@warlock.js/ai 4.8.2 → 4.9.1

package/esm/rag/loaders/loader.type.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"loader.type.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/loaders/loader.type.ts"],"mappings":";;;;;;AAgBA;;;;AAAyC;AASzC;;;;;;KATY,eAAA,GAAkB,WAAW;;;;;;AAwB5B;AAIb;KAnBY,iBAAA;;;AAmBa;AAQzB;EAtBE,MAAA;EAEA,KAAA,WA0BA;EAxBA,MAAA,GAAS,aAAa,EA6BX;EA3BX,IAAA,WA6BI;EA3BJ,SAAA,WA+BU;EA7BV,WAAA;AAAA;;KAIU,aAAA;AA4BZ;;;;AAA8C;AAG9C;AAHA,KApBY,gBAAA;;;AA6BH;AAST;;EAhCE,EAAA;EAsCuB;;;;EAjCvB,QAAA,GAAW,MAAM,mBAiCM;EA/BvB,IAAA;AAAA;;KAIU,eAAA,GAAkB,gBAAgB;;KAGlC,eAAA,GAAkB,gBAAgB;;KAGlC,cAAA,GAAiB,gBAAgB;;;;;;EAM3C,OAAO;AAAA;;;;;;;KASG,cAAA,GAAiB,gBAAA;;;;;;EAM3B,MAAA,GAAS,cAAc;AAAA"}
+{"version":3,"file":"loader.type.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/loaders/loader.type.ts"],"mappings":";;;;;;AAgBA;;;;AAAyC;AASzC;;;;;;KATY,eAAA,GAAkB,WAAW;;;;;;AAwB5B;AAIb;KAnBY,iBAAA;;;AAmBa;AAQzB;EAtBE,MAAA;EAEA,KAAA,WA0BA;EAxBA,MAAA,GAAS,aAAa,EA6BX;EA3BX,IAAA,WA6BI;EA3BJ,SAAA,WA+BU;EA7BV,WAAA;AAAA;;KAIU,aAAA;AA4BZ;;;;AAA8C;AAG9C;AAHA,KApBY,gBAAA;;;AA6BH;AAST;;EAhCE,EAAA;EAsCuB;;;;EAjCvB,QAAA,GAAW,MAAM,mBAiCM;EA/BvB,IAAA;AAAA;;KAIU,eAAA,GAAkB,gBAAgB;;KAGlC,eAAA,GAAkB,gBAAgB;;KAGlC,cAAA,GAAiB,gBAAgB;;;;;;EAM3C,OAAO;AAAA;;;;;;;KASG,cAAA,GAAiB,gBAAA;;;;;;EAM3B,MAAA,GAAS,cAAc;AAAA"}

package/esm/rag/rag.d.mts

@@ -1,6 +1,6 @@
 import { Rag, RagConfig } from "./contracts/rag-config.type.mjs";
 
-//#region ../@warlock.js/ai/src/rag/rag.d.ts
+//#region ../ai/src/rag/rag.d.ts
 /**
  * Create a RAG pipeline: **chunk → embed → vector store → retrieve →
  * rerank → cite**, reusing the app's `ai.embedder` for embedding, a

package/esm/rag/rag.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"rag.d.mts","names":[],"sources":["../../../../../../../@warlock.js/ai/src/rag/rag.ts"],"mappings":";;;;;AAyDA;;;;;;;;AAA2C;;;;;;;;;;;;;;;;;;;;;iBAA3B,GAAA,CAAI,MAAA,EAAQ,SAAA,GAAY,GAAG"}
+{"version":3,"file":"rag.d.mts","names":[],"sources":["../../../../../../../ai/src/rag/rag.ts"],"mappings":";;;;;AAyDA;;;;;;;;AAA2C;;;;;;;;;;;;;;;;;;;;;iBAA3B,GAAA,CAAI,MAAA,EAAQ,SAAA,GAAY,GAAG"}

package/esm/rag/rag.mjs

@@ -4,7 +4,7 @@ import { ragAsTool } from "./as-tool.mjs";
 import { retrieve } from "./retrieve.mjs";
 import { cacheVectorStore } from "./store/cache-vector-store.mjs";
 
-//#region ../@warlock.js/ai/src/rag/rag.ts
+//#region ../ai/src/rag/rag.ts
 const DEFAULT_NAME = "rag";
 const DEFAULT_NAMESPACE_PREFIX = "ai.rag";
 /**

