe.ts"],"names":[],"mappings":"AAAA,MAAM,MAAM,wBAAwB,GACjC;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,cAAc,EAAE,EAAE,CAAA;CAAE,GACtC;IAAE,MAAM,EAAE,UAAU,GAAG,UAAU,CAAC;IAAC,cAAc,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC;AAMjE,mGAAmG;AACnG,wBAAsB,wBAAwB,CAC7C,SAAS,EAAE,MAAM,EACjB,eAAe,EAAE,MAAM,EACvB,OAAO,GAAE,OAAO,KAAa,EAC7B,SAAS,GAAE,MAAiC,GAC1C,OAAO,CAAC,wBAAwB,CAAC,CA6BnC"}

package/dist/core/models/prism-llamacpp-server-probe.js

@@ -0,0 +1,36 @@
+const DEFAULT_PROBE_TIMEOUT_MS = 2_000;
+const MAX_REPORTED_MODEL_IDS = 8;
+const MAX_MODEL_ID_CHARS = 256;
+/** Health alone is insufficient on a shared port: readiness requires the exact requested alias. */
+export async function probePrismLlamaCppServer(serverUrl, expectedModelId, fetchFn = fetch, timeoutMs = DEFAULT_PROBE_TIMEOUT_MS) {
+    try {
+        const health = await fetchFn(`${serverUrl}/health`, { signal: AbortSignal.timeout(timeoutMs) });
+        if (!health.ok)
+            return { status: "down", servedModelIds: [] };
+    }
+    catch {
+        return { status: "down", servedModelIds: [] };
+    }
+    try {
+        const models = await fetchFn(`${serverUrl}/v1/models`, { signal: AbortSignal.timeout(timeoutMs) });
+        if (!models.ok)
+            return { status: "conflict", servedModelIds: [] };
+        const payload = (await models.json());
+        const servedModelIds = Array.isArray(payload.data)
+            ? payload.data
+                .map((entry) => typeof entry === "object" && entry !== null && typeof entry.id === "string"
+                ? entry.id.slice(0, MAX_MODEL_ID_CHARS)
+                : undefined)
+                .filter((id) => id !== undefined)
+                .slice(0, MAX_REPORTED_MODEL_IDS)
+            : [];
+        return {
+            status: servedModelIds.includes(expectedModelId) ? "matching" : "conflict",
+            servedModelIds,
+        };
+    }
+    catch {
+        return { status: "conflict", servedModelIds: [] };
+    }
+}
+//# sourceMappingURL=prism-llamacpp-server-probe.js.map

package/dist/core/models/prism-llamacpp-server-probe.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"prism-llamacpp-server-probe.js","sourceRoot":"","sources":["../../../src/core/models/prism-llamacpp-server-probe.ts"],"names":[],"mappings":"AAIA,MAAM,wBAAwB,GAAG,KAAK,CAAC;AACvC,MAAM,sBAAsB,GAAG,CAAC,CAAC;AACjC,MAAM,kBAAkB,GAAG,GAAG,CAAC;AAE/B,mGAAmG;AACnG,MAAM,CAAC,KAAK,UAAU,wBAAwB,CAC7C,SAAiB,EACjB,eAAuB,EACvB,OAAO,GAAiB,KAAK,EAC7B,SAAS,GAAW,wBAAwB;IAE5C,IAAI,CAAC;QACJ,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,SAAS,SAAS,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;QAChG,IAAI,CAAC,MAAM,CAAC,EAAE;YAAE,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,EAAE,EAAE,CAAC;IAC/D,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,cAAc,EAAE,EAAE,EAAE,CAAC;IAC/C,CAAC;IAED,IAAI,CAAC;QACJ,MAAM,MAAM,GAAG,MAAM,OAAO,CAAC,GAAG,SAAS,YAAY,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC;QACnG,IAAI,CAAC,MAAM,CAAC,EAAE;YAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,cAAc,EAAE,EAAE,EAAE,CAAC;QAClE,MAAM,OAAO,GAAG,CAAC,MAAM,MAAM,CAAC,IAAI,EAAE,CAAuB,CAAC;QAC5D,MAAM,cAAc,GAAG,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC;YACjD,CAAC,CAAC,OAAO,CAAC,IAAI;iBACX,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CACd,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,KAAK,IAAI,IAAI,OAAQ,KAA0B,CAAC,EAAE,KAAK,QAAQ;gBAChG,CAAC,CAAE,KAAwB,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,kBAAkB,CAAC;gBAC3D,CAAC,CAAC,SAAS,CACZ;iBACA,MAAM,CAAC,CAAC,EAAE,EAAgB,EAAE,CAAC,EAAE,KAAK,SAAS,CAAC;iBAC9C,KAAK,CAAC,CAAC,EAAE,sBAAsB,CAAC;YACnC,CAAC,CAAC,EAAE,CAAC;QACN,OAAO;YACN,MAAM,EAAE,cAAc,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,UAAU;YAC1E,cAAc;SACd,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,cAAc,EAAE,EAAE,EAAE,CAAC;IACnD,CAAC;AACF,CAAC","sourcesContent":["export type PrismLlamaCppServerProbe =\n\t| { status: \"down\"; servedModelIds: [] }\n\t| { status: \"matching\" | \"conflict\"; servedModelIds: string[] };\n\nconst DEFAULT_PROBE_TIMEOUT_MS = 2_000;\nconst MAX_REPORTED_MODEL_IDS = 8;\nconst MAX_MODEL_ID_CHARS = 256;\n\n/** Health alone is insufficient on a shared port: readiness requires the exact requested alias. */\nexport async function probePrismLlamaCppServer(\n\tserverUrl: string,\n\texpectedModelId: string,\n\tfetchFn: typeof fetch = fetch,\n\ttimeoutMs: number = DEFAULT_PROBE_TIMEOUT_MS,\n): Promise<PrismLlamaCppServerProbe> {\n\ttry {\n\t\tconst health = await fetchFn(`${serverUrl}/health`, { signal: AbortSignal.timeout(timeoutMs) });\n\t\tif (!health.ok) return { status: \"down\", servedModelIds: [] };\n\t} catch {\n\t\treturn { status: \"down\", servedModelIds: [] };\n\t}\n\n\ttry {\n\t\tconst models = await fetchFn(`${serverUrl}/v1/models`, { signal: AbortSignal.timeout(timeoutMs) });\n\t\tif (!models.ok) return { status: \"conflict\", servedModelIds: [] };\n\t\tconst payload = (await models.json()) as { data?: unknown };\n\t\tconst servedModelIds = Array.isArray(payload.data)\n\t\t\t? payload.data\n\t\t\t\t\t.map((entry) =>\n\t\t\t\t\t\ttypeof entry === \"object\" && entry !== null && typeof (entry as { id?: unknown }).id === \"string\"\n\t\t\t\t\t\t\t? (entry as { id: string }).id.slice(0, MAX_MODEL_ID_CHARS)\n\t\t\t\t\t\t\t: undefined,\n\t\t\t\t\t)\n\t\t\t\t\t.filter((id): id is string => id !== undefined)\n\t\t\t\t\t.slice(0, MAX_REPORTED_MODEL_IDS)\n\t\t\t: [];\n\t\treturn {\n\t\t\tstatus: servedModelIds.includes(expectedModelId) ? \"matching\" : \"conflict\",\n\t\t\tservedModelIds,\n\t\t};\n\t} catch {\n\t\treturn { status: \"conflict\", servedModelIds: [] };\n\t}\n}\n"]}

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

@@ -16,6 +16,9 @@ export interface RuntimeLoadRequest {
     role: RuntimeRole;
     priority: number;
     nowMs: number;
+    /** Refuse an immediate reload of a model that was just evicted for the current resident. */
+    antiThrashWindowMs?: number;
+    /** Protect every newly-used resident from eviction for this duration. */
     minDwellMs?: number;
     pinActiveModel?: string;
     /** Adapter identity for this request; set by RuntimeResidencyArbiter before planning. */

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

@@ -1 +1 @@