package/esm/rag/rag.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"rag.mjs","names":["chunkText","runRetrieve"],"sources":["../../../../../../../@warlock.js/ai/src/rag/rag.ts"],"sourcesContent":["import { resolveDefaultStore } from \"../config\";\nimport { chunk as chunkText } from \"./chunk/chunk\";\nimport type { ChunkOptions } from \"./contracts/chunk-options.type\";\nimport type { RetrieveOptions, RetrieveResult } from \"./contracts/citation.type\";\nimport type {\n  Rag,\n  RagAsToolOptions,\n  RagConfig,\n} from \"./contracts/rag-config.type\";\nimport type { RagDocument } from \"./contracts/rag-document.type\";\nimport { ragAsTool } from \"./as-tool\";\nimport { retrieve as runRetrieve, type StoredChunk } from \"./retrieve\";\nimport { cacheVectorStore } from \"./store/cache-vector-store\";\nimport type { VectorStore } from \"./store/vector-store.contract\";\n\nconst DEFAULT_NAME = \"rag\";\nconst DEFAULT_NAMESPACE_PREFIX = \"ai.rag\";\n\n/**\n * Max chunk texts embedded per `embedder.embedMany()` call. One call is\n * one provider request, so a giant document is sub-batched to stay under\n * the provider's per-request token cap (the design's \"chunk larger than\n * provider per-request cap\" guard).\n */\nconst DEFAULT_MAX_BATCH = 96;\n\n/**\n * Create a RAG pipeline: **chunk → embed → vector store → retrieve →\n * rerank → cite**, reusing the app's `ai.embedder` for embedding, a\n * `@warlock.js/cache` `CacheDriver` as the vector store, and the\n * composite-as-tool engine to expose retrieval as a tool.\n *\n * Resolution is loud at construction (mirroring `memory()`):\n * - `embedder` is required — a provider with no embedder must be caught\n *   here, not at first index.\n * - `store` falls back to `ai.config({ defaultStore })`; if neither\n *   resolves, construction throws.\n *\n * `retrieve()` is return-only — it never auto-injects into a prompt; the\n * caller formats the cited chunks (or uses `asTool()` for the agent loop).\n * The reranker is OFF by default (cosine-only) unless `config.reranker`\n * is set.\n *\n * @example\n * import { ai } from \"@warlock.js/ai\";\n * import { MemoryCacheDriver } from \"@warlock.js/cache\";\n *\n * const kb = ai.rag({\n *   name: \"docs\",\n *   embedder: openai.embedder({ name: \"text-embedding-3-small\" }),\n *   store: new MemoryCacheDriver(),\n *   chunk: { type: \"markdown\", size: 800, overlap: 120 },\n * });\n *\n * await kb.index([{ id: \"guide\", text: longMarkdown, metadata: { url: \"/guide\" } }]);\n * const { chunks } = await kb.retrieve(\"how do I configure caching?\", { topK: 4 });\n */\nexport function rag(config: RagConfig): Rag {\n  const name = config.name ?? DEFAULT_NAME;\n\n  if (!config.embedder) {\n    throw new Error(\n      `rag(\"${name}\"): an \\`embedder\\` is required — pass one from a provider that supports embeddings (e.g. openai.embedder({ name: \"text-embedding-3-small\" }))`,\n    );\n  }\n\n  const driver = config.store ?? resolveDefaultStore();\n\n  if (!driver) {\n    throw new Error(\n      `rag(\"${name}\"): no store — pass \\`store\\` (a vector-capable @warlock.js/cache CacheDriver) or call \\`ai.config({ defaultStore })\\` at app boot before constructing the rag`,\n    );\n  }\n\n  const store: VectorStore = cacheVectorStore(driver);\n  const namespace = config.namespace ?? `${DEFAULT_NAMESPACE_PREFIX}.${name}`;\n  const embedder = config.embedder;\n\n  // Captured at first index for the dimension-mismatch guard in retrieve().\n  let indexedDimensions: number | undefined;\n\n  const instance: Rag = {\n    name,\n\n    async index(\n      docs: RagDocument[],\n      chunkOverride?: ChunkOptions,\n    ): Promise<{ chunks: number }> {\n      const chunkOptions = chunkOverride ?? config.chunk;\n\n      // Ingestion guardrails (D5) — fail BEFORE any embedding spend.\n      const limits = config.limits;\n      if (limits?.maxDocuments !== undefined && docs.length > limits.maxDocuments) {\n        throw new Error(\n          `rag(\"${name}\"): index() received ${docs.length} documents, exceeding the configured maxDocuments of ${limits.maxDocuments}`,\n        );\n      }\n      if (limits?.maxBytes !== undefined) {\n        const totalBytes = docs.reduce(\n          (sum, doc) => sum + Buffer.byteLength(doc.text ?? \"\"),\n          0,\n        );\n        if (totalBytes > limits.maxBytes) {\n          throw new Error(\n            `rag(\"${name}\"): index() received ${totalBytes} bytes of document text, exceeding the configured maxBytes of ${limits.maxBytes}`,\n          );\n        }\n      }\n\n      // Flatten every document into stored-chunk records + their texts,\n      // preserving document order so a single batched embed maps back 1:1.\n      const records: { key: string; value: StoredChunk; text: string; tags?: string[] }[] = [];\n\n      for (const doc of docs) {\n        const pieces = chunkText(doc.text, chunkOptions);\n\n        for (const piece of pieces) {\n          const value: StoredChunk = {\n            sourceId: doc.id,\n            chunkIndex: piece.index,\n            span: piece.span,\n            text: piece.text,\n            metadata: doc.metadata,\n          };\n\n          records.push({\n            key: keyFor(namespace, doc.id, piece.index),\n            value,\n            text: piece.text,\n            tags: doc.tags,\n          });\n        }\n      }\n\n      // Empty / whitespace-only documents yield zero chunks — write\n      // nothing and never embed an empty batch.\n      if (records.length === 0) {\n        return { chunks: 0 };\n      }\n\n      // Chunk cap (D5) — checked after chunking, still before embedding.\n      if (limits?.maxChunks !== undefined && records.length > limits.maxChunks) {\n        throw new Error(\n          `rag(\"${name}\"): index() produced ${records.length} chunks, exceeding the configured maxChunks of ${limits.maxChunks}`,\n        );\n      }\n\n      // Sub-batch the embed calls so one giant document does not blow the\n      // provider's per-request token cap.\n      for (let offset = 0; offset < records.length; offset += DEFAULT_MAX_BATCH) {\n        const batch = records.slice(offset, offset + DEFAULT_MAX_BATCH);\n        const { vectors, dimensions } = await embedder.embedMany(\n          batch.map((record) => record.text),\n        );\n\n        if (indexedDimensions === undefined && dimensions !== 0) {\n          indexedDimensions = dimensions;\n        }\n\n        await Promise.all(\n          batch.map((record, position) =>\n            store.upsert(record.key, record.value, vectors[position], record.tags),\n          ),\n        );\n      }\n\n      return { chunks: records.length };\n    },\n\n    async retrieve(query: string, options?: RetrieveOptions): Promise<RetrieveResult> {\n      return runRetrieve(\n        query,\n        {\n          embedder,\n          store,\n          namespace,\n          reranker: config.reranker,\n          defaults: config.retrieve,\n          indexedDimensions,\n        },\n        options,\n      );\n    },\n\n    async clear(): Promise<void> {\n      await store.removeNamespace(namespace);\n    },\n\n    asTool(options?: RagAsToolOptions) {\n      return ragAsTool(name, (query, retrieveOptions) => instance.retrieve(query, retrieveOptions), options);\n    },\n  };\n\n  return instance;\n}\n\n/**\n * Namespaced key for a stored chunk. Uses the `.` separator (matching\n * `SemanticMemory.keyFor`) so namespace-prefix filtering on the returned\n * `hit.key` stays aligned with the cache's `parseKey` normalization.\n */\nfunction keyFor(namespace: string, sourceId: string, chunkIndex: number): string {\n  return `${namespace}.${sourceId}.${chunkIndex}`;\n}\n"],"mappings":";;;;;;;AAeA,MAAM,eAAe;AACrB,MAAM,2BAA2B;;;;;;;AAQjC,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiC1B,SAAgB,IAAI,QAAwB;CAC1C,MAAM,OAAO,OAAO,QAAQ;CAE5B,IAAI,CAAC,OAAO,UACV,MAAM,IAAI,MACR,QAAQ,KAAK,+IACf;CAGF,MAAM,SAAS,OAAO,SAAS,oBAAoB;CAEnD,IAAI,CAAC,QACH,MAAM,IAAI,MACR,QAAQ,KAAK,+JACf;CAGF,MAAM,QAAqB,iBAAiB,MAAM;CAClD,MAAM,YAAY,OAAO,aAAa,GAAG,yBAAyB,GAAG;CACrE,MAAM,WAAW,OAAO;CAGxB,IAAI;CAEJ,MAAM,WAAgB;EACpB;EAEA,MAAM,MACJ,MACA,eAC6B;GAC7B,MAAM,eAAe,iBAAiB,OAAO;GAG7C,MAAM,SAAS,OAAO;GACtB,IAAI,QAAQ,iBAAiB,UAAa,KAAK,SAAS,OAAO,cAC7D,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,KAAK,OAAO,uDAAuD,OAAO,cAChH;GAEF,IAAI,QAAQ,aAAa,QAAW;IAClC,MAAM,aAAa,KAAK,QACrB,KAAK,QAAQ,MAAM,OAAO,WAAW,IAAI,QAAQ,EAAE,GACpD,CACF;IACA,IAAI,aAAa,OAAO,UACtB,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,WAAW,gEAAgE,OAAO,UACxH;GAEJ;GAIA,MAAM,UAAgF,CAAC;GAEvF,KAAK,MAAM,OAAO,MAAM;IACtB,MAAM,SAASA,MAAU,IAAI,MAAM,YAAY;IAE/C,KAAK,MAAM,SAAS,QAAQ;KAC1B,MAAM,QAAqB;MACzB,UAAU,IAAI;MACd,YAAY,MAAM;MAClB,MAAM,MAAM;MACZ,MAAM,MAAM;MACZ,UAAU,IAAI;KAChB;KAEA,QAAQ,KAAK;MACX,KAAK,OAAO,WAAW,IAAI,IAAI,MAAM,KAAK;MAC1C;MACA,MAAM,MAAM;MACZ,MAAM,IAAI;KACZ,CAAC;IACH;GACF;GAIA,IAAI,QAAQ,WAAW,GACrB,OAAO,EAAE,QAAQ,EAAE;GAIrB,IAAI,QAAQ,cAAc,UAAa,QAAQ,SAAS,OAAO,WAC7D,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,QAAQ,OAAO,iDAAiD,OAAO,WAC7G;GAKF,KAAK,IAAI,SAAS,GAAG,SAAS,QAAQ,QAAQ,UAAU,mBAAmB;IACzE,MAAM,QAAQ,QAAQ,MAAM,QAAQ,SAAS,iBAAiB;IAC9D,MAAM,EAAE,SAAS,eAAe,MAAM,SAAS,UAC7C,MAAM,KAAK,WAAW,OAAO,IAAI,CACnC;IAEA,IAAI,sBAAsB,UAAa,eAAe,GACpD,oBAAoB;IAGtB,MAAM,QAAQ,IACZ,MAAM,KAAK,QAAQ,aACjB,MAAM,OAAO,OAAO,KAAK,OAAO,OAAO,QAAQ,WAAW,OAAO,IAAI,CACvE,CACF;GACF;GAEA,OAAO,EAAE,QAAQ,QAAQ,OAAO;EAClC;EAEA,MAAM,SAAS,OAAe,SAAoD;GAChF,OAAOC,SACL,OACA;IACE;IACA;IACA;IACA,UAAU,OAAO;IACjB,UAAU,OAAO;IACjB;GACF,GACA,OACF;EACF;EAEA,MAAM,QAAuB;GAC3B,MAAM,MAAM,gBAAgB,SAAS;EACvC;EAEA,OAAO,SAA4B;GACjC,OAAO,UAAU,OAAO,OAAO,oBAAoB,SAAS,SAAS,OAAO,eAAe,GAAG,OAAO;EACvG;CACF;CAEA,OAAO;AACT;;;;;;AAOA,SAAS,OAAO,WAAmB,UAAkB,YAA4B;CAC/E,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG;AACrC"}
+{"version":3,"file":"rag.mjs","names":["chunkText","runRetrieve"],"sources":["../../../../../../../ai/src/rag/rag.ts"],"sourcesContent":["import { resolveDefaultStore } from \"../config\";\nimport { chunk as chunkText } from \"./chunk/chunk\";\nimport type { ChunkOptions } from \"./contracts/chunk-options.type\";\nimport type { RetrieveOptions, RetrieveResult } from \"./contracts/citation.type\";\nimport type {\n  Rag,\n  RagAsToolOptions,\n  RagConfig,\n} from \"./contracts/rag-config.type\";\nimport type { RagDocument } from \"./contracts/rag-document.type\";\nimport { ragAsTool } from \"./as-tool\";\nimport { retrieve as runRetrieve, type StoredChunk } from \"./retrieve\";\nimport { cacheVectorStore } from \"./store/cache-vector-store\";\nimport type { VectorStore } from \"./store/vector-store.contract\";\n\nconst DEFAULT_NAME = \"rag\";\nconst DEFAULT_NAMESPACE_PREFIX = \"ai.rag\";\n\n/**\n * Max chunk texts embedded per `embedder.embedMany()` call. One call is\n * one provider request, so a giant document is sub-batched to stay under\n * the provider's per-request token cap (the design's \"chunk larger than\n * provider per-request cap\" guard).\n */\nconst DEFAULT_MAX_BATCH = 96;\n\n/**\n * Create a RAG pipeline: **chunk → embed → vector store → retrieve →\n * rerank → cite**, reusing the app's `ai.embedder` for embedding, a\n * `@warlock.js/cache` `CacheDriver` as the vector store, and the\n * composite-as-tool engine to expose retrieval as a tool.\n *\n * Resolution is loud at construction (mirroring `memory()`):\n * - `embedder` is required — a provider with no embedder must be caught\n *   here, not at first index.\n * - `store` falls back to `ai.config({ defaultStore })`; if neither\n *   resolves, construction throws.\n *\n * `retrieve()` is return-only — it never auto-injects into a prompt; the\n * caller formats the cited chunks (or uses `asTool()` for the agent loop).\n * The reranker is OFF by default (cosine-only) unless `config.reranker`\n * is set.\n *\n * @example\n * import { ai } from \"@warlock.js/ai\";\n * import { MemoryCacheDriver } from \"@warlock.js/cache\";\n *\n * const kb = ai.rag({\n *   name: \"docs\",\n *   embedder: openai.embedder({ name: \"text-embedding-3-small\" }),\n *   store: new MemoryCacheDriver(),\n *   chunk: { type: \"markdown\", size: 800, overlap: 120 },\n * });\n *\n * await kb.index([{ id: \"guide\", text: longMarkdown, metadata: { url: \"/guide\" } }]);\n * const { chunks } = await kb.retrieve(\"how do I configure caching?\", { topK: 4 });\n */\nexport function rag(config: RagConfig): Rag {\n  const name = config.name ?? DEFAULT_NAME;\n\n  if (!config.embedder) {\n    throw new Error(\n      `rag(\"${name}\"): an \\`embedder\\` is required — pass one from a provider that supports embeddings (e.g. openai.embedder({ name: \"text-embedding-3-small\" }))`,\n    );\n  }\n\n  const driver = config.store ?? resolveDefaultStore();\n\n  if (!driver) {\n    throw new Error(\n      `rag(\"${name}\"): no store — pass \\`store\\` (a vector-capable @warlock.js/cache CacheDriver) or call \\`ai.config({ defaultStore })\\` at app boot before constructing the rag`,\n    );\n  }\n\n  const store: VectorStore = cacheVectorStore(driver);\n  const namespace = config.namespace ?? `${DEFAULT_NAMESPACE_PREFIX}.${name}`;\n  const embedder = config.embedder;\n\n  // Captured at first index for the dimension-mismatch guard in retrieve().\n  let indexedDimensions: number | undefined;\n\n  const instance: Rag = {\n    name,\n\n    async index(\n      docs: RagDocument[],\n      chunkOverride?: ChunkOptions,\n    ): Promise<{ chunks: number }> {\n      const chunkOptions = chunkOverride ?? config.chunk;\n\n      // Ingestion guardrails (D5) — fail BEFORE any embedding spend.\n      const limits = config.limits;\n      if (limits?.maxDocuments !== undefined && docs.length > limits.maxDocuments) {\n        throw new Error(\n          `rag(\"${name}\"): index() received ${docs.length} documents, exceeding the configured maxDocuments of ${limits.maxDocuments}`,\n        );\n      }\n      if (limits?.maxBytes !== undefined) {\n        const totalBytes = docs.reduce(\n          (sum, doc) => sum + Buffer.byteLength(doc.text ?? \"\"),\n          0,\n        );\n        if (totalBytes > limits.maxBytes) {\n          throw new Error(\n            `rag(\"${name}\"): index() received ${totalBytes} bytes of document text, exceeding the configured maxBytes of ${limits.maxBytes}`,\n          );\n        }\n      }\n\n      // Flatten every document into stored-chunk records + their texts,\n      // preserving document order so a single batched embed maps back 1:1.\n      const records: { key: string; value: StoredChunk; text: string; tags?: string[] }[] = [];\n\n      for (const doc of docs) {\n        const pieces = chunkText(doc.text, chunkOptions);\n\n        for (const piece of pieces) {\n          const value: StoredChunk = {\n            sourceId: doc.id,\n            chunkIndex: piece.index,\n            span: piece.span,\n            text: piece.text,\n            metadata: doc.metadata,\n          };\n\n          records.push({\n            key: keyFor(namespace, doc.id, piece.index),\n            value,\n            text: piece.text,\n            tags: doc.tags,\n          });\n        }\n      }\n\n      // Empty / whitespace-only documents yield zero chunks — write\n      // nothing and never embed an empty batch.\n      if (records.length === 0) {\n        return { chunks: 0 };\n      }\n\n      // Chunk cap (D5) — checked after chunking, still before embedding.\n      if (limits?.maxChunks !== undefined && records.length > limits.maxChunks) {\n        throw new Error(\n          `rag(\"${name}\"): index() produced ${records.length} chunks, exceeding the configured maxChunks of ${limits.maxChunks}`,\n        );\n      }\n\n      // Sub-batch the embed calls so one giant document does not blow the\n      // provider's per-request token cap.\n      for (let offset = 0; offset < records.length; offset += DEFAULT_MAX_BATCH) {\n        const batch = records.slice(offset, offset + DEFAULT_MAX_BATCH);\n        const { vectors, dimensions } = await embedder.embedMany(\n          batch.map((record) => record.text),\n        );\n\n        if (indexedDimensions === undefined && dimensions !== 0) {\n          indexedDimensions = dimensions;\n        }\n\n        await Promise.all(\n          batch.map((record, position) =>\n            store.upsert(record.key, record.value, vectors[position], record.tags),\n          ),\n        );\n      }\n\n      return { chunks: records.length };\n    },\n\n    async retrieve(query: string, options?: RetrieveOptions): Promise<RetrieveResult> {\n      return runRetrieve(\n        query,\n        {\n          embedder,\n          store,\n          namespace,\n          reranker: config.reranker,\n          defaults: config.retrieve,\n          indexedDimensions,\n        },\n        options,\n      );\n    },\n\n    async clear(): Promise<void> {\n      await store.removeNamespace(namespace);\n    },\n\n    asTool(options?: RagAsToolOptions) {\n      return ragAsTool(name, (query, retrieveOptions) => instance.retrieve(query, retrieveOptions), options);\n    },\n  };\n\n  return instance;\n}\n\n/**\n * Namespaced key for a stored chunk. Uses the `.` separator (matching\n * `SemanticMemory.keyFor`) so namespace-prefix filtering on the returned\n * `hit.key` stays aligned with the cache's `parseKey` normalization.\n */\nfunction keyFor(namespace: string, sourceId: string, chunkIndex: number): string {\n  return `${namespace}.${sourceId}.${chunkIndex}`;\n}\n"],"mappings":";;;;;;;AAeA,MAAM,eAAe;AACrB,MAAM,2BAA2B;;;;;;;AAQjC,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAiC1B,SAAgB,IAAI,QAAwB;CAC1C,MAAM,OAAO,OAAO,QAAQ;CAE5B,IAAI,CAAC,OAAO,UACV,MAAM,IAAI,MACR,QAAQ,KAAK,+IACf;CAGF,MAAM,SAAS,OAAO,SAAS,oBAAoB;CAEnD,IAAI,CAAC,QACH,MAAM,IAAI,MACR,QAAQ,KAAK,+JACf;CAGF,MAAM,QAAqB,iBAAiB,MAAM;CAClD,MAAM,YAAY,OAAO,aAAa,GAAG,yBAAyB,GAAG;CACrE,MAAM,WAAW,OAAO;CAGxB,IAAI;CAEJ,MAAM,WAAgB;EACpB;EAEA,MAAM,MACJ,MACA,eAC6B;GAC7B,MAAM,eAAe,iBAAiB,OAAO;GAG7C,MAAM,SAAS,OAAO;GACtB,IAAI,QAAQ,iBAAiB,UAAa,KAAK,SAAS,OAAO,cAC7D,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,KAAK,OAAO,uDAAuD,OAAO,cAChH;GAEF,IAAI,QAAQ,aAAa,QAAW;IAClC,MAAM,aAAa,KAAK,QACrB,KAAK,QAAQ,MAAM,OAAO,WAAW,IAAI,QAAQ,EAAE,GACpD,CACF;IACA,IAAI,aAAa,OAAO,UACtB,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,WAAW,gEAAgE,OAAO,UACxH;GAEJ;GAIA,MAAM,UAAgF,CAAC;GAEvF,KAAK,MAAM,OAAO,MAAM;IACtB,MAAM,SAASA,MAAU,IAAI,MAAM,YAAY;IAE/C,KAAK,MAAM,SAAS,QAAQ;KAC1B,MAAM,QAAqB;MACzB,UAAU,IAAI;MACd,YAAY,MAAM;MAClB,MAAM,MAAM;MACZ,MAAM,MAAM;MACZ,UAAU,IAAI;KAChB;KAEA,QAAQ,KAAK;MACX,KAAK,OAAO,WAAW,IAAI,IAAI,MAAM,KAAK;MAC1C;MACA,MAAM,MAAM;MACZ,MAAM,IAAI;KACZ,CAAC;IACH;GACF;GAIA,IAAI,QAAQ,WAAW,GACrB,OAAO,EAAE,QAAQ,EAAE;GAIrB,IAAI,QAAQ,cAAc,UAAa,QAAQ,SAAS,OAAO,WAC7D,MAAM,IAAI,MACR,QAAQ,KAAK,uBAAuB,QAAQ,OAAO,iDAAiD,OAAO,WAC7G;GAKF,KAAK,IAAI,SAAS,GAAG,SAAS,QAAQ,QAAQ,UAAU,mBAAmB;IACzE,MAAM,QAAQ,QAAQ,MAAM,QAAQ,SAAS,iBAAiB;IAC9D,MAAM,EAAE,SAAS,eAAe,MAAM,SAAS,UAC7C,MAAM,KAAK,WAAW,OAAO,IAAI,CACnC;IAEA,IAAI,sBAAsB,UAAa,eAAe,GACpD,oBAAoB;IAGtB,MAAM,QAAQ,IACZ,MAAM,KAAK,QAAQ,aACjB,MAAM,OAAO,OAAO,KAAK,OAAO,OAAO,QAAQ,WAAW,OAAO,IAAI,CACvE,CACF;GACF;GAEA,OAAO,EAAE,QAAQ,QAAQ,OAAO;EAClC;EAEA,MAAM,SAAS,OAAe,SAAoD;GAChF,OAAOC,SACL,OACA;IACE;IACA;IACA;IACA,UAAU,OAAO;IACjB,UAAU,OAAO;IACjB;GACF,GACA,OACF;EACF;EAEA,MAAM,QAAuB;GAC3B,MAAM,MAAM,gBAAgB,SAAS;EACvC;EAEA,OAAO,SAA4B;GACjC,OAAO,UAAU,OAAO,OAAO,oBAAoB,SAAS,SAAS,OAAO,eAAe,GAAG,OAAO;EACvG;CACF;CAEA,OAAO;AACT;;;;;;AAOA,SAAS,OAAO,WAAmB,UAAkB,YAA4B;CAC/E,OAAO,GAAG,UAAU,GAAG,SAAS,GAAG;AACrC"}