-{"version":3,"file":"runtime-arbiter.d.ts","sourceRoot":"","sources":["../../../src/core/models/runtime-arbiter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAE7E,MAAM,MAAM,gBAAgB,GAAG,MAAM,GAAG,YAAY,GAAG,UAAU,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,QAAQ,GAAG,OAAO,GAAG,OAAO,CAAC;AAElE,MAAM,WAAW,oBAAoB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,gBAAgB,EAAE,gBAAgB,CAAC;CACnC;AAED,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,WAAW,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,yFAAyF;IACzF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mFAAmF;IACnF,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,YAAY,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACpC,eAAe,CAAC,EAAE,qBAAqB,EAAE,CAAC;IAC1C,sGAAsG;IACtG,SAAS,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,+EAA+E;IAC/E,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,qBAAqB;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,mFAAmF;IACnF,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,eAAe,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,oBAAoB,GAC7B;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,oBAAoB,EAAE,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,GAC7F;IAAE,MAAM,EAAE,QAAQ,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,oBAAoB,EAAE,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;AAEnH,MAAM,WAAW,uBAAuB;IACvC,EAAE,EAAE,MAAM,CAAC;IACX,gBAAgB,EAAE,gBAAgB,CAAC;IACnC,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAAC;IACxC,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7C,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACtC;AAED,qBAAa,uBAAuB;IACnC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAuC;IAEhE,YAAY,IAAI,EAAE;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,uBAAuB,EAAE,CAAA;KAAE,EAG7E;IAEK,cAAc,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,oBAAoB,CAAC,CA6BlG;CACD;AAED,qBAAa,6BAA8B,YAAW,uBAAuB;IAC5E,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,YAAY,CAAU;IAClD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAmF;IAC3G,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IAErC,YACC,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,IAAI,CAAC,aAAa,EAAE,oBAAoB,GAAG,gBAAgB,GAAG,iBAAiB,CAAC,EACzF,OAAO,GAAE;QAAE,KAAK,CAAC,EAAE,MAAM,MAAM,CAAA;KAAO,EAKtC;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAU5C;IAEK,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAGjD;IAEK,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAE1C;CACD;AAED,qBAAa,mCAAoC,YAAW,uBAAuB;IAClF,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,MAAM,CAAU;IAC5C,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAyD;IACjF,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAS;IAC/B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAS;IAC/B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IAErC,YACC,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,IAAI,CAAC,mBAAmB,EAAE,QAAQ,GAAG,OAAO,GAAG,MAAM,CAAC,EAC/D,KAAK,EAAE,MAAM,EACb,KAAK,EAAE,MAAM,EACb,OAAO,GAAE;QAAE,KAAK,CAAC,EAAE,MAAM,MAAM,CAAA;KAAO,EAOtC;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAa5C;IAEK,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAGlD;IAEK,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAE3C;CACD;AAED,qBAAa,+BAAgC,YAAW,uBAAuB;IAC9E,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,UAAU,CAAU;IAChD,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAyB;IAEnD,YAAY,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,oBAAoB,EAAE,WAAW,GAAG,kBAAkB,CAAC,EAAE,EAOhG;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAE5C;IAEK,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAG;IAEhD,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAG;CAC/C;AAED,wBAAgB,oBAAoB,CAAC,IAAI,EAAE;IAC1C,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,SAAS,oBAAoB,EAAE,CAAC;IAC3C,OAAO,EAAE,kBAAkB,CAAC;CAC5B,GAAG,oBAAoB,CAsEvB"}
+{"version":3,"file":"runtime-arbiter.d.ts","sourceRoot":"","sources":["../../../src/core/models/runtime-arbiter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAC;AAE7E,MAAM,MAAM,gBAAgB,GAAG,MAAM,GAAG,YAAY,GAAG,UAAU,CAAC;AAClE,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,QAAQ,GAAG,OAAO,GAAG,OAAO,CAAC;AAElE,MAAM,WAAW,oBAAoB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,gBAAgB,EAAE,gBAAgB,CAAC;CACnC;AAED,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,WAAW,CAAC;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,4FAA4F;IAC5F,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,yEAAyE;IACzE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,yFAAyF;IACzF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,mFAAmF;IACnF,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,YAAY,CAAC,EAAE,kBAAkB,EAAE,CAAC;IACpC,eAAe,CAAC,EAAE,qBAAqB,EAAE,CAAC;IAC1C,sGAAsG;IACtG,SAAS,CAAC,EAAE,OAAO,CAAC;CACpB;AAED,MAAM,WAAW,kBAAkB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,+EAA+E;IAC/E,SAAS,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,qBAAqB;IACrC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,IAAI,EAAE,MAAM,CAAC;IACb,mFAAmF;IACnF,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,eAAe,CAAC,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,MAAM,oBAAoB,GAC7B;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,oBAAoB,EAAE,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,GAC7F;IAAE,MAAM,EAAE,QAAQ,CAAC;IAAC,MAAM,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,oBAAoB,EAAE,CAAC;IAAC,aAAa,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC;AAEnH,MAAM,WAAW,uBAAuB;IACvC,EAAE,EAAE,MAAM,CAAC;IACX,gBAAgB,EAAE,gBAAgB,CAAC;IACnC,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAAC;IACxC,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7C,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CACtC;AAED,qBAAa,uBAAuB;IACnC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAuC;IAEhE,YAAY,IAAI,EAAE;QAAE,WAAW,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,uBAAuB,EAAE,CAAA;KAAE,EAG7E;IAEK,cAAc,CAAC,SAAS,EAAE,MAAM,EAAE,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,oBAAoB,CAAC,CA6BlG;CACD;AAED,qBAAa,6BAA8B,YAAW,uBAAuB;IAC5E,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,YAAY,CAAU;IAClD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAmF;IAC3G,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IAErC,YACC,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,IAAI,CAAC,aAAa,EAAE,oBAAoB,GAAG,gBAAgB,GAAG,iBAAiB,CAAC,EACzF,OAAO,GAAE;QAAE,KAAK,CAAC,EAAE,MAAM,MAAM,CAAA;KAAO,EAKtC;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAU5C;IAEK,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAGjD;IAEK,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAE1C;CACD;AAED,qBAAa,mCAAoC,YAAW,uBAAuB;IAClF,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,MAAM,CAAU;IAC5C,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAyD;IACjF,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAS;IAC/B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAS;IAC/B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAe;IAErC,YACC,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,IAAI,CAAC,mBAAmB,EAAE,QAAQ,GAAG,OAAO,GAAG,MAAM,CAAC,EAC/D,KAAK,EAAE,MAAM,EACb,KAAK,EAAE,MAAM,EACb,OAAO,GAAE;QAAE,KAAK,CAAC,EAAE,MAAM,MAAM,CAAA;KAAO,EAOtC;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAa5C;IAEK,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAGlD;IAEK,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAE3C;CACD;AAED,qBAAa,+BAAgC,YAAW,uBAAuB;IAC9E,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,QAAQ,CAAC,gBAAgB,EAAG,UAAU,CAAU;IAChD,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAyB;IAEnD,YAAY,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,oBAAoB,EAAE,WAAW,GAAG,kBAAkB,CAAC,EAAE,EAOhG;IAEK,IAAI,IAAI,OAAO,CAAC,oBAAoB,EAAE,CAAC,CAE5C;IAEK,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAG;IAEhD,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAAG;CAC/C;AAED,wBAAgB,oBAAoB,CAAC,IAAI,EAAE;IAC1C,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,SAAS,oBAAoB,EAAE,CAAC;IAC3C,OAAO,EAAE,kBAAkB,CAAC;CAC5B,GAAG,oBAAoB,CAsEvB"}

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

@@ -165,10 +165,11 @@ export function planRuntimeResidency(args) {
     };
 }
 function hasPingPongRisk(request) {
+    const windowMs = request.antiThrashWindowMs ?? request.minDwellMs ?? 0;
     return (request.recentEvictions ?? []).some((record) => record.evicted === request.model &&
         (request.adapterId === undefined ||
             record.evictedAdapterId === undefined ||
             record.evictedAdapterId === request.adapterId) &&
-        request.nowMs - record.atMs <= (request.minDwellMs ?? 0));
+        request.nowMs - record.atMs <= windowMs);
 }
 //# sourceMappingURL=runtime-arbiter.js.map

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

@@ -1 +1 @@
-{"version":3,"file":"runtime-arbiter.js","sourceRoot":"","sources":["../../../src/core/models/runtime-arbiter.ts"],"names":[],"mappings":"AA8DA,MAAM,OAAO,uBAAuB;IAInC,YAAY,IAAkE;QAC7E,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC;QACpC,IAAI,CAAC,QAAQ,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChF,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,SAAiB,EAAE,OAA2B;QAClE,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,SAAS,GAAG,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACvG,IAAI,CAAC,OAAO,EAAE,CAAC;YACd,OAAO;gBACN,MAAM,EAAE,QAAQ;gBAChB,MAAM,EAAE,iBAAiB;gBACzB,KAAK,EAAE,EAAE;gBACT,aAAa,EAAE,OAAO,CAAC,KAAK;gBAC5B,WAAW,EAAE,IAAI,CAAC,WAAW;aAC7B,CAAC;QACH,CAAC;QACD,MAAM,iBAAiB,GAAuB;YAC7C,GAAG,OAAO;YACV,SAAS;YACT,GAAG,CAAC,OAAO,CAAC,cAAc,KAAK,SAAS,IAAI,OAAO,CAAC,kBAAkB,KAAK,SAAS;gBACnF,CAAC,CAAC,EAAE,kBAAkB,EAAE,SAAS,EAAE;gBACnC,CAAC,CAAC,EAAE,CAAC;SACN,CAAC;QACF,MAAM,IAAI,GAAG,oBAAoB,CAAC,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,iBAAiB,EAAE,CAAC,CAAC;QAC5G,IAAI,IAAI,CAAC,MAAM,KAAK,MAAM;YAAE,OAAO,IAAI,CAAC;QACxC,KAAK,MAAM,QAAQ,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACnC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QACtE,CAAC;QACD,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CACrC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,SAAS,KAAK,SAAS,IAAI,QAAQ,CAAC,KAAK,KAAK,OAAO,CAAC,KAAK,CAClF,CAAC;QACF,IAAI,OAAO,CAAC,SAAS,KAAK,KAAK,IAAI,CAAC,eAAe;YAAE,MAAM,OAAO,CAAC,cAAc,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QACjG,OAAO,IAAI,CAAC;IACb,CAAC;CACD;AAED,MAAM,OAAO,6BAA6B;IAMzC,YACC,EAAU,EACV,OAAyF,EACzF,OAAO,GAA6B,EAAE;QAP9B,qBAAgB,GAAG,YAAqB,CAAC;QASjD,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC;IACxC,CAAC;IAED,KAAK,CAAC,IAAI;QACT,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC;QACzB,OAAO,CAAC,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YAChE,SAAS,EAAE,IAAI,CAAC,EAAE;YAClB,KAAK,EAAE,KAAK,CAAC,IAAI;YACjB,KAAK,EAAE,KAAK,CAAC,SAAS;YACtB,YAAY,EAAE,GAAG;YACjB,QAAQ,EAAE,EAAE;YACZ,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;SACvC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,KAAa;QACjC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QACxD,IAAI,CAAC,MAAM,CAAC,EAAE;YAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC/C,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAAa;QAC1B,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC;IAC3C,CAAC;CACD;AAED,MAAM,OAAO,mCAAmC;IAQ/C,YACC,EAAU,EACV,OAA+D,EAC/D,KAAa,EACb,KAAa,EACb,OAAO,GAA6B,EAAE;QAX9B,qBAAgB,GAAG,MAAe,CAAC;QAa3C,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC;IACxC,CAAC;IAED,KAAK,CAAC,IAAI;QACT,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;QAC3C,IAAI,CAAC,MAAM,CAAC,QAAQ;YAAE,OAAO,EAAE,CAAC;QAChC,OAAO;YACN;gBACC,SAAS,EAAE,IAAI,CAAC,EAAE;gBAClB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,YAAY,EAAE,IAAI,CAAC,KAAK,EAAE;gBAC1B,QAAQ,EAAE,EAAE;gBACZ,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;aACvC;SACD,CAAC;IACH,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,MAAc;QAClC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QAC1C,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,MAAM,KAAK,iBAAiB;YAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAC5F,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,MAAc;QAC3B,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;IACrB,CAAC;CACD;AAED,MAAM,OAAO,+BAA+B;IAK3C,YAAY,EAAU,EAAE,SAAyE;QAHxF,qBAAgB,GAAG,UAAmB,CAAC;QAI/C,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;YAC7C,GAAG,QAAQ;YACX,SAAS,EAAE,EAAE;YACb,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;SACvC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,IAAI;QACT,OAAO,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC;IAC5B,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,MAAc,IAAkB,CAAC;IAEtD,KAAK,CAAC,OAAO,CAAC,MAAc,IAAkB,CAAC;CAC/C;AAED,MAAM,UAAU,oBAAoB,CAAC,IAIpC;IACA,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkF,CAAC;IAClH,MAAM,iBAAiB,GAAG,CAAC,KAA6E,EAAQ,EAAE;QACjH,MAAM,GAAG,GAAG,GAAG,KAAK,CAAC,SAAS,IAAI,GAAG,KAAK,KAAK,CAAC,KAAK,EAAE,CAAC;QACxD,MAAM,QAAQ,GAAG,eAAe,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC1C,IAAI,CAAC,QAAQ,IAAI,KAAK,CAAC,KAAK,GAAG,QAAQ,CAAC,KAAK;YAAE,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;IAChF,CAAC,CAAC;IACF,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAChC,KAAK,MAAM,WAAW,IAAI,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC;QAC3D,iBAAiB,CAAC;YACjB,GAAG,WAAW;YACd,SAAS,EACR,WAAW,CAAC,SAAS,IAAI,CAAC,WAAW,CAAC,KAAK,KAAK,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;SACzG,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,oBAAoB,GAAG,CAC5B,QAA8B,EAC9B,SAAgD,EACtC,EAAE,CACZ,QAAQ,CAAC,KAAK,KAAK,SAAS,CAAC,KAAK;QAClC,CAAC,SAAS,CAAC,SAAS,KAAK,SAAS,IAAI,QAAQ,CAAC,SAAS,KAAK,SAAS,CAAC,SAAS,CAAC,CAAC;IACnF,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACxF,MAAM,YAAY,GAAG,CAAC,GAAG,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,SAAS,EAAE,EAAE;QAC5E,OAAO,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,oBAAoB,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,SAAS,CAAC,KAAK,CAAC;IACnH,CAAC,EAAE,CAAC,CAAC,CAAC;IACN,MAAM,aAAa,GAAG,aAAa,GAAG,YAAY,CAAC;IACnD,IAAI,aAAa,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACvC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;IACpF,CAAC;IAED,IAAI,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,EAAE,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;IAC7G,CAAC;IAED,IAAI,cAAc,GAAG,aAAa,CAAC;IACnC,MAAM,KAAK,GAA2B,EAAE,CAAC;IACzC,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS;SAC/B,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,oBAAoB,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC,CAC3G;SACA,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CACZ,CAAC,QAAQ,CAAC,MAAM;QAChB,CAAC,CACA,QAAQ,CAAC,KAAK,KAAK,IAAI,CAAC,OAAO,CAAC,cAAc;YAC9C,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KAAK,SAAS,IAAI,QAAQ,CAAC,SAAS,KAAK,IAAI,CAAC,OAAO,CAAC,kBAAkB,CAAC,CACzG,CACF;SACA,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,gBAAgB,KAAK,UAAU,CAAC;SAC9D,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CACZ,IAAI,CAAC,OAAO,CAAC,UAAU,KAAK,SAAS;QACrC,IAAI,CAAC,OAAO,CAAC,KAAK,GAAG,QAAQ,CAAC,YAAY,IAAI,IAAI,CAAC,OAAO,CAAC,UAAU,CACtE;SACA,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,IAAI,IAAI,CAAC,YAAY,GAAG,KAAK,CAAC,YAAY,CAAC,CAAC;IAClG,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,cAAc,IAAI,SAAS,CAAC,KAAK,CAAC;QAClC,IAAI,cAAc,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YACxC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;QAChF,CAAC;IACF,CAAC;IACD,OAAO;QACN,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,+BAA+B;QACvC,KAAK;QACL,aAAa;QACb,WAAW,EAAE,IAAI,CAAC,WAAW;KAC7B,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,OAA2B;IACnD,OAAO,CAAC,OAAO,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC,IAAI,CAC1C,CAAC,MAAM,EAAE,EAAE,CACV,MAAM,CAAC,OAAO,KAAK,OAAO,CAAC,KAAK;QAChC,CAAC,OAAO,CAAC,SAAS,KAAK,SAAS;YAC/B,MAAM,CAAC,gBAAgB,KAAK,SAAS;YACrC,MAAM,CAAC,gBAAgB,KAAK,OAAO,CAAC,SAAS,CAAC;QAC/C,OAAO,CAAC,KAAK,GAAG,MAAM,CAAC,IAAI,IAAI,CAAC,OAAO,CAAC,UAAU,IAAI,CAAC,CAAC,CACzD,CAAC;AACH,CAAC","sourcesContent":["import