package/esm/rag/rerank/keyword-reranker.d.mts

@@ -1,6 +1,6 @@
 import { RagReranker } from "./reranker.contract.mjs";
 
-//#region ../@warlock.js/ai/src/rag/rerank/keyword-reranker.d.ts
+//#region ../ai/src/rag/rerank/keyword-reranker.d.ts
 /** Options for the {@link keywordReranker}. */
 type KeywordRerankerOptions = {
   /**

package/esm/rag/rerank/keyword-reranker.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"keyword-reranker.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/rerank/keyword-reranker.ts"],"mappings":";;;;KAIY,sBAAA;EAAA;;;;AAMJ;EAAN,MAAM;AAAA;;;;;;;AA2B0E;;;;;;;;;;iBAAlE,eAAA,CAAgB,OAAA,GAAS,sBAAA,GAA8B,WAAW"}
+{"version":3,"file":"keyword-reranker.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/rerank/keyword-reranker.ts"],"mappings":";;;;KAIY,sBAAA;EAAA;;;;AAMJ;EAAN,MAAM;AAAA;;;;;;;AA2B0E;;;;;;;;;;iBAAlE,eAAA,CAAgB,OAAA,GAAS,sBAAA,GAA8B,WAAW"}

package/esm/rag/rerank/keyword-reranker.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/rerank/keyword-reranker.ts
+//#region ../ai/src/rag/rerank/keyword-reranker.ts
 /** Splits text into lowercase alphanumeric terms. */
 function tokenize(text) {
 	return text.toLowerCase().split(/[^a-z0-9]+/).filter((term) => term.length > 0);