type { OllamaRuntime, TransformersRuntime } from \"./local-runtime.ts\";\n\nexport type ResidencyControl = \"full\" | \"keep-alive\" | \"advisory\";\nexport type RuntimeRole = \"active\" | \"router\" | \"bench\" | \"probe\";\n\nexport interface RuntimeResidentModel {\n\tadapterId: string;\n\tmodel: string;\n\tbytes: number;\n\tlastUsedAtMs: number;\n\tpriority: number;\n\tpinned?: boolean;\n\tresidencyControl: ResidencyControl;\n}\n\nexport interface RuntimeLoadRequest {\n\tmodel: string;\n\tbytes: number;\n\trole: RuntimeRole;\n\tpriority: number;\n\tnowMs: number;\n\tminDwellMs?: number;\n\tpinActiveModel?: string;\n\t/** Adapter identity for this request; set by RuntimeResidencyArbiter before planning. */\n\tadapterId?: string;\n\t/** When pinActiveModel is set, constrain that pin to this adapter when present. */\n\tpinActiveAdapterId?: string;\n\treservations?: RuntimeReservation[];\n\trecentEvictions?: RuntimeEvictionRecord[];\n\t/** False performs admission/eviction only; the real model request owns cold loading. Default true. */\n\tloadModel?: boolean;\n}\n\nexport interface RuntimeReservation {\n\tmodel: string;\n\tbytes: number;\n\tpriority: number;\n\t/** Optional runtime identity; omitted means any adapter serving this model. */\n\tadapterId?: string;\n}\n\nexport interface RuntimeEvictionRecord {\n\tevicted: string;\n\tloaded: string;\n\tatMs: number;\n\t/** Optional identities preserve anti-thrash separation across runtime adapters. */\n\tevictedAdapterId?: string;\n\tloadedAdapterId?: string;\n}\n\nexport type RuntimeResidencyPlan =\n\t| { status: \"fits\"; evict: RuntimeResidentModel[]; requiredBytes: number; budgetBytes: number }\n\t| { status: \"refuse\"; reason: string; evict: RuntimeResidentModel[]; requiredBytes: number; budgetBytes: number };\n\nexport interface RuntimeResidencyAdapter {\n\tid: string;\n\tresidencyControl: ResidencyControl;\n\tlist(): Promise<RuntimeResidentModel[]>;\n\tensureResident(model: string): Promise<void>;\n\trelease(model: string): Promise<void>;\n}\n\nexport class RuntimeResidencyArbiter {\n\tprivate readonly budgetBytes: number;\n\tprivate readonly adapters: Map<string, RuntimeResidencyAdapter>;\n\n\tconstructor(args: { budgetBytes: number; adapters: RuntimeResidencyAdapter[] }) {\n\t\tthis.budgetBytes = args.budgetBytes;\n\t\tthis.adapters = new Map(args.adapters.map((adapter) => [adapter.id, adapter]));\n\t}\n\n\tasync ensureResident(adapterId: string, request: RuntimeLoadRequest): Promise<RuntimeResidencyPlan> {\n\t\tconst adapter = this.adapters.get(adapterId);\n\t\tconst residents = (await Promise.all([...this.adapters.values()].map((entry) => entry.list()))).flat();\n\t\tif (!adapter) {\n\t\t\treturn {\n\t\t\t\tstatus: \"refuse\",\n\t\t\t\treason: \"unknown-adapter\",\n\t\t\t\tevict: [],\n\t\t\t\trequiredBytes: request.bytes,\n\t\t\t\tbudgetBytes: this.budgetBytes,\n\t\t\t};\n\t\t}\n\t\tconst requestForAdapter: RuntimeLoadRequest = {\n\t\t\t...request,\n\t\t\tadapterId,\n\t\t\t...(request.pinActiveModel !== undefined && request.pinActiveAdapterId === undefined\n\t\t\t\t? { pinActiveAdapterId: adapterId }\n\t\t\t\t: {}),\n\t\t};\n\t\tconst plan = planRuntimeResidency({ budgetBytes: this.budgetBytes, residents, request: requestForAdapter });\n\t\tif (plan.status !== \"fits\") return plan;\n\t\tfor (const resident of plan.evict) {\n\t\t\tawait this.adapters.get(resident.adapterId)?.release(resident.model);\n\t\t}\n\t\tconst alreadyResident = residents.some(\n\t\t\t(resident) => resident.adapterId === adapterId && resident.model === request.model,\n\t\t);\n\t\tif (request.loadModel !== false && !alreadyResident) await adapter.ensureResident(request.model);\n\t\treturn plan;\n\t}\n}\n\nexport class OllamaRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"keep-alive\" as const;\n\tprivate readonly runtime: Pick<OllamaRuntime, \"listResidentModels\" | \"ensureResident\" | \"releaseResident\">;\n\tprivate readonly nowMs: () => number;\n\n\tconstructor(\n\t\tid: string,\n\t\truntime: Pick<OllamaRuntime, \"listResidentModels\" | \"ensureResident\" | \"releaseResident\">,\n\t\toptions: { nowMs?: () => number } = {},\n\t) {\n\t\tthis.id = id;\n\t\tthis.runtime = runtime;\n\t\tthis.nowMs = options.nowMs ?? Date.now;\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\tconst now = this.nowMs();\n\t\treturn (await this.runtime.listResidentModels()).map((model) => ({\n\t\t\tadapterId: this.id,\n\t\t\tmodel: model.name,\n\t\t\tbytes: model.sizeBytes,\n\t\t\tlastUsedAtMs: now,\n\t\t\tpriority: 10,\n\t\t\tresidencyControl: this.residencyControl,\n\t\t}));\n\t}\n\n\tasync ensureResident(model: string): Promise<void> {\n\t\tconst result = await this.runtime.ensureResident(model);\n\t\tif (!result.ok) throw new Error(result.error);\n\t}\n\n\tasync release(model: string): Promise<void> {\n\t\tawait this.runtime.releaseResident(model);\n\t}\n}\n\nexport class TransformersRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"full\" as const;\n\tprivate readonly runtime: Pick<TransformersRuntime, \"detect\" | \"start\" | \"stop\">;\n\tprivate readonly model: string;\n\tprivate readonly bytes: number;\n\tprivate readonly nowMs: () => number;\n\n\tconstructor(\n\t\tid: string,\n\t\truntime: Pick<TransformersRuntime, \"detect\" | \"start\" | \"stop\">,\n\t\tmodel: string,\n\t\tbytes: number,\n\t\toptions: { nowMs?: () => number } = {},\n\t) {\n\t\tthis.id = id;\n\t\tthis.runtime = runtime;\n\t\tthis.model = model;\n\t\tthis.bytes = bytes;\n\t\tthis.nowMs = options.nowMs ?? Date.now;\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\tconst status = await this.runtime.detect();\n\t\tif (!status.serverUp) return [];\n\t\treturn [\n\t\t\t{\n\t\t\t\tadapterId: this.id,\n\t\t\t\tmodel: this.model,\n\t\t\t\tbytes: this.bytes,\n\t\t\t\tlastUsedAtMs: this.nowMs(),\n\t\t\t\tpriority: 10,\n\t\t\t\tresidencyControl: this.residencyControl,\n\t\t\t},\n\t\t];\n\t}\n\n\tasync ensureResident(_model: string): Promise<void> {\n\t\tconst result = await this.runtime.start();\n\t\tif (!result.started && result.reason !== \"already_running\") throw new Error(result.reason);\n\t}\n\n\tasync release(_model: string): Promise<void> {\n\t\tthis.runtime.stop();\n\t}\n}\n\nexport class AdvisoryRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"advisory\" as const;\n\tprivate readonly residents: RuntimeResidentModel[];\n\n\tconstructor(id: string, residents: Omit<RuntimeResidentModel, \"adapterId\" | \"residencyControl\">[]) {\n\t\tthis.id = id;\n\t\tthis.residents = residents.map((resident) => ({\n\t\t\t...resident,\n\t\t\tadapterId: id,\n\t\t\tresidencyControl: this.residencyControl,\n\t\t}));\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\treturn [...this.residents];\n\t}\n\n\tasync ensureResident(_model: string): Promise<void> {}\n\n\tasync release(_model: string): Promise<void> {}\n}\n\nexport function planRuntimeResidency(args: {\n\tbudgetBytes: number;\n\tresidents: readonly RuntimeResidentModel[];\n\trequest: RuntimeLoadRequest;\n}): RuntimeResidencyPlan {\n\tconst requestedModels = new Map<string, { model: string; adapterId?: string; bytes: number; priority: number }>();\n\tconst addRequestedModel = (entry: { model: string; adapterId?: string; bytes: number; priority: number }): void => {\n\t\tconst key = `${entry.adapterId ?? \"*\"}\\0${entry.model}`;\n\t\tconst existing = requestedModels.get(key);\n\t\tif (!existing || entry.bytes > existing.bytes) requestedModels.set(key, entry);\n\t};\n\taddRequestedModel(args.request);\n\tfor (const reservation of args.request.reservations ?? []) {\n\t\taddRequestedModel({\n\t\t\t...reservation,\n\t\t\tadapterId:\n\t\t\t\treservation.adapterId ?? (reservation.model === args.request.model ? args.request.adapterId : undefined),\n\t\t});\n\t}\n\n\tconst isResidentForRequest = (\n\t\tresident: RuntimeResidentModel,\n\t\trequested: { model: string; adapterId?: string },\n\t): boolean =>\n\t\tresident.model === requested.model &&\n\t\t(requested.adapterId === undefined || resident.adapterId === requested.adapterId);\n\tconst residentBytes = args.residents.reduce((sum, resident) => sum + resident.bytes, 0);\n\tconst missingBytes = [...requestedModels.values()].reduce((sum, requested) => {\n\t\treturn args.residents.some((resident) => isResidentForRequest(resident, requested)) ? sum : sum + requested.bytes;\n\t}, 0);\n\tconst requiredBytes = residentBytes + missingBytes;\n\tif (requiredBytes <= args.budgetBytes) {\n\t\treturn { status: \"fits\", evict: [], requiredBytes, budgetBytes: args.budgetBytes };\n\t}\n\n\tif (hasPingPongRisk(args.request)) {\n\t\treturn { status: \"refuse\", reason: \"anti-thrash\", evict: [], requiredBytes, budgetBytes: args.budgetBytes };\n\t}\n\n\tlet projectedBytes = requiredBytes;\n\tconst evict: RuntimeResidentModel[] = [];\n\tconst candidates = args.residents\n\t\t.filter(\n\t\t\t(resident) => ![...requestedModels.values()].some((requested) => isResidentForRequest(resident, requested)),\n\t\t)\n\t\t.filter(\n\t\t\t(resident) =>\n\t\t\t\t!resident.pinned &&\n\t\t\t\t!(\n\t\t\t\t\tresident.model === args.request.pinActiveModel &&\n\t\t\t\t\t(args.request.pinActiveAdapterId === undefined || resident.adapterId === args.request.pinActiveAdapterId)\n\t\t\t\t),\n\t\t)\n\t\t.filter((resident) => resident.residencyControl !== \"advisory\")\n\t\t.filter(\n\t\t\t(resident) =>\n\t\t\t\targs.request.minDwellMs === undefined ||\n\t\t\t\targs.request.nowMs - resident.lastUsedAtMs >= args.request.minDwellMs,\n\t\t)\n\t\t.sort((left, right) => left.priority - right.priority || left.lastUsedAtMs - right.lastUsedAtMs);\n\tfor (const candidate of candidates) {\n\t\tevict.push(candidate);\n\t\tprojectedBytes -= candidate.bytes;\n\t\tif (projectedBytes <= args.budgetBytes) {\n\t\t\treturn { status: \"fits\", evict, requiredBytes, budgetBytes: args.budgetBytes };\n\t\t}\n\t}\n\treturn {\n\t\tstatus: \"refuse\",\n\t\treason: \"insufficient-evictable-memory\",\n\t\tevict,\n\t\trequiredBytes,\n\t\tbudgetBytes: args.budgetBytes,\n\t};\n}\n\nfunction hasPingPongRisk(request: RuntimeLoadRequest): boolean {\n\treturn (request.recentEvictions ?? []).some(\n\t\t(record) =>\n\t\t\trecord.evicted === request.model &&\n\t\t\t(request.adapterId === undefined ||\n\t\t\t\trecord.evictedAdapterId === undefined ||\n\t\t\t\trecord.evictedAdapterId === request.adapterId) &&\n\t\t\trequest.nowMs - record.atMs <= (request.minDwellMs ?? 0),\n\t);\n}\n"]}
+{"version":3,"file":"runtime-arbiter.js","sourceRoot":"","sources":["../../../src/core/models/runtime-arbiter.ts"],"names":[],"mappings":"AAiEA,MAAM,OAAO,uBAAuB;IAInC,YAAY,IAAkE;QAC7E,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,CAAC;QACpC,IAAI,CAAC,QAAQ,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChF,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,SAAiB,EAAE,OAA2B;QAClE,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAC7C,MAAM,SAAS,GAAG,CAAC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACvG,IAAI,CAAC,OAAO,EAAE,CAAC;YACd,OAAO;gBACN,MAAM,EAAE,QAAQ;gBAChB,MAAM,EAAE,iBAAiB;gBACzB,KAAK,EAAE,EAAE;gBACT,aAAa,EAAE,OAAO,CAAC,KAAK;gBAC5B,WAAW,EAAE,IAAI,CAAC,WAAW;aAC7B,CAAC;QACH,CAAC;QACD,MAAM,iBAAiB,GAAuB;YAC7C,GAAG,OAAO;YACV,SAAS;YACT,GAAG,CAAC,OAAO,CAAC,cAAc,KAAK,SAAS,IAAI,OAAO,CAAC,kBAAkB,KAAK,SAAS;gBACnF,CAAC,CAAC,EAAE,kBAAkB,EAAE,SAAS,EAAE;gBACnC,CAAC,CAAC,EAAE,CAAC;SACN,CAAC;QACF,MAAM,IAAI,GAAG,oBAAoB,CAAC,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,SAAS,EAAE,OAAO,EAAE,iBAAiB,EAAE,CAAC,CAAC;QAC5G,IAAI,IAAI,CAAC,MAAM,KAAK,MAAM;YAAE,OAAO,IAAI,CAAC;QACxC,KAAK,MAAM,QAAQ,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACnC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;QACtE,CAAC;QACD,MAAM,eAAe,GAAG,SAAS,CAAC,IAAI,CACrC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,SAAS,KAAK,SAAS,IAAI,QAAQ,CAAC,KAAK,KAAK,OAAO,CAAC,KAAK,CAClF,CAAC;QACF,IAAI,OAAO,CAAC,SAAS,KAAK,KAAK,IAAI,CAAC,eAAe;YAAE,MAAM,OAAO,CAAC,cAAc,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;QACjG,OAAO,IAAI,CAAC;IACb,CAAC;CACD;AAED,MAAM,OAAO,6BAA6B;IAMzC,YACC,EAAU,EACV,OAAyF,EACzF,OAAO,GAA6B,EAAE;QAP9B,qBAAgB,GAAG,YAAqB,CAAC;QASjD,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC;IACxC,CAAC;IAED,KAAK,CAAC,IAAI;QACT,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC;QACzB,OAAO,CAAC,MAAM,IAAI,CAAC,OAAO,CAAC,kBAAkB,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;YAChE,SAAS,EAAE,IAAI,CAAC,EAAE;YAClB,KAAK,EAAE,KAAK,CAAC,IAAI;YACjB,KAAK,EAAE,KAAK,CAAC,SAAS;YACtB,YAAY,EAAE,GAAG;YACjB,QAAQ,EAAE,EAAE;YACZ,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;SACvC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,KAAa;QACjC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;QACxD,IAAI,CAAC,MAAM,CAAC,EAAE;YAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IAC/C,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,KAAa;QAC1B,MAAM,IAAI,CAAC,OAAO,CAAC,eAAe,CAAC,KAAK,CAAC,CAAC;IAC3C,CAAC;CACD;AAED,MAAM,OAAO,mCAAmC;IAQ/C,YACC,EAAU,EACV,OAA+D,EAC/D,KAAa,EACb,KAAa,EACb,OAAO,GAA6B,EAAE;QAX9B,qBAAgB,GAAG,MAAe,CAAC;QAa3C,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;QACvB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,GAAG,CAAC;IACxC,CAAC;IAED,KAAK,CAAC,IAAI;QACT,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;QAC3C,IAAI,CAAC,MAAM,CAAC,QAAQ;YAAE,OAAO,EAAE,CAAC;QAChC,OAAO;YACN;gBACC,SAAS,EAAE,IAAI,CAAC,EAAE;gBAClB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,YAAY,EAAE,IAAI,CAAC,KAAK,EAAE;gBAC1B,QAAQ,EAAE,EAAE;gBACZ,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;aACvC;SACD,CAAC;IACH,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,MAAc;QAClC,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QAC1C,IAAI,CAAC,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC,MAAM,KAAK,iBAAiB;YAAE,MAAM,IAAI,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IAC5F,CAAC;IAED,KAAK,CAAC,OAAO,CAAC,MAAc;QAC3B,IAAI,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC;IACrB,CAAC;CACD;AAED,MAAM,OAAO,+BAA+B;IAK3C,YAAY,EAAU,EAAE,SAAyE;QAHxF,qBAAgB,GAAG,UAAmB,CAAC;QAI/C,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC;QACb,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;YAC7C,GAAG,QAAQ;YACX,SAAS,EAAE,EAAE;YACb,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;SACvC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,CAAC,IAAI;QACT,OAAO,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC;IAC5B,CAAC;IAED,KAAK,CAAC,cAAc,CAAC,MAAc,IAAkB,CAAC;IAEtD,KAAK,CAAC,OAAO,CAAC,MAAc,IAAkB,CAAC;CAC/C;AAED,MAAM,UAAU,oBAAoB,CAAC,IAIpC;IACA,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkF,CAAC;IAClH,MAAM,iBAAiB,GAAG,CAAC,KAA6E,EAAQ,EAAE;QACjH,MAAM,GAAG,GAAG,GAAG,KAAK,CAAC,SAAS