package/esm/rag/rerank/keyword-reranker.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"keyword-reranker.mjs","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/rerank/keyword-reranker.ts"],"sourcesContent":["import type { RetrievedChunk } from \"../contracts/citation.type\";\nimport type { RagReranker } from \"./reranker.contract\";\n\n/** Options for the {@link keywordReranker}. */\nexport type KeywordRerankerOptions = {\n  /**\n   * Weight of the lexical-overlap signal blended with the original cosine\n   * score, in `[0, 1]`. `1` ranks purely by keyword overlap; `0` keeps the\n   * cosine order. Default `0.5`.\n   */\n  weight?: number;\n};\n\n/** Splits text into lowercase alphanumeric terms. */\nfunction tokenize(text: string): string[] {\n  return text\n    .toLowerCase()\n    .split(/[^a-z0-9]+/)\n    .filter((term) => term.length > 0);\n}\n\n/**\n * Zero-dependency lexical reranker (a BM25-lite, IDF-free keyword overlap).\n *\n * For each candidate it computes the fraction of distinct query terms that\n * appear in the chunk, blends that with the candidate's original cosine\n * score by `weight`, and sorts descending. A pure-lexical pass costs\n * nothing beyond string splits — no peer, no model — so it is the\n * recommended opt-in reranker when an embedding-only ranking surfaces a\n * keyword-rich chunk too low.\n *\n * Ties (equal blended score) preserve the incoming order, so the cosine\n * ranking breaks ties deterministically.\n *\n * @example\n * const kb = ai.rag({ embedder, store, reranker: ai.rag.keywordReranker() });\n */\nexport function keywordReranker(options: KeywordRerankerOptions = {}): RagReranker {\n  const weight = options.weight ?? 0.5;\n\n  return {\n    name: \"keyword\",\n    async rerank(query: string, candidates: RetrievedChunk[]): Promise<RetrievedChunk[]> {\n      if (candidates.length === 0) {\n        return [];\n      }\n\n      const queryTerms = new Set(tokenize(query));\n\n      if (queryTerms.size === 0) {\n        return [...candidates];\n      }\n\n      const scored = candidates.map((candidate, position) => {\n        const chunkTerms = new Set(tokenize(candidate.text));\n\n        let overlap = 0;\n        for (const term of queryTerms) {\n          if (chunkTerms.has(term)) {\n            overlap += 1;\n          }\n        }\n\n        const lexical = overlap / queryTerms.size;\n        const blended = weight * lexical + (1 - weight) * candidate.score;\n\n        return { candidate, blended, position };\n      });\n\n      scored.sort((first, second) => {\n        if (second.blended !== first.blended) {\n          return second.blended - first.blended;\n        }\n\n        // Stable on ties: keep the incoming (cosine) order.\n        return first.position - second.position;\n      });\n\n      return scored.map((entry) => ({\n        ...entry.candidate,\n        score: entry.blended,\n        citation: { ...entry.candidate.citation, score: entry.blended },\n      }));\n    },\n  };\n}\n"],"mappings":";;AAcA,SAAS,SAAS,MAAwB;CACxC,OAAO,KACJ,YAAY,CAAC,CACb,MAAM,YAAY,CAAC,CACnB,QAAQ,SAAS,KAAK,SAAS,CAAC;AACrC;;;;;;;;;;;;;;;;;AAkBA,SAAgB,gBAAgB,UAAkC,CAAC,GAAgB;CACjF,MAAM,SAAS,QAAQ,UAAU;CAEjC,OAAO;EACL,MAAM;EACN,MAAM,OAAO,OAAe,YAAyD;GACnF,IAAI,WAAW,WAAW,GACxB,OAAO,CAAC;GAGV,MAAM,aAAa,IAAI,IAAI,SAAS,KAAK,CAAC;GAE1C,IAAI,WAAW,SAAS,GACtB,OAAO,CAAC,GAAG,UAAU;GAGvB,MAAM,SAAS,WAAW,KAAK,WAAW,aAAa;IACrD,MAAM,aAAa,IAAI,IAAI,SAAS,UAAU,IAAI,CAAC;IAEnD,IAAI,UAAU;IACd,KAAK,MAAM,QAAQ,YACjB,IAAI,WAAW,IAAI,IAAI,GACrB,WAAW;IAOf,OAAO;KAAE;KAAW,SAFJ,UADA,UAAU,WAAW,SACD,IAAI,UAAU,UAAU;KAE/B;IAAS;GACxC,CAAC;GAED,OAAO,MAAM,OAAO,WAAW;IAC7B,IAAI,OAAO,YAAY,MAAM,SAC3B,OAAO,OAAO,UAAU,MAAM;IAIhC,OAAO,MAAM,WAAW,OAAO;GACjC,CAAC;GAED,OAAO,OAAO,KAAK,WAAW;IAC5B,GAAG,MAAM;IACT,OAAO,MAAM;IACb,UAAU;KAAE,GAAG,MAAM,UAAU;KAAU,OAAO,MAAM;IAAQ;GAChE,EAAE;EACJ;CACF;AACF"}
+{"version":3,"file":"keyword-reranker.mjs","names":[],"sources":["../../../../../../../../ai/src/rag/rerank/keyword-reranker.ts"],"sourcesContent":["import type { RetrievedChunk } from \"../contracts/citation.type\";\nimport type { RagReranker } from \"./reranker.contract\";\n\n/** Options for the {@link keywordReranker}. */\nexport type KeywordRerankerOptions = {\n  /**\n   * Weight of the lexical-overlap signal blended with the original cosine\n   * score, in `[0, 1]`. `1` ranks purely by keyword overlap; `0` keeps the\n   * cosine order. Default `0.5`.\n   */\n  weight?: number;\n};\n\n/** Splits text into lowercase alphanumeric terms. */\nfunction tokenize(text: string): string[] {\n  return text\n    .toLowerCase()\n    .split(/[^a-z0-9]+/)\n    .filter((term) => term.length > 0);\n}\n\n/**\n * Zero-dependency lexical reranker (a BM25-lite, IDF-free keyword overlap).\n *\n * For each candidate it computes the fraction of distinct query terms that\n * appear in the chunk, blends that with the candidate's original cosine\n * score by `weight`, and sorts descending. A pure-lexical pass costs\n * nothing beyond string splits — no peer, no model — so it is the\n * recommended opt-in reranker when an embedding-only ranking surfaces a\n * keyword-rich chunk too low.\n *\n * Ties (equal blended score) preserve the incoming order, so the cosine\n * ranking breaks ties deterministically.\n *\n * @example\n * const kb = ai.rag({ embedder, store, reranker: ai.rag.keywordReranker() });\n */\nexport function keywordReranker(options: KeywordRerankerOptions = {}): RagReranker {\n  const weight = options.weight ?? 0.5;\n\n  return {\n    name: \"keyword\",\n    async rerank(query: string, candidates: RetrievedChunk[]): Promise<RetrievedChunk[]> {\n      if (candidates.length === 0) {\n        return [];\n      }\n\n      const queryTerms = new Set(tokenize(query));\n\n      if (queryTerms.size === 0) {\n        return [...candidates];\n      }\n\n      const scored = candidates.map((candidate, position) => {\n        const chunkTerms = new Set(tokenize(candidate.text));\n\n        let overlap = 0;\n        for (const term of queryTerms) {\n          if (chunkTerms.has(term)) {\n            overlap += 1;\n          }\n        }\n\n        const lexical = overlap / queryTerms.size;\n        const blended = weight * lexical + (1 - weight) * candidate.score;\n\n        return { candidate, blended, position };\n      });\n\n      scored.sort((first, second) => {\n        if (second.blended !== first.blended) {\n          return second.blended - first.blended;\n        }\n\n        // Stable on ties: keep the incoming (cosine) order.\n        return first.position - second.position;\n      });\n\n      return scored.map((entry) => ({\n        ...entry.candidate,\n        score: entry.blended,\n        citation: { ...entry.candidate.citation, score: entry.blended },\n      }));\n    },\n  };\n}\n"],"mappings":";;AAcA,SAAS,SAAS,MAAwB;CACxC,OAAO,KACJ,YAAY,CAAC,CACb,MAAM,YAAY,CAAC,CACnB,QAAQ,SAAS,KAAK,SAAS,CAAC;AACrC;;;;;;;;;;;;;;;;;AAkBA,SAAgB,gBAAgB,UAAkC,CAAC,GAAgB;CACjF,MAAM,SAAS,QAAQ,UAAU;CAEjC,OAAO;EACL,MAAM;EACN,MAAM,OAAO,OAAe,YAAyD;GACnF,IAAI,WAAW,WAAW,GACxB,OAAO,CAAC;GAGV,MAAM,aAAa,IAAI,IAAI,SAAS,KAAK,CAAC;GAE1C,IAAI,WAAW,SAAS,GACtB,OAAO,CAAC,GAAG,UAAU;GAGvB,MAAM,SAAS,WAAW,KAAK,WAAW,aAAa;IACrD,MAAM,aAAa,IAAI,IAAI,SAAS,UAAU,IAAI,CAAC;IAEnD,IAAI,UAAU;IACd,KAAK,MAAM,QAAQ,YACjB,IAAI,WAAW,IAAI,IAAI,GACrB,WAAW;IAOf,OAAO;KAAE;KAAW,SAFJ,UADA,UAAU,WAAW,SACD,IAAI,UAAU,UAAU;KAE/B;IAAS;GACxC,CAAC;GAED,OAAO,MAAM,OAAO,WAAW;IAC7B,IAAI,OAAO,YAAY,MAAM,SAC3B,OAAO,OAAO,UAAU,MAAM;IAIhC,OAAO,MAAM,WAAW,OAAO;GACjC,CAAC;GAED,OAAO,OAAO,KAAK,WAAW;IAC5B,GAAG,MAAM;IACT,OAAO,MAAM;IACb,UAAU;KAAE,GAAG,MAAM,UAAU;KAAU,OAAO,MAAM;IAAQ;GAChE,EAAE;EACJ;CACF;AACF"}

package/esm/rag/rerank/llm-reranker.d.mts

@@ -1,7 +1,7 @@
 import { ModelContract } from "../../contracts/model.contract.mjs";
 import { RagReranker } from "./reranker.contract.mjs";
 
-//#region ../@warlock.js/ai/src/rag/rerank/llm-reranker.d.ts
+//#region ../ai/src/rag/rerank/llm-reranker.d.ts
 /** Options for the {@link llmReranker}. */
 type LlmRerankerOptions = {
   /** The model used to score candidate relevance. Required. */model: ModelContract;

package/esm/rag/rerank/llm-reranker.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"llm-reranker.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/rerank/llm-reranker.ts"],"mappings":";;;;;KAMY,kBAAA;EAAA,6DAEV,KAAA,EAAO,aAAa;;;;;EAKpB,SAAA;AAAA;AAAS;AA8FX;;;;;;;;AAAqE;;;;;;;;;;AA9F1D,iBA8FK,WAAA,CAAY,OAAA,EAAS,kBAAA,GAAqB,WAAW"}
+{"version":3,"file":"llm-reranker.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/rerank/llm-reranker.ts"],"mappings":";;;;;KAMY,kBAAA;EAAA,6DAEV,KAAA,EAAO,aAAa;;;;;EAKpB,SAAA;AAAA;AAAS;AA8FX;;;;;;;;AAAqE;;;;;;;;;;AA9F1D,iBA8FK,WAAA,CAAY,OAAA,EAAS,kBAAA,GAAqB,WAAW"}

package/esm/rag/rerank/llm-reranker.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/rerank/llm-reranker.ts
+//#region ../ai/src/rag/rerank/llm-reranker.ts
 /**
 * Build the scoring prompt — the model rates each candidate's relevance to
 * the query on a `0..1` scale and replies with one `index: score` line per

package/esm/rag/rerank/llm-reranker.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"llm-reranker.mjs","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/rerank/llm-reranker.ts"],"sourcesContent":["import type { Message } from \"../../contracts/conversation-message.type\";\nimport type { ModelContract } from \"../../contracts/model.contract\";\nimport type { RetrievedChunk } from \"../contracts/citation.type\";\nimport type { RagReranker } from \"./reranker.contract\";\n\n/** Options for the {@link llmReranker}. */\nexport type LlmRerankerOptions = {\n  /** The model used to score candidate relevance. Required. */\n  model: ModelContract;\n  /**\n   * How many candidates to score per model call. Larger batches mean\n   * fewer round-trips but a longer prompt. Default `10`.\n   */\n  batchSize?: number;\n};\n\n/**\n * A single relevance score the model returns for a candidate, in `[0, 1]`,\n * keyed by the candidate's position in the batch.\n */\ntype ScoreLine = {\n  index: number;\n  score: number;\n};\n\n/**\n * Build the scoring prompt — the model rates each candidate's relevance to\n * the query on a `0..1` scale and replies with one `index: score` line per\n * candidate. Kept terse and JSON-light so any chat model can answer.\n */\nfunction buildPrompt(query: string, candidates: RetrievedChunk[]): Message[] {\n  const lines = candidates\n    .map((candidate, index) => `[${index}] ${candidate.text}`)\n    .join(\"\\n\\n\");\n\n  return [\n    {\n      role: \"system\",\n      content:\n        \"You are a relevance grader. For each numbered passage, rate how well it answers the query on a scale from 0 (irrelevant) to 1 (fully relevant). Reply with ONLY a JSON array of objects like [{\\\"index\\\":0,\\\"score\\\":0.9}], one entry per passage, no prose.\",\n    },\n    {\n      role: \"user\",\n      content: `Query: ${query}\\n\\nPassages:\\n${lines}`,\n    },\n  ];\n}\n\n/**\n * Parse the model's reply into a score map. Tolerant of surrounding prose:\n * extracts the first JSON array and reads `{ index, score }` entries.\n * Returns an empty map when nothing parseable is found, so the caller can\n * fall back to the original order.\n */\nfunction parseScores(reply: string): Map<number, number> {\n  const scores = new Map<number, number>();\n  const start = reply.indexOf(\"[\");\n  const end = reply.lastIndexOf(\"]\");\n\n  if (start === -1 || end === -1 || end <= start) {\n    return scores;\n  }\n\n  let parsed: unknown;\n  try {\n    parsed = JSON.parse(reply.slice(start, end + 1));\n  } catch {\n    return scores;\n  }\n\n  if (!Array.isArray(parsed)) {\n    return scores;\n  }\n\n  for (const entry of parsed as ScoreLine[]) {\n    if (\n      entry &&\n      typeof entry.index === \"number\" &&\n      typeof entry.score === \"number\" &&\n      Number.isFinite(entry.score)\n    ) {\n      scores.set(entry.index, Math.max(0, Math.min(1, entry.score)));\n    }\n  }\n\n  return scores;\n}\n\n/**\n * Optional model-backed reranker.\n *\n * Asks an LLM to grade each over-fetched candidate's relevance to the\n * query on a `0..1` scale, then sorts descending by the model's score.\n * Candidates the model does not score keep their original cosine score, so\n * a partial/garbled reply degrades gracefully rather than dropping hits.\n * Scoring is batched (`batchSize`) to bound prompt length.\n *\n * Unlike {@link keywordReranker}, this costs one or more model calls per\n * retrieval — opt in only when precision matters more than latency/cost.\n *\n * @example\n * const kb = ai.rag({\n *   embedder,\n *   store,\n *   reranker: ai.rag.llmReranker({ model: openai.model({ name: \"gpt-4o-mini\" }) }),\n * });\n */\nexport function llmReranker(options: LlmRerankerOptions): RagReranker {\n  const batchSize = options.batchSize ?? 10;\n\n  return {\n    name: \"llm\",\n    async rerank(query: string, candidates: RetrievedChunk[]): Promise<RetrievedChunk[]> {\n      if (candidates.length === 0) {\n        return [];\n      }\n\n      const rescored: RetrievedChunk[] = [];\n\n      for (let offset = 0; offset < candidates.length; offset += batchSize) {\n        const batch = candidates.slice(offset, offset + batchSize);\n        const response = await options.model.complete(buildPrompt(query, batch));\n        const scores = parseScores(response.content);\n\n        batch.forEach((candidate, index) => {\n          const score = scores.has(index) ? (scores.get(index) as number) : candidate.score;\n\n          rescored.push({\n            ...candidate,\n            score,\n            citation: { ...candidate.citation, score },\n          });\n        });\n      }\n\n      return rescored.sort((first, second) => second.score - first.score);\n    },\n  };\n}\n"],"mappings":";;;;;;AA8BA,SAAS,YAAY,OAAe,YAAyC;CAK3E,OAAO,CACL;EACE,MAAM;EACN,SACE;CACJ,GACA;EACE,MAAM;EACN,SAAS,UAAU,MAAM,iBAZf,WACX,KAAK,WAAW,UAAU,IAAI,MAAM,IAAI,UAAU,MAAM,CAAC,CACzD,KAAK,MAU0C;CAChD,CACF;AACF;;;;;;;AAQA,SAAS,YAAY,OAAoC;CACvD,MAAM,yBAAS,IAAI,IAAoB;CACvC,MAAM,QAAQ,MAAM,QAAQ,GAAG;CAC/B,MAAM,MAAM,MAAM,YAAY,GAAG;CAEjC,IAAI,UAAU,MAAM,QAAQ,MAAM,OAAO,OACvC,OAAO;CAGT,IAAI;CACJ,IAAI;EACF,SAAS,KAAK,MAAM,MAAM,MAAM,OAAO,MAAM,CAAC,CAAC;CACjD,QAAQ;EACN,OAAO;CACT;CAEA,IAAI,CAAC,MAAM,QAAQ,MAAM,GACvB,OAAO;CAGT,KAAK,MAAM,SAAS,QAClB,IACE,SACA,OAAO,MAAM,UAAU,YACvB,OAAO,MAAM,UAAU,YACvB,OAAO,SAAS,MAAM,KAAK,GAE3B,OAAO,IAAI,MAAM,OAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,MAAM,KAAK,CAAC,CAAC;CAIjE,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,YAAY,SAA0C;CACpE,MAAM,YAAY,QAAQ,aAAa;CAEvC,OAAO;EACL,MAAM;EACN,MAAM,OAAO,OAAe,YAAyD;GACnF,IAAI,WAAW,WAAW,GACxB,OAAO,CAAC;GAGV,MAAM,WAA6B,CAAC;GAEpC,KAAK,IAAI,SAAS,GAAG,SAAS,WAAW,QAAQ,UAAU,WAAW;IACpE,MAAM,QAAQ,WAAW,MAAM,QAAQ,SAAS,SAAS;IAEzD,MAAM,SAAS,aAAY,MADJ,QAAQ,MAAM,SAAS,YAAY,OAAO,KAAK,CAAC,EACpC,CAAC,OAAO;IAE3C,MAAM,SAAS,WAAW,UAAU;KAClC,MAAM,QAAQ,OAAO,IAAI,KAAK,IAAK,OAAO,IAAI,KAAK,IAAe,UAAU;KAE5E,SAAS,KAAK;MACZ,GAAG;MACH;MACA,UAAU;OAAE,GAAG,UAAU;OAAU;MAAM;KAC3C,CAAC;IACH,CAAC;GACH;GAEA,OAAO,SAAS,MAAM,OAAO,WAAW,OAAO,QAAQ,MAAM,KAAK;EACpE;CACF;AACF"}
+{"version":3,"file":"llm-reranker.mjs","names":[],"sources":["../../../../../../../../ai/src/rag/rerank/llm-reranker.ts"],"sourcesContent":["import type { Message } from \"../../contracts/conversation-message.type\";\nimport type { ModelContract } from \"../../contracts/model.contract\";\nimport type { RetrievedChunk } from \"../contracts/citation.type\";\nimport type { RagReranker } from \"./reranker.contract\";\n\n/** Options for the {@link llmReranker}. */\nexport type LlmRerankerOptions = {\n  /** The model used to score candidate relevance. Required. */\n  model: ModelContract;\n  /**\n   * How many candidates to score per model call. Larger batches mean\n   * fewer round-trips but a longer prompt. Default `10`.\n   */\n  batchSize?: number;\n};\n\n/**\n * A single relevance score the model returns for a candidate, in `[0, 1]`,\n * keyed by the candidate's position in the batch.\n */\ntype ScoreLine = {\n  index: number;\n  score: number;\n};\n\n/**\n * Build the scoring prompt — the model rates each candidate's relevance to\n * the query on a `0..1` scale and replies with one `index: score` line per\n * candidate. Kept terse and JSON-light so any chat model can answer.\n */\nfunction buildPrompt(query: string, candidates: RetrievedChunk[]): Message[] {\n  const lines = candidates\n    .map((candidate, index) => `[${index}] ${candidate.text}`)\n    .join(\"\\n\\n\");\n\n  return [\n    {\n      role: \"system\",\n      content:\n        \"You are a relevance grader. For each numbered passage, rate how well it answers the query on a scale from 0 (irrelevant) to 1 (fully relevant). Reply with ONLY a JSON array of objects like [{\\\"index\\\":0,\\\"score\\\":0.9}], one entry per passage, no prose.\",\n    },\n    {\n      role: \"user\",\n      content: `Query: ${query}\\n\\nPassages:\\n${lines}`,\n    },\n  ];\n}\n\n/**\n * Parse the model's reply into a score map. Tolerant of surrounding prose:\n * extracts the first JSON array and reads `{ index, score }` entries.\n * Returns an empty map when nothing parseable is found, so the caller can\n * fall back to the original order.\n */\nfunction parseScores(reply: string): Map<number, number> {\n  const scores = new Map<number, number>();\n  const start = reply.indexOf(\"[\");\n  const end = reply.lastIndexOf(\"]\");\n\n  if (start === -1 || end === -1 || end <= start) {\n    return scores;\n  }\n\n  let parsed: unknown;\n  try {\n    parsed = JSON.parse(reply.slice(start, end + 1));\n  } catch {\n    return scores;\n  }\n\n  if (!Array.isArray(parsed)) {\n    return scores;\n  }\n\n  for (const entry of parsed as ScoreLine[]) {\n    if (\n      entry &&\n      typeof entry.index === \"number\" &&\n      typeof entry.score === \"number\" &&\n      Number.isFinite(entry.score)\n    ) {\n      scores.set(entry.index, Math.max(0, Math.min(1, entry.score)));\n    }\n  }\n\n  return scores;\n}\n\n/**\n * Optional model-backed reranker.\n *\n * Asks an LLM to grade each over-fetched candidate's relevance to the\n * query on a `0..1` scale, then sorts descending by the model's score.\n * Candidates the model does not score keep their original cosine score, so\n * a partial/garbled reply degrades gracefully rather than dropping hits.\n * Scoring is batched (`batchSize`) to bound prompt length.\n *\n * Unlike {@link keywordReranker}, this costs one or more model calls per\n * retrieval — opt in only when precision matters more than latency/cost.\n *\n * @example\n * const kb = ai.rag({\n *   embedder,\n *   store,\n *   reranker: ai.rag.llmReranker({ model: openai.model({ name: \"gpt-4o-mini\" }) }),\n * });\n */\nexport function llmReranker(options: LlmRerankerOptions): RagReranker {\n  const batchSize = options.batchSize ?? 10;\n\n  return {\n    name: \"llm\",\n    async rerank(query: string, candidates: RetrievedChunk[]): Promise<RetrievedChunk[]> {\n      if (candidates.length === 0) {\n        return [];\n      }\n\n      const rescored: RetrievedChunk[] = [];\n\n      for (let offset = 0; offset < candidates.length; offset += batchSize) {\n        const batch = candidates.slice(offset, offset + batchSize);\n        const response = await options.model.complete(buildPrompt(query, batch));\n        const scores = parseScores(response.content);\n\n        batch.forEach((candidate, index) => {\n          const score = scores.has(index) ? (scores.get(index) as number) : candidate.score;\n\n          rescored.push({\n            ...candidate,\n            score,\n            citation: { ...candidate.citation, score },\n          });\n        });\n      }\n\n      return rescored.sort((first, second) => second.score - first.score);\n    },\n  };\n}\n"],"mappings":";;;;;;AA8BA,SAAS,YAAY,OAAe,YAAyC;CAK3E,OAAO,CACL;EACE,MAAM;EACN,SACE;CACJ,GACA;EACE,MAAM;EACN,SAAS,UAAU,MAAM,iBAZf,WACX,KAAK,WAAW,UAAU,IAAI,MAAM,IAAI,UAAU,MAAM,CAAC,CACzD,KAAK,MAU0C;CAChD,CACF;AACF;;;;;;;AAQA,SAAS,YAAY,OAAoC;CACvD,MAAM,yBAAS,IAAI,IAAoB;CACvC,MAAM,QAAQ,MAAM,QAAQ,GAAG;CAC/B,MAAM,MAAM,MAAM,YAAY,GAAG;CAEjC,IAAI,UAAU,MAAM,QAAQ,MAAM,OAAO,OACvC,OAAO;CAGT,IAAI;CACJ,IAAI;EACF,SAAS,KAAK,MAAM,MAAM,MAAM,OAAO,MAAM,CAAC,CAAC;CACjD,QAAQ;EACN,OAAO;CACT;CAEA,IAAI,CAAC,MAAM,QAAQ,MAAM,GACvB,OAAO;CAGT,KAAK,MAAM,SAAS,QAClB,IACE,SACA,OAAO,MAAM,UAAU,YACvB,OAAO,MAAM,UAAU,YACvB,OAAO,SAAS,MAAM,KAAK,GAE3B,OAAO,IAAI,MAAM,OAAO,KAAK,IAAI,GAAG,KAAK,IAAI,GAAG,MAAM,KAAK,CAAC,CAAC;CAIjE,OAAO;AACT;;;;;;;;;;;;;;;;;;;;AAqBA,SAAgB,YAAY,SAA0C;CACpE,MAAM,YAAY,QAAQ,aAAa;CAEvC,OAAO;EACL,MAAM;EACN,MAAM,OAAO,OAAe,YAAyD;GACnF,IAAI,WAAW,WAAW,GACxB,OAAO,CAAC;GAGV,MAAM,WAA6B,CAAC;GAEpC,KAAK,IAAI,SAAS,GAAG,SAAS,WAAW,QAAQ,UAAU,WAAW;IACpE,MAAM,QAAQ,WAAW,MAAM,QAAQ,SAAS,SAAS;IAEzD,MAAM,SAAS,aAAY,MADJ,QAAQ,MAAM,SAAS,YAAY,OAAO,KAAK,CAAC,EACpC,CAAC,OAAO;IAE3C,MAAM,SAAS,WAAW,UAAU;KAClC,MAAM,QAAQ,OAAO,IAAI,KAAK,IAAK,OAAO,IAAI,KAAK,IAAe,UAAU;KAE5E,SAAS,KAAK;MACZ,GAAG;MACH;MACA,UAAU;OAAE,GAAG,UAAU;OAAU;MAAM;KAC3C,CAAC;IACH,CAAC;GACH;GAEA,OAAO,SAAS,MAAM,OAAO,WAAW,OAAO,QAAQ,MAAM,KAAK;EACpE;CACF;AACF"}