,IAAI,GAAG,KAAK,KAAK,CAAC,KAAK,EAAE,CAAC;QACxD,MAAM,QAAQ,GAAG,eAAe,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAC1C,IAAI,CAAC,QAAQ,IAAI,KAAK,CAAC,KAAK,GAAG,QAAQ,CAAC,KAAK;YAAE,eAAe,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;IAChF,CAAC,CAAC;IACF,iBAAiB,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAChC,KAAK,MAAM,WAAW,IAAI,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,EAAE,EAAE,CAAC;QAC3D,iBAAiB,CAAC;YACjB,GAAG,WAAW;YACd,SAAS,EACR,WAAW,CAAC,SAAS,IAAI,CAAC,WAAW,CAAC,KAAK,KAAK,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;SACzG,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,oBAAoB,GAAG,CAC5B,QAA8B,EAC9B,SAAgD,EACtC,EAAE,CACZ,QAAQ,CAAC,KAAK,KAAK,SAAS,CAAC,KAAK;QAClC,CAAC,SAAS,CAAC,SAAS,KAAK,SAAS,IAAI,QAAQ,CAAC,SAAS,KAAK,SAAS,CAAC,SAAS,CAAC,CAAC;IACnF,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAG,GAAG,QAAQ,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;IACxF,MAAM,YAAY,GAAG,CAAC,GAAG,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,SAAS,EAAE,EAAE;QAC5E,OAAO,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,oBAAoB,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,GAAG,SAAS,CAAC,KAAK,CAAC;IACnH,CAAC,EAAE,CAAC,CAAC,CAAC;IACN,MAAM,aAAa,GAAG,aAAa,GAAG,YAAY,CAAC;IACnD,IAAI,aAAa,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACvC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,EAAE,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;IACpF,CAAC;IAED,IAAI,eAAe,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;QACnC,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,aAAa,EAAE,KAAK,EAAE,EAAE,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;IAC7G,CAAC;IAED,IAAI,cAAc,GAAG,aAAa,CAAC;IACnC,MAAM,KAAK,GAA2B,EAAE,CAAC;IACzC,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS;SAC/B,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,oBAAoB,CAAC,QAAQ,EAAE,SAAS,CAAC,CAAC,CAC3G;SACA,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CACZ,CAAC,QAAQ,CAAC,MAAM;QAChB,CAAC,CACA,QAAQ,CAAC,KAAK,KAAK,IAAI,CAAC,OAAO,CAAC,cAAc;YAC9C,CAAC,IAAI,CAAC,OAAO,CAAC,kBAAkB,KAAK,SAAS,IAAI,QAAQ,CAAC,SAAS,KAAK,IAAI,CAAC,OAAO,CAAC,kBAAkB,CAAC,CACzG,CACF;SACA,MAAM,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,gBAAgB,KAAK,UAAU,CAAC;SAC9D,MAAM,CACN,CAAC,QAAQ,EAAE,EAAE,CACZ,IAAI,CAAC,OAAO,CAAC,UAAU,KAAK,SAAS;QACrC,IAAI,CAAC,OAAO,CAAC,KAAK,GAAG,QAAQ,CAAC,YAAY,IAAI,IAAI,CAAC,OAAO,CAAC,UAAU,CACtE;SACA,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC,QAAQ,IAAI,IAAI,CAAC,YAAY,GAAG,KAAK,CAAC,YAAY,CAAC,CAAC;IAClG,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,cAAc,IAAI,SAAS,CAAC,KAAK,CAAC;QAClC,IAAI,cAAc,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;YACxC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,EAAE,WAAW,EAAE,IAAI,CAAC,WAAW,EAAE,CAAC;QAChF,CAAC;IACF,CAAC;IACD,OAAO;QACN,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,+BAA+B;QACvC,KAAK;QACL,aAAa;QACb,WAAW,EAAE,IAAI,CAAC,WAAW;KAC7B,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,OAA2B;IACnD,MAAM,QAAQ,GAAG,OAAO,CAAC,kBAAkB,IAAI,OAAO,CAAC,UAAU,IAAI,CAAC,CAAC;IACvE,OAAO,CAAC,OAAO,CAAC,eAAe,IAAI,EAAE,CAAC,CAAC,IAAI,CAC1C,CAAC,MAAM,EAAE,EAAE,CACV,MAAM,CAAC,OAAO,KAAK,OAAO,CAAC,KAAK;QAChC,CAAC,OAAO,CAAC,SAAS,KAAK,SAAS;YAC/B,MAAM,CAAC,gBAAgB,KAAK,SAAS;YACrC,MAAM,CAAC,gBAAgB,KAAK,OAAO,CAAC,SAAS,CAAC;QAC/C,OAAO,CAAC,KAAK,GAAG,MAAM,CAAC,IAAI,IAAI,QAAQ,CACxC,CAAC;AACH,CAAC","sourcesContent":["import type { OllamaRuntime, TransformersRuntime } from \"./local-runtime.ts\";\n\nexport type ResidencyControl = \"full\" | \"keep-alive\" | \"advisory\";\nexport type RuntimeRole = \"active\" | \"router\" | \"bench\" | \"probe\";\n\nexport interface RuntimeResidentModel {\n\tadapterId: string;\n\tmodel: string;\n\tbytes: number;\n\tlastUsedAtMs: number;\n\tpriority: number;\n\tpinned?: boolean;\n\tresidencyControl: ResidencyControl;\n}\n\nexport interface RuntimeLoadRequest {\n\tmodel: string;\n\tbytes: number;\n\trole: RuntimeRole;\n\tpriority: number;\n\tnowMs: number;\n\t/** Refuse an immediate reload of a model that was just evicted for the current resident. */\n\tantiThrashWindowMs?: number;\n\t/** Protect every newly-used resident from eviction for this duration. */\n\tminDwellMs?: number;\n\tpinActiveModel?: string;\n\t/** Adapter identity for this request; set by RuntimeResidencyArbiter before planning. */\n\tadapterId?: string;\n\t/** When pinActiveModel is set, constrain that pin to this adapter when present. */\n\tpinActiveAdapterId?: string;\n\treservations?: RuntimeReservation[];\n\trecentEvictions?: RuntimeEvictionRecord[];\n\t/** False performs admission/eviction only; the real model request owns cold loading. Default true. */\n\tloadModel?: boolean;\n}\n\nexport interface RuntimeReservation {\n\tmodel: string;\n\tbytes: number;\n\tpriority: number;\n\t/** Optional runtime identity; omitted means any adapter serving this model. */\n\tadapterId?: string;\n}\n\nexport interface RuntimeEvictionRecord {\n\tevicted: string;\n\tloaded: string;\n\tatMs: number;\n\t/** Optional identities preserve anti-thrash separation across runtime adapters. */\n\tevictedAdapterId?: string;\n\tloadedAdapterId?: string;\n}\n\nexport type RuntimeResidencyPlan =\n\t| { status: \"fits\"; evict: RuntimeResidentModel[]; requiredBytes: number; budgetBytes: number }\n\t| { status: \"refuse\"; reason: string; evict: RuntimeResidentModel[]; requiredBytes: number; budgetBytes: number };\n\nexport interface RuntimeResidencyAdapter {\n\tid: string;\n\tresidencyControl: ResidencyControl;\n\tlist(): Promise<RuntimeResidentModel[]>;\n\tensureResident(model: string): Promise<void>;\n\trelease(model: string): Promise<void>;\n}\n\nexport class RuntimeResidencyArbiter {\n\tprivate readonly budgetBytes: number;\n\tprivate readonly adapters: Map<string, RuntimeResidencyAdapter>;\n\n\tconstructor(args: { budgetBytes: number; adapters: RuntimeResidencyAdapter[] }) {\n\t\tthis.budgetBytes = args.budgetBytes;\n\t\tthis.adapters = new Map(args.adapters.map((adapter) => [adapter.id, adapter]));\n\t}\n\n\tasync ensureResident(adapterId: string, request: RuntimeLoadRequest): Promise<RuntimeResidencyPlan> {\n\t\tconst adapter = this.adapters.get(adapterId);\n\t\tconst residents = (await Promise.all([...this.adapters.values()].map((entry) => entry.list()))).flat();\n\t\tif (!adapter) {\n\t\t\treturn {\n\t\t\t\tstatus: \"refuse\",\n\t\t\t\treason: \"unknown-adapter\",\n\t\t\t\tevict: [],\n\t\t\t\trequiredBytes: request.bytes,\n\t\t\t\tbudgetBytes: this.budgetBytes,\n\t\t\t};\n\t\t}\n\t\tconst requestForAdapter: RuntimeLoadRequest = {\n\t\t\t...request,\n\t\t\tadapterId,\n\t\t\t...(request.pinActiveModel !== undefined && request.pinActiveAdapterId === undefined\n\t\t\t\t? { pinActiveAdapterId: adapterId }\n\t\t\t\t: {}),\n\t\t};\n\t\tconst plan = planRuntimeResidency({ budgetBytes: this.budgetBytes, residents, request: requestForAdapter });\n\t\tif (plan.status !== \"fits\") return plan;\n\t\tfor (const resident of plan.evict) {\n\t\t\tawait this.adapters.get(resident.adapterId)?.release(resident.model);\n\t\t}\n\t\tconst alreadyResident = residents.some(\n\t\t\t(resident) => resident.adapterId === adapterId && resident.model === request.model,\n\t\t);\n\t\tif (request.loadModel !== false && !alreadyResident) await adapter.ensureResident(request.model);\n\t\treturn plan;\n\t}\n}\n\nexport class OllamaRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"keep-alive\" as const;\n\tprivate readonly runtime: Pick<OllamaRuntime, \"listResidentModels\" | \"ensureResident\" | \"releaseResident\">;\n\tprivate readonly nowMs: () => number;\n\n\tconstructor(\n\t\tid: string,\n\t\truntime: Pick<OllamaRuntime, \"listResidentModels\" | \"ensureResident\" | \"releaseResident\">,\n\t\toptions: { nowMs?: () => number } = {},\n\t) {\n\t\tthis.id = id;\n\t\tthis.runtime = runtime;\n\t\tthis.nowMs = options.nowMs ?? Date.now;\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\tconst now = this.nowMs();\n\t\treturn (await this.runtime.listResidentModels()).map((model) => ({\n\t\t\tadapterId: this.id,\n\t\t\tmodel: model.name,\n\t\t\tbytes: model.sizeBytes,\n\t\t\tlastUsedAtMs: now,\n\t\t\tpriority: 10,\n\t\t\tresidencyControl: this.residencyControl,\n\t\t}));\n\t}\n\n\tasync ensureResident(model: string): Promise<void> {\n\t\tconst result = await this.runtime.ensureResident(model);\n\t\tif (!result.ok) throw new Error(result.error);\n\t}\n\n\tasync release(model: string): Promise<void> {\n\t\tawait this.runtime.releaseResident(model);\n\t}\n}\n\nexport class TransformersRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"full\" as const;\n\tprivate readonly runtime: Pick<TransformersRuntime, \"detect\" | \"start\" | \"stop\">;\n\tprivate readonly model: string;\n\tprivate readonly bytes: number;\n\tprivate readonly nowMs: () => number;\n\n\tconstructor(\n\t\tid: string,\n\t\truntime: Pick<TransformersRuntime, \"detect\" | \"start\" | \"stop\">,\n\t\tmodel: string,\n\t\tbytes: number,\n\t\toptions: { nowMs?: () => number } = {},\n\t) {\n\t\tthis.id = id;\n\t\tthis.runtime = runtime;\n\t\tthis.model = model;\n\t\tthis.bytes = bytes;\n\t\tthis.nowMs = options.nowMs ?? Date.now;\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\tconst status = await this.runtime.detect();\n\t\tif (!status.serverUp) return [];\n\t\treturn [\n\t\t\t{\n\t\t\t\tadapterId: this.id,\n\t\t\t\tmodel: this.model,\n\t\t\t\tbytes: this.bytes,\n\t\t\t\tlastUsedAtMs: this.nowMs(),\n\t\t\t\tpriority: 10,\n\t\t\t\tresidencyControl: this.residencyControl,\n\t\t\t},\n\t\t];\n\t}\n\n\tasync ensureResident(_model: string): Promise<void> {\n\t\tconst result = await this.runtime.start();\n\t\tif (!result.started && result.reason !== \"already_running\") throw new Error(result.reason);\n\t}\n\n\tasync release(_model: string): Promise<void> {\n\t\tthis.runtime.stop();\n\t}\n}\n\nexport class AdvisoryRuntimeResidencyAdapter implements RuntimeResidencyAdapter {\n\treadonly id: string;\n\treadonly residencyControl = \"advisory\" as const;\n\tprivate readonly residents: RuntimeResidentModel[];\n\n\tconstructor(id: string, residents: Omit<RuntimeResidentModel, \"adapterId\" | \"residencyControl\">[]) {\n\t\tthis.id = id;\n\t\tthis.residents = residents.map((resident) => ({\n\t\t\t...resident,\n\t\t\tadapterId: id,\n\t\t\tresidencyControl: this.residencyControl,\n\t\t}));\n\t}\n\n\tasync list(): Promise<RuntimeResidentModel[]> {\n\t\treturn [...this.residents];\n\t}\n\n\tasync ensureResident(_model: string): Promise<void> {}\n\n\tasync release(_model: string): Promise<void> {}\n}\n\nexport function planRuntimeResidency(args: {\n\tbudgetBytes: number;\n\tresidents: readonly RuntimeResidentModel[];\n\trequest: RuntimeLoadRequest;\n}): RuntimeResidencyPlan {\n\tconst requestedModels = new Map<string, { model: string; adapterId?: string; bytes: number; priority: number }>();\n\tconst addRequestedModel = (entry: { model: string; adapterId?: string; bytes: number; priority: number }): void => {\n\t\tconst key = `${entry.adapterId ?? \"*\"}\\0${entry.model}`;\n\t\tconst existing = requestedModels.get(key);\n\t\tif (!existing || entry.bytes > existing.bytes) requestedModels.set(key, entry);\n\t};\n\taddRequestedModel(args.request);\n\tfor (const reservation of args.request.reservations ?? []) {\n\t\taddRequestedModel({\n\t\t\t...reservation,\n\t\t\tadapterId:\n\t\t\t\treservation.adapterId ?? (reservation.model === args.request.model ? args.request.adapterId : undefined),\n\t\t});\n\t}\n\n\tconst isResidentForRequest = (\n\t\tresident: RuntimeResidentModel,\n\t\trequested: { model: string; adapterId?: string },\n\t): boolean =>\n\t\tresident.model === requested.model &&\n\t\t(requested.adapterId === undefined || resident.adapterId === requested.adapterId);\n\tconst residentBytes = args.residents.reduce((sum, resident) => sum + resident.bytes, 0);\n\tconst missingBytes = [...requestedModels.values()].reduce((sum, requested) => {\n\t\treturn args.residents.some((resident) => isResidentForRequest(resident, requested)) ? sum : sum + requested.bytes;\n\t}, 0);\n\tconst requiredBytes = residentBytes + missingBytes;\n\tif (requiredBytes <= args.budgetBytes) {\n\t\treturn { status: \"fits\", evict: [], requiredBytes, budgetBytes: args.budgetBytes };\n\t}\n\n\tif (hasPingPongRisk(args.request)) {\n\t\treturn { status: \"refuse\", reason: \"anti-thrash\", evict: [], requiredBytes, budgetBytes: args.budgetBytes };\n\t}\n\n\tlet projectedBytes = requiredBytes;\n\tconst evict: RuntimeResidentModel[] = [];\n\tconst candidates = args.residents\n\t\t.filter(\n\t\t\t(resident) => ![...requestedModels.values()].some((requested) => isResidentForRequest(resident, requested)),\n\t\t)\n\t\t.filter(\n\t\t\t(resident) =>\n\t\t\t\t!resident.pinned &&\n\t\t\t\t!(\n\t\t\t\t\tresident.model === args.request.pinActiveModel &&\n\t\t\t\t\t(args.request.pinActiveAdapterId === undefined || resident.adapterId === args.request.pinActiveAdapterId)\n\t\t\t\t),\n\t\t)\n\t\t.filter((resident) => resident.residencyControl !== \"advisory\")\n\t\t.filter(\n\t\t\t(resident) =>\n\t\t\t\targs.request.minDwellMs === undefined ||\n\t\t\t\targs.request.nowMs - resident.lastUsedAtMs >= args.request.minDwellMs,\n\t\t)\n\t\t.sort((left, right) => left.priority - right.priority || left.lastUsedAtMs - right.lastUsedAtMs);\n\tfor (const candidate of candidates) {\n\t\tevict.push(candidate);\n\t\tprojectedBytes -= candidate.bytes;\n\t\tif (projectedBytes <= args.budgetBytes) {\n\t\t\treturn { status: \"fits\", evict, requiredBytes, budgetBytes: args.budgetBytes };\n\t\t}\n\t}\n\treturn {\n\t\tstatus: \"refuse\",\n\t\treason: \"insufficient-evictable-memory\",\n\t\tevict,\n\t\trequiredBytes,\n\t\tbudgetBytes: args.budgetBytes,\n\t};\n}\n\nfunction hasPingPongRisk(request: RuntimeLoadRequest): boolean {\n\tconst windowMs = request.antiThrashWindowMs ?? request.minDwellMs ?? 0;\n\treturn (request.recentEvictions ?? []).some(\n\t\t(record) =>\n\t\t\trecord.evicted === request.model &&\n\t\t\t(request.adapterId === undefined ||\n\t\t\t\trecord.evictedAdapterId === undefined ||\n\t\t\t\trecord.evictedAdapterId === request.adapterId) &&\n\t\t\trequest.nowMs - record.atMs <= windowMs,\n\t);\n}\n"]}