package/esm/rag/rerank/reranker.contract.d.mts

@@ -1,6 +1,6 @@
 import { RetrievedChunk } from "../contracts/citation.type.mjs";
 
-//#region ../@warlock.js/ai/src/rag/rerank/reranker.contract.d.ts
+//#region ../ai/src/rag/rerank/reranker.contract.d.ts
 /**
  * Reranks an over-fetched candidate set against the query before the
  * pipeline slices to `topK`. Optional — when no reranker is configured the

package/esm/rag/rerank/reranker.contract.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"reranker.contract.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/rerank/reranker.contract.ts"],"mappings":";;;;;AAcA;;;;;;;;;;UAAiB,WAAA;EAQmB;EAAA,SANzB,IAAA;EAM4C;;;AAAsB;;EAA3E,MAAA,CAAO,KAAA,UAAe,UAAA,EAAY,cAAA,KAAmB,OAAA,CAAQ,cAAA;AAAA"}
+{"version":3,"file":"reranker.contract.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/rerank/reranker.contract.ts"],"mappings":";;;;;AAcA;;;;;;;;;;UAAiB,WAAA;EAQmB;EAAA,SANzB,IAAA;EAM4C;;;AAAsB;;EAA3E,MAAA,CAAO,KAAA,UAAe,UAAA,EAAY,cAAA,KAAmB,OAAA,CAAQ,cAAA;AAAA"}

package/esm/rag/retrieve.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/retrieve.ts
+//#region ../ai/src/rag/retrieve.ts
 /** Default number of chunks returned after reranking. */
 const DEFAULT_TOP_K = 5;
 /** Default cosine floor applied at the vector-store stage. */

package/esm/rag/retrieve.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"retrieve.mjs","names":[],"sources":["../../../../../../../@warlock.js/ai/src/rag/retrieve.ts"],"sourcesContent":["import type { EmbedderContract } from \"../contracts/embedder.contract\";\nimport type {\n  Citation,\n  RetrievedChunk,\n  RetrieveOptions,\n  RetrieveResult,\n} from \"./contracts/citation.type\";\nimport type { RagReranker } from \"./rerank/reranker.contract\";\nimport type { VectorStore } from \"./store/vector-store.contract\";\n\n/** Default number of chunks returned after reranking. */\nexport const DEFAULT_TOP_K = 5;\n\n/** Default cosine floor applied at the vector-store stage. */\nexport const DEFAULT_THRESHOLD = 0.5;\n\n/**\n * Shape persisted per chunk in the vector store. The vector itself is held\n * by the driver's own index (passed via `set({ vector })`), so it is not\n * duplicated here.\n */\nexport type StoredChunk = {\n  sourceId: string;\n  chunkIndex: number;\n  span: [start: number, end: number];\n  text: string;\n  metadata?: Record<string, unknown>;\n};\n\n/** Dependencies the retrieve pipeline needs, resolved once by `rag()`. */\nexport type RetrieveDeps = {\n  embedder: EmbedderContract;\n  store: VectorStore;\n  /** Namespace prefix every stored key carries (e.g. `\"ai.rag.docs\"`). */\n  namespace: string;\n  /** Optional reranker; when absent the cosine order is kept. */\n  reranker?: RagReranker;\n  /** Pipeline-level retrieval defaults. */\n  defaults?: RetrieveOptions;\n  /**\n   * Dimension count captured at first index for the mismatch guard. When\n   * set, the query embedder's `dimensions` must equal it.\n   */\n  indexedDimensions?: number;\n};\n\n/**\n * The cite pipeline: embed the query → over-fetch candidates from the\n * store → filter to this rag's namespace → map to {@link RetrievedChunk}s\n * with a {@link Citation} → optionally rerank → slice `topK`.\n *\n * Behavior matches the design's failure modes:\n * - No hits clearing the threshold → `{ query, chunks: [] }`, never throws.\n * - Namespace-prefix filtering keeps two rags sharing one driver isolated.\n * - A reranker that throws is caught; the raw cosine order is used instead.\n * - A dimension mismatch (indexed with model A, queried with model B)\n *   throws a clear error rather than returning garbage hits.\n */\nexport async function retrieve(\n  query: string,\n  deps: RetrieveDeps,\n  options: RetrieveOptions = {},\n): Promise<RetrieveResult> {\n  const topK = options.topK ?? deps.defaults?.topK ?? DEFAULT_TOP_K;\n  const threshold = options.threshold ?? deps.defaults?.threshold ?? DEFAULT_THRESHOLD;\n  const tags = options.tags ?? deps.defaults?.tags;\n  const candidates = options.candidates ?? deps.defaults?.candidates ?? Math.max(topK * 4, topK);\n\n  const { vector, dimensions } = await deps.embedder.embed(query);\n\n  if (\n    deps.indexedDimensions !== undefined &&\n    dimensions !== 0 &&\n    deps.indexedDimensions !== 0 &&\n    dimensions !== deps.indexedDimensions\n  ) {\n    throw new Error(\n      `rag.retrieve(): query embedder dimensions (${dimensions}) do not match the dimensions captured at index time (${deps.indexedDimensions}); index and query must use the same embedding model`,\n    );\n  }\n\n  const hits = await deps.store.query<StoredChunk>(vector, {\n    topK: candidates,\n    threshold,\n    tags,\n  });\n\n  const prefix = `${deps.namespace}.`;\n\n  let retrieved: RetrievedChunk[] = hits\n    .filter((hit) => hit.key.startsWith(prefix))\n    .map((hit) => toRetrievedChunk(hit.value, hit.score));\n\n  retrieved = await applyReranker(query, retrieved, deps.reranker);\n\n  return { query, chunks: retrieved.slice(0, topK) };\n}\n\n/** Build a cited {@link RetrievedChunk} from a stored chunk + its cosine score. */\nfunction toRetrievedChunk(stored: StoredChunk, score: number): RetrievedChunk {\n  const citation: Citation = {\n    sourceId: stored.sourceId,\n    chunkIndex: stored.chunkIndex,\n    span: stored.span,\n    score,\n    metadata: stored.metadata,\n  };\n\n  return { text: stored.text, score, citation };\n}\n\n/**\n * Run the optional reranker, degrading to the raw cosine order if it\n * throws — a flaky optional reranker must never fail the whole retrieval.\n */\nasync function applyReranker(\n  query: string,\n  candidates: RetrievedChunk[],\n  reranker: RagReranker | undefined,\n): Promise<RetrievedChunk[]> {\n  if (!reranker) {\n    return candidates;\n  }\n\n  try {\n    return await reranker.rerank(query, candidates);\n  } catch {\n    // Logged at the call site in a richer build; here we degrade silently\n    // to vector-only ranking rather than aborting the retrieval.\n    return candidates;\n  }\n}\n"],"mappings":";;AAWA,MAAa,gBAAgB;;AAG7B,MAAa,oBAAoB;;;;;;;;;;;;;AA4CjC,eAAsB,SACpB,OACA,MACA,UAA2B,CAAC,GACH;CACzB,MAAM,OAAO,QAAQ,QAAQ,KAAK,UAAU;CAC5C,MAAM,YAAY,QAAQ,aAAa,KAAK,UAAU;CACtD,MAAM,OAAO,QAAQ,QAAQ,KAAK,UAAU;CAC5C,MAAM,aAAa,QAAQ,cAAc,KAAK,UAAU,cAAc,KAAK,IAAI,OAAO,GAAG,IAAI;CAE7F,MAAM,EAAE,QAAQ,eAAe,MAAM,KAAK,SAAS,MAAM,KAAK;CAE9D,IACE,KAAK,sBAAsB,UAC3B,eAAe,KACf,KAAK,sBAAsB,KAC3B,eAAe,KAAK,mBAEpB,MAAM,IAAI,MACR,8CAA8C,WAAW,wDAAwD,KAAK,kBAAkB,qDAC1I;CAGF,MAAM,OAAO,MAAM,KAAK,MAAM,MAAmB,QAAQ;EACvD,MAAM;EACN;EACA;CACF,CAAC;CAED,MAAM,SAAS,GAAG,KAAK,UAAU;CAEjC,IAAI,YAA8B,KAC/B,QAAQ,QAAQ,IAAI,IAAI,WAAW,MAAM,CAAC,CAAC,CAC3C,KAAK,QAAQ,iBAAiB,IAAI,OAAO,IAAI,KAAK,CAAC;CAEtD,YAAY,MAAM,cAAc,OAAO,WAAW,KAAK,QAAQ;CAE/D,OAAO;EAAE;EAAO,QAAQ,UAAU,MAAM,GAAG,IAAI;CAAE;AACnD;;AAGA,SAAS,iBAAiB,QAAqB,OAA+B;CAC5E,MAAM,WAAqB;EACzB,UAAU,OAAO;EACjB,YAAY,OAAO;EACnB,MAAM,OAAO;EACb;EACA,UAAU,OAAO;CACnB;CAEA,OAAO;EAAE,MAAM,OAAO;EAAM;EAAO;CAAS;AAC9C;;;;;AAMA,eAAe,cACb,OACA,YACA,UAC2B;CAC3B,IAAI,CAAC,UACH,OAAO;CAGT,IAAI;EACF,OAAO,MAAM,SAAS,OAAO,OAAO,UAAU;CAChD,QAAQ;EAGN,OAAO;CACT;AACF"}
+{"version":3,"file":"retrieve.mjs","names":[],"sources":["../../../../../../../ai/src/rag/retrieve.ts"],"sourcesContent":["import type { EmbedderContract } from \"../contracts/embedder.contract\";\nimport type {\n  Citation,\n  RetrievedChunk,\n  RetrieveOptions,\n  RetrieveResult,\n} from \"./contracts/citation.type\";\nimport type { RagReranker } from \"./rerank/reranker.contract\";\nimport type { VectorStore } from \"./store/vector-store.contract\";\n\n/** Default number of chunks returned after reranking. */\nexport const DEFAULT_TOP_K = 5;\n\n/** Default cosine floor applied at the vector-store stage. */\nexport const DEFAULT_THRESHOLD = 0.5;\n\n/**\n * Shape persisted per chunk in the vector store. The vector itself is held\n * by the driver's own index (passed via `set({ vector })`), so it is not\n * duplicated here.\n */\nexport type StoredChunk = {\n  sourceId: string;\n  chunkIndex: number;\n  span: [start: number, end: number];\n  text: string;\n  metadata?: Record<string, unknown>;\n};\n\n/** Dependencies the retrieve pipeline needs, resolved once by `rag()`. */\nexport type RetrieveDeps = {\n  embedder: EmbedderContract;\n  store: VectorStore;\n  /** Namespace prefix every stored key carries (e.g. `\"ai.rag.docs\"`). */\n  namespace: string;\n  /** Optional reranker; when absent the cosine order is kept. */\n  reranker?: RagReranker;\n  /** Pipeline-level retrieval defaults. */\n  defaults?: RetrieveOptions;\n  /**\n   * Dimension count captured at first index for the mismatch guard. When\n   * set, the query embedder's `dimensions` must equal it.\n   */\n  indexedDimensions?: number;\n};\n\n/**\n * The cite pipeline: embed the query → over-fetch candidates from the\n * store → filter to this rag's namespace → map to {@link RetrievedChunk}s\n * with a {@link Citation} → optionally rerank → slice `topK`.\n *\n * Behavior matches the design's failure modes:\n * - No hits clearing the threshold → `{ query, chunks: [] }`, never throws.\n * - Namespace-prefix filtering keeps two rags sharing one driver isolated.\n * - A reranker that throws is caught; the raw cosine order is used instead.\n * - A dimension mismatch (indexed with model A, queried with model B)\n *   throws a clear error rather than returning garbage hits.\n */\nexport async function retrieve(\n  query: string,\n  deps: RetrieveDeps,\n  options: RetrieveOptions = {},\n): Promise<RetrieveResult> {\n  const topK = options.topK ?? deps.defaults?.topK ?? DEFAULT_TOP_K;\n  const threshold = options.threshold ?? deps.defaults?.threshold ?? DEFAULT_THRESHOLD;\n  const tags = options.tags ?? deps.defaults?.tags;\n  const candidates = options.candidates ?? deps.defaults?.candidates ?? Math.max(topK * 4, topK);\n\n  const { vector, dimensions } = await deps.embedder.embed(query);\n\n  if (\n    deps.indexedDimensions !== undefined &&\n    dimensions !== 0 &&\n    deps.indexedDimensions !== 0 &&\n    dimensions !== deps.indexedDimensions\n  ) {\n    throw new Error(\n      `rag.retrieve(): query embedder dimensions (${dimensions}) do not match the dimensions captured at index time (${deps.indexedDimensions}); index and query must use the same embedding model`,\n    );\n  }\n\n  const hits = await deps.store.query<StoredChunk>(vector, {\n    topK: candidates,\n    threshold,\n    tags,\n  });\n\n  const prefix = `${deps.namespace}.`;\n\n  let retrieved: RetrievedChunk[] = hits\n    .filter((hit) => hit.key.startsWith(prefix))\n    .map((hit) => toRetrievedChunk(hit.value, hit.score));\n\n  retrieved = await applyReranker(query, retrieved, deps.reranker);\n\n  return { query, chunks: retrieved.slice(0, topK) };\n}\n\n/** Build a cited {@link RetrievedChunk} from a stored chunk + its cosine score. */\nfunction toRetrievedChunk(stored: StoredChunk, score: number): RetrievedChunk {\n  const citation: Citation = {\n    sourceId: stored.sourceId,\n    chunkIndex: stored.chunkIndex,\n    span: stored.span,\n    score,\n    metadata: stored.metadata,\n  };\n\n  return { text: stored.text, score, citation };\n}\n\n/**\n * Run the optional reranker, degrading to the raw cosine order if it\n * throws — a flaky optional reranker must never fail the whole retrieval.\n */\nasync function applyReranker(\n  query: string,\n  candidates: RetrievedChunk[],\n  reranker: RagReranker | undefined,\n): Promise<RetrievedChunk[]> {\n  if (!reranker) {\n    return candidates;\n  }\n\n  try {\n    return await reranker.rerank(query, candidates);\n  } catch {\n    // Logged at the call site in a richer build; here we degrade silently\n    // to vector-only ranking rather than aborting the retrieval.\n    return candidates;\n  }\n}\n"],"mappings":";;AAWA,MAAa,gBAAgB;;AAG7B,MAAa,oBAAoB;;;;;;;;;;;;;AA4CjC,eAAsB,SACpB,OACA,MACA,UAA2B,CAAC,GACH;CACzB,MAAM,OAAO,QAAQ,QAAQ,KAAK,UAAU;CAC5C,MAAM,YAAY,QAAQ,aAAa,KAAK,UAAU;CACtD,MAAM,OAAO,QAAQ,QAAQ,KAAK,UAAU;CAC5C,MAAM,aAAa,QAAQ,cAAc,KAAK,UAAU,cAAc,KAAK,IAAI,OAAO,GAAG,IAAI;CAE7F,MAAM,EAAE,QAAQ,eAAe,MAAM,KAAK,SAAS,MAAM,KAAK;CAE9D,IACE,KAAK,sBAAsB,UAC3B,eAAe,KACf,KAAK,sBAAsB,KAC3B,eAAe,KAAK,mBAEpB,MAAM,IAAI,MACR,8CAA8C,WAAW,wDAAwD,KAAK,kBAAkB,qDAC1I;CAGF,MAAM,OAAO,MAAM,KAAK,MAAM,MAAmB,QAAQ;EACvD,MAAM;EACN;EACA;CACF,CAAC;CAED,MAAM,SAAS,GAAG,KAAK,UAAU;CAEjC,IAAI,YAA8B,KAC/B,QAAQ,QAAQ,IAAI,IAAI,WAAW,MAAM,CAAC,CAAC,CAC3C,KAAK,QAAQ,iBAAiB,IAAI,OAAO,IAAI,KAAK,CAAC;CAEtD,YAAY,MAAM,cAAc,OAAO,WAAW,KAAK,QAAQ;CAE/D,OAAO;EAAE;EAAO,QAAQ,UAAU,MAAM,GAAG,IAAI;CAAE;AACnD;;AAGA,SAAS,iBAAiB,QAAqB,OAA+B;CAC5E,MAAM,WAAqB;EACzB,UAAU,OAAO;EACjB,YAAY,OAAO;EACnB,MAAM,OAAO;EACb;EACA,UAAU,OAAO;CACnB;CAEA,OAAO;EAAE,MAAM,OAAO;EAAM;EAAO;CAAS;AAC9C;;;;;AAMA,eAAe,cACb,OACA,YACA,UAC2B;CAC3B,IAAI,CAAC,UACH,OAAO;CAGT,IAAI;EACF,OAAO,MAAM,SAAS,OAAO,OAAO,UAAU;CAChD,QAAQ;EAGN,OAAO;CACT;AACF"}

package/esm/rag/store/cache-vector-store.d.mts

@@ -1,7 +1,7 @@
 import { VectorStore } from "./vector-store.contract.mjs";
 import { CacheDriver } from "@warlock.js/cache";
 
-//#region ../@warlock.js/ai/src/rag/store/cache-vector-store.d.ts
+//#region ../ai/src/rag/store/cache-vector-store.d.ts
 /**
  * Adapt any `@warlock.js/cache` `CacheDriver` to the {@link VectorStore}
  * narrowing the RAG pipeline depends on. The cache driver IS the vector

package/esm/rag/store/cache-vector-store.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"cache-vector-store.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/store/cache-vector-store.ts"],"mappings":";;;;;;AAsBA;;;;;;;;AAA4E;;;;;;;;;iBAA5D,gBAAA,CAAiB,MAAA,EAAQ,WAAA,aAAwB,WAAW"}
+{"version":3,"file":"cache-vector-store.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/store/cache-vector-store.ts"],"mappings":";;;;;;AAsBA;;;;;;;;AAA4E;;;;;;;;;iBAA5D,gBAAA,CAAiB,MAAA,EAAQ,WAAA,aAAwB,WAAW"}

package/esm/rag/store/cache-vector-store.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/store/cache-vector-store.ts
+//#region ../ai/src/rag/store/cache-vector-store.ts
 /**
 * Adapt any `@warlock.js/cache` `CacheDriver` to the {@link VectorStore}
 * narrowing the RAG pipeline depends on. The cache driver IS the vector