package/dist/core/research/research-lane-controller.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"research-lane-controller.d.ts","sourceRoot":"","sources":["../../../src/core/research/research-lane-controller.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,gCAAgC,CAAC;AACrE,OAAO,EAA6B,KAAK,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAC1E,OAAO,KAAK,EACX,iBAAiB,EACjB,yBAAyB,EACzB,wBAAwB,EACxB,sBAAsB,EACtB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,KAAK,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAGnF,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,6BAA6B,CAAC;AAG/D,OAAO,EAAkC,KAAK,sBAAsB,EAAE,MAAM,iCAAiC,CAAC;AAC9G,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAC;AAGxD,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAmB,KAAK,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAEnF,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,0BAA0B,CAAC;AAExE,MAAM,WAAW,0BAA0B;IAC1C,UAAU,IAAI,OAAO,CAAC;IACtB,cAAc,IAAI,OAAO,CAAC;IAC1B,YAAY,IAAI,MAAM,CAAC;IACvB,MAAM,IAAI,MAAM,CAAC;IACjB,WAAW,IAAI,MAAM,CAAC;IACtB,iBAAiB,IAAI,cAAc,CAAC;IACpC,kBAAkB,IAAI,eAAe,CAAC;IACtC,qBAAqB,IAAI,kBAAkB,GAAG,SAAS,CAAC;IACxD,IAAI,CAAC,KAAK,EAAE,iBAAiB,GAAG,IAAI,CAAC;IACrC,qBAAqB,CAAC,KAAK,EAAE,sBAAsB,GAAG,IAAI,CAAC;IAC3D,oBAAoB,IAAI,SAAS,GAAG,SAAS,CAAC;IAC9C,yBAAyB,IAAI,cAAc,GAAG,SAAS,CAAC;IACxD,0BAA0B,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAAC;IAC3D,eAAe,CACd,KAAK,EAAE,KAAK,EACZ,IAAI,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GAClE,MAAM,GAAG,SAAS,CAAC;IACtB,qBAAqB,CAAC,IAAI,EAAE,yBAAyB,GAAG,OAAO,CAAC,wBAAwB,CAAC,CAAC;IAC1F,uBAAuB,EAAE,OAAO,uBAAuB,CAAC;CACxD;AAED,6FAA6F;AAC7F,qBAAa,sBAAsB;IAClC,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,eAAe,CAAqB;IAC5C,OAAO,CAAC,cAAc,CAAS;IAC/B,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAyB;IACzD,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAqB;IAC7D,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA6B;IAClD,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAc;IACpC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAoB;IAE3C,YAAY,IAAI,EAAE,0BAA0B,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,iBAAiB,EAI1F;IAED,WAAW,IAAI,IAAI,CAMlB;IAED,iBAAiB,IAAI,MAAM,GAAG,SAAS,CAEtC;IAED,KAAK,IAAI,IAAI,CAGZ;IAED,UAAU,IAAI,IAAI,CAKjB;IAED,gBAAgB,IAAI,IAAI,CAsCvB;IAEK,OAAO,CAAC,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,sBAAsB,CAAC,CA8H9G;IAED,OAAO,CAAC,WAAW;YA2BL,YAAY;IAa1B,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,qBAAqB;IAW7B,OAAO,CAAC,qBAAqB;CAc7B"}
+{"version":3,"file":"research-lane-controller.d.ts","sourceRoot":"","sources":["../../../src/core/research/research-lane-controller.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,gCAAgC,CAAC;AACrE,OAAO,EAA6B,KAAK,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAC1E,OAAO,KAAK,EACX,iBAAiB,EACjB,yBAAyB,EACzB,wBAAwB,EACxB,sBAAsB,EACtB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,KAAK,EAAE,kBAAkB,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAGnF,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,6BAA6B,CAAC;AAG/D,OAAO,EAAkC,KAAK,sBAAsB,EAAE,MAAM,iCAAiC,CAAC;AAC9G,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,wBAAwB,CAAC;AAGxD,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAC9D,OAAO,EAAmB,KAAK,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAEnF,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,0BAA0B,CAAC;AAExE,MAAM,WAAW,0BAA0B;IAC1C,UAAU,IAAI,OAAO,CAAC;IACtB,cAAc,IAAI,OAAO,CAAC;IAC1B,YAAY,IAAI,MAAM,CAAC;IACvB,MAAM,IAAI,MAAM,CAAC;IACjB,WAAW,IAAI,MAAM,CAAC;IACtB,iBAAiB,IAAI,cAAc,CAAC;IACpC,kBAAkB,IAAI,eAAe,CAAC;IACtC,qBAAqB,IAAI,kBAAkB,GAAG,SAAS,CAAC;IACxD,IAAI,CAAC,KAAK,EAAE,iBAAiB,GAAG,IAAI,CAAC;IACrC,qBAAqB,CAAC,KAAK,EAAE,sBAAsB,GAAG,IAAI,CAAC;IAC3D,oBAAoB,IAAI,SAAS,GAAG,SAAS,CAAC;IAC9C,yBAAyB,IAAI,cAAc,GAAG,SAAS,CAAC;IACxD,0BAA0B,CAAC,MAAM,EAAE,cAAc,GAAG,MAAM,CAAC;IAC3D,eAAe,CACd,KAAK,EAAE,KAAK,EACZ,IAAI,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GAClE,MAAM,GAAG,SAAS,CAAC;IACtB,qBAAqB,CAAC,IAAI,EAAE,yBAAyB,GAAG,OAAO,CAAC,wBAAwB,CAAC,CAAC;IAC1F,uBAAuB,EAAE,OAAO,uBAAuB,CAAC;CACxD;AAED,6FAA6F;AAC7F,qBAAa,sBAAsB;IAClC,OAAO,CAAC,MAAM,CAA4C;IAC1D,OAAO,CAAC,UAAU,CAAS;IAC3B,OAAO,CAAC,eAAe,CAAqB;IAC5C,OAAO,CAAC,cAAc,CAAS;IAC/B,OAAO,CAAC,kBAAkB,CAAK;IAC/B,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAyB;IACzD,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAAqB;IAC7D,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA6B;IAClD,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAc;IACpC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAoB;IAE3C,YAAY,IAAI,EAAE,0BAA0B,EAAE,KAAK,EAAE,WAAW,EAAE,MAAM,EAAE,iBAAiB,EAI1F;IAED,WAAW,IAAI,IAAI,CAMlB;IAED,iBAAiB,IAAI,MAAM,GAAG,SAAS,CAEtC;IAED,KAAK,IAAI,IAAI,CAGZ;IAED,UAAU,IAAI,IAAI,CAKjB;IAED,gBAAgB,IAAI,IAAI,CAsCvB;IAEK,OAAO,CAAC,OAAO,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,sBAAsB,CAAC,CAsI9G;IAED,OAAO,CAAC,WAAW;YA2BL,YAAY;IAa1B,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,qBAAqB;IAW7B,OAAO,CAAC,qBAAqB;CAc7B"}

package/dist/core/research/research-lane-controller.js

@@ -107,6 +107,7 @@ export class ResearchLaneController {
         const startedRecord = this.lanes.start({ type: "research", goalId: demand.goalId });
         this._persistedRunCount++;
         const deregisterInFlight = registerInFlightWork(this.deps.getAgentDir(), "lane", `research:${startedRecord.laneId}`);
+        let toolSurface;
         try {
             let spentUsage;
             const workspaceSources = await this.deps.collectWorkspaceSources({
@@ -115,17 +116,18 @@ export class ResearchLaneController {
                 maxSources: settings.maxSources,
             });
             const maxUsd = clampLaneMaxUsd(settings.maxUsd, this.deps.getCapabilityEnvelope()?.maxEstimatedUsd);
-            const toolSurface = createLaneToolSurface({
+            const activeToolSurface = createLaneToolSurface({
                 cwd: this.deps.getCwd(),
                 profile: laneProfile,
                 deniedPaths: getPrivateLaneDeniedPaths(this.deps.getCwd(), this.deps.getAgentDir()),
             });
-            this.warnUnboundToolGrants(laneProfile, toolSurface);
+            toolSurface = activeToolSurface;
+            this.warnUnboundToolGrants(laneProfile, activeToolSurface);
             const result = await runResearch({
                 query: demand.query,
                 context: demand.context,
                 sources: workspaceSources,
-                envelope: this.buildEnvelope(maxUsd, laneProfile, toolSurface),
+                envelope: this.buildEnvelope(maxUsd, laneProfile, activeToolSurface),
                 maxUsd,
                 maxSources: settings.maxSources,
                 maxFindings: settings.maxFindings,
@@ -142,9 +144,9 @@ export class ResearchLaneController {
                         model,
                         thinkingLevel: resolveModelThinkingLevel(model, laneProfile?.thinking),
                         maxTokens: laneCapability.laneMaxOutputTokens,
-                        tools: toolSurface.tools,
+                        tools: activeToolSurface.tools,
                         maxTurns: 6,
-                        beforeToolCall: toolSurface.beforeToolCall,
+                        beforeToolCall: activeToolSurface.beforeToolCall,
                         signal,
                         cacheRetention: "short",
                         laneKind: "research",
@@ -199,6 +201,13 @@ export class ResearchLaneController {
             return { started: true, record };
         }
         finally {
+            try {
+                await toolSurface?.dispose();
+            }
+            catch (error) {
+                const message = error instanceof Error ? error.message : String(error);
+                this.deps.emit({ type: "warning", message: `Research lane payload cleanup failed: ${message}` });
+            }
             this._isRunning = false;
             deregisterInFlight();
         }