package/esm/rag/store/cache-vector-store.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"cache-vector-store.mjs","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/store/cache-vector-store.ts"],"sourcesContent":["import type { CacheDriver } from \"@warlock.js/cache\";\nimport type { VectorStore } from \"./vector-store.contract\";\n\n/**\n * Adapt any `@warlock.js/cache` `CacheDriver` to the {@link VectorStore}\n * narrowing the RAG pipeline depends on. The cache driver IS the vector\n * store — exactly as `SemanticMemory` and `semanticCache` already use it:\n *\n * - `upsert` → `driver.set(key, value, { vector, tags })`\n * - `query`  → `driver.similar<T>(vector, { topK, threshold, tags })`\n * - `removeNamespace` → `driver.removeNamespace(namespace)`\n *\n * Drivers without similarity support throw `CacheUnsupportedError` from\n * `set({ vector })` / `similar()`; the error surfaces unchanged so the\n * caller sees the cache layer's own message (pointing at the `pg` /\n * `redis` drivers for production-scale similarity).\n *\n * @example\n * const store = cacheVectorStore(new MemoryCacheDriver());\n * await store.upsert(\"ai.rag.docs.guide.0\", { text: \"…\" }, vector);\n * const hits = await store.query(queryVector, { topK: 5, threshold: 0.5 });\n */\nexport function cacheVectorStore(driver: CacheDriver<any, any>): VectorStore {\n  return {\n    async upsert(\n      key: string,\n      value: unknown,\n      vector: number[],\n      tags?: string[],\n    ): Promise<void> {\n      await driver.set(key, value, tags && tags.length > 0 ? { vector, tags } : { vector });\n    },\n\n    async query<T>(\n      vector: number[],\n      options: { topK: number; threshold?: number; tags?: string[] },\n    ): Promise<{ key: string; value: T; score: number }[]> {\n      const hits = await driver.similar<T>(vector, {\n        topK: options.topK,\n        threshold: options.threshold,\n        tags: options.tags,\n      });\n\n      return hits.map((hit: { key: string; value: T; score: number }) => ({\n        key: hit.key,\n        value: hit.value,\n        score: hit.score,\n      }));\n    },\n\n    async removeNamespace(namespace: string): Promise<void> {\n      await driver.removeNamespace(namespace);\n    },\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAsBA,SAAgB,iBAAiB,QAA4C;CAC3E,OAAO;EACL,MAAM,OACJ,KACA,OACA,QACA,MACe;GACf,MAAM,OAAO,IAAI,KAAK,OAAO,QAAQ,KAAK,SAAS,IAAI;IAAE;IAAQ;GAAK,IAAI,EAAE,OAAO,CAAC;EACtF;EAEA,MAAM,MACJ,QACA,SACqD;GAOrD,QAAO,MANY,OAAO,QAAW,QAAQ;IAC3C,MAAM,QAAQ;IACd,WAAW,QAAQ;IACnB,MAAM,QAAQ;GAChB,CAAC,EAEU,CAAC,KAAK,SAAmD;IAClE,KAAK,IAAI;IACT,OAAO,IAAI;IACX,OAAO,IAAI;GACb,EAAE;EACJ;EAEA,MAAM,gBAAgB,WAAkC;GACtD,MAAM,OAAO,gBAAgB,SAAS;EACxC;CACF;AACF"}
+{"version":3,"file":"cache-vector-store.mjs","names":[],"sources":["../../../../../../../../ai/src/rag/store/cache-vector-store.ts"],"sourcesContent":["import type { CacheDriver } from \"@warlock.js/cache\";\nimport type { VectorStore } from \"./vector-store.contract\";\n\n/**\n * Adapt any `@warlock.js/cache` `CacheDriver` to the {@link VectorStore}\n * narrowing the RAG pipeline depends on. The cache driver IS the vector\n * store — exactly as `SemanticMemory` and `semanticCache` already use it:\n *\n * - `upsert` → `driver.set(key, value, { vector, tags })`\n * - `query`  → `driver.similar<T>(vector, { topK, threshold, tags })`\n * - `removeNamespace` → `driver.removeNamespace(namespace)`\n *\n * Drivers without similarity support throw `CacheUnsupportedError` from\n * `set({ vector })` / `similar()`; the error surfaces unchanged so the\n * caller sees the cache layer's own message (pointing at the `pg` /\n * `redis` drivers for production-scale similarity).\n *\n * @example\n * const store = cacheVectorStore(new MemoryCacheDriver());\n * await store.upsert(\"ai.rag.docs.guide.0\", { text: \"…\" }, vector);\n * const hits = await store.query(queryVector, { topK: 5, threshold: 0.5 });\n */\nexport function cacheVectorStore(driver: CacheDriver<any, any>): VectorStore {\n  return {\n    async upsert(\n      key: string,\n      value: unknown,\n      vector: number[],\n      tags?: string[],\n    ): Promise<void> {\n      await driver.set(key, value, tags && tags.length > 0 ? { vector, tags } : { vector });\n    },\n\n    async query<T>(\n      vector: number[],\n      options: { topK: number; threshold?: number; tags?: string[] },\n    ): Promise<{ key: string; value: T; score: number }[]> {\n      const hits = await driver.similar<T>(vector, {\n        topK: options.topK,\n        threshold: options.threshold,\n        tags: options.tags,\n      });\n\n      return hits.map((hit: { key: string; value: T; score: number }) => ({\n        key: hit.key,\n        value: hit.value,\n        score: hit.score,\n      }));\n    },\n\n    async removeNamespace(namespace: string): Promise<void> {\n      await driver.removeNamespace(namespace);\n    },\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAsBA,SAAgB,iBAAiB,QAA4C;CAC3E,OAAO;EACL,MAAM,OACJ,KACA,OACA,QACA,MACe;GACf,MAAM,OAAO,IAAI,KAAK,OAAO,QAAQ,KAAK,SAAS,IAAI;IAAE;IAAQ;GAAK,IAAI,EAAE,OAAO,CAAC;EACtF;EAEA,MAAM,MACJ,QACA,SACqD;GAOrD,QAAO,MANY,OAAO,QAAW,QAAQ;IAC3C,MAAM,QAAQ;IACd,WAAW,QAAQ;IACnB,MAAM,QAAQ;GAChB,CAAC,EAEU,CAAC,KAAK,SAAmD;IAClE,KAAK,IAAI;IACT,OAAO,IAAI;IACX,OAAO,IAAI;GACb,EAAE;EACJ;EAEA,MAAM,gBAAgB,WAAkC;GACtD,MAAM,OAAO,gBAAgB,SAAS;EACxC;CACF;AACF"}

package/esm/rag/store/pg-vector-store.d.mts

@@ -1,6 +1,6 @@
 import { VectorStore } from "./vector-store.contract.mjs";
 
-//#region ../@warlock.js/ai/src/rag/store/pg-vector-store.d.ts
+//#region ../ai/src/rag/store/pg-vector-store.d.ts
 /**
  * Minimal `pg`-compatible client surface the Postgres {@link VectorStore}
  * depends on. Both `pg.Pool` and `pg.Client` satisfy it — the store only

package/esm/rag/store/pg-vector-store.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"pg-vector-store.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/store/pg-vector-store.ts"],"mappings":";;;;;AAaA;;;;;;;;;UAAiB,YAAA;EACf,KAAA,CAAM,IAAA,UAAc,MAAA,eAAqB,OAAO;IAAG,IAAA;EAAA;AAAA;;;;;;;;;;AAuDvC;AA4Fd;;;;AAA8C;AA6U9C;UA7ciB,oBAAA;;EAEf,MAAA,GAAS,YAAY;EA2cwB;EAxc7C,gBAAA;EA4cA;;AAAY;AA0Cd;EAhfE,KAAA;;;;;;;AAgfiF;;EAtejF,UAAA;;;;;;;;EASA,KAAA;;;;;EAMA,YAAA;AAAA;;;;;;;;;;;;;;;;iBA4Fc,aAAA,CAAc,MAAgB;;;;;;;UA6U7B,qBAAA,SAA8B,WAAW;;EAExD,MAAA;;EAEA,YAAA;AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA0Cc,aAAA,CAAc,OAAA,EAAS,oBAAA,GAAuB,qBAAqB"}
+{"version":3,"file":"pg-vector-store.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/store/pg-vector-store.ts"],"mappings":";;;;;AAaA;;;;;;;;;UAAiB,YAAA;EACf,KAAA,CAAM,IAAA,UAAc,MAAA,eAAqB,OAAO;IAAG,IAAA;EAAA;AAAA;;;;;;;;;;AAuDvC;AA4Fd;;;;AAA8C;AA6U9C;UA7ciB,oBAAA;;EAEf,MAAA,GAAS,YAAY;EA2cwB;EAxc7C,gBAAA;EA4cA;;AAAY;AA0Cd;EAhfE,KAAA;;;;;;;AAgfiF;;EAtejF,UAAA;;;;;;;;EASA,KAAA;;;;;EAMA,YAAA;AAAA;;;;;;;;;;;;;;;;iBA4Fc,aAAA,CAAc,MAAgB;;;;;;;UA6U7B,qBAAA,SAA8B,WAAW;;EAExD,MAAA;;EAEA,YAAA;AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA0Cc,aAAA,CAAc,OAAA,EAAS,oBAAA,GAAuB,qBAAqB"}

package/esm/rag/store/pg-vector-store.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/store/pg-vector-store.ts
+//#region ../ai/src/rag/store/pg-vector-store.ts
 /**
 * Default backing table — provisions the store with no extra config when
 * the dev runs {@link VectorStore.schema | ensureSchema} through their

package/esm/rag/store/pg-vector-store.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"pg-vector-store.mjs","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/store/pg-vector-store.ts"],"sourcesContent":["import type { VectorStore } from \"./vector-store.contract\";\n\n/**\n * Minimal `pg`-compatible client surface the Postgres {@link VectorStore}\n * depends on. Both `pg.Pool` and `pg.Client` satisfy it — the store only\n * ever calls `query`.\n *\n * `@warlock.js/ai` takes **no** hard dependency on `pg`; the dev installs\n * it (an optional peer) and passes the client in. Structurally identical\n * to the snapshot / human-interrupt stores' `PgClientLike`, so a single\n * pool can back the orchestrator checkpoint/snapshot tables, the\n * interrupt table, and this vectors table alike.\n */\nexport interface PgClientLike {\n  query(text: string, params?: unknown[]): Promise<{ rows: unknown[] }>;\n}\n\n/**\n * Options for the Postgres {@link VectorStore}.\n *\n * Two mutually-supportive ways to supply the connection (mirroring\n * `ai.human.interrupt.pg`):\n * - **`client`** — pass an already-built `pg.Pool` / `pg.Client` (anything\n *   satisfying {@link PgClientLike}). The store only ever calls `query`\n *   and never opens or closes it; one pool can back several stores.\n * - **`connectionString`** — let the store lazily `import(\"pg\")` and build\n *   its own `Pool`. `@warlock.js/ai` takes **no** hard dependency on\n *   `pg` (an optional peer); when it is absent the store throws a curated\n *   install string at first use, never a raw module-resolution stack trace\n *   at import.\n *\n * Exactly one of the two must be present.\n */\nexport interface PgVectorStoreOptions {\n  /** An already-built `pg.Pool` / `pg.Client` — anything matching {@link PgClientLike}. */\n  client?: PgClientLike;\n\n  /** Connection string the store passes to a lazily-imported `pg.Pool`. */\n  connectionString?: string;\n\n  /**\n   * Backing table name. Defaults to `warlock_ai_rag_vectors`. Must be a\n   * safe SQL identifier — it is interpolated into DDL/DML.\n   */\n  table?: string;\n\n  /**\n   * Embedding dimensionality used in the `CREATE TABLE` DDL emitted by\n   * {@link VectorStore.schema | ensureSchema}. Defaults to `1536`\n   * (OpenAI `text-embedding-3-small`). The column is declared\n   * `vector(N)`; queries and upserts never re-state it, so an existing\n   * table provisioned at a different size is unaffected — only the DDL\n   * helper reads this.\n   */\n  dimensions?: number;\n\n  /**\n   * Approximate-nearest-neighbour index strategy baked into the DDL\n   * emitted by {@link VectorStore.schema | ensureSchema}. Defaults to\n   * `\"hnsw\"` (better recall/latency on modern pgvector). Use `\"ivfflat\"`\n   * for the classic list-partitioned index, or `\"none\"` to emit no ANN\n   * index (exact scan — correct, but linear in row count).\n   */\n  index?: \"hnsw\" | \"ivfflat\" | \"none\";\n\n  /**\n   * `lists` parameter for an `ivfflat` index (ignored for `hnsw` / `none`).\n   * Defaults to `100`. Tune toward `rows / 1000` for large tables.\n   */\n  ivfflatLists?: number;\n}\n\n/**\n * Default backing table — provisions the store with no extra config when\n * the dev runs {@link VectorStore.schema | ensureSchema} through their\n * migration tool.\n */\nconst DEFAULT_TABLE = \"warlock_ai_rag_vectors\";\n\n/** Default embedding width baked into the DDL (OpenAI `text-embedding-3-small`). */\nconst DEFAULT_DIMENSIONS = 1536;\n\n/** Default `ivfflat` list count when that index strategy is chosen. */\nconst DEFAULT_IVFFLAT_LISTS = 100;\n\n/**\n * Allowed characters in a Postgres identifier (table name). The table name\n * is interpolated into DDL/DML, so anything outside this conservative\n * ASCII subset is rejected — interpolating an arbitrary string would be a\n * SQL-injection footgun (mirrors the snapshot / human-interrupt pg stores\n * and `@warlock.js/cache`'s `PgCacheDriver`).\n */\nconst SAFE_IDENTIFIER = /^[A-Za-z_][A-Za-z0-9_]*$/;\n\n/**\n * Module specifier for the optional `pg` driver. Held in a `string`\n * variable so the dynamic `import()` is not statically resolved at compile\n * time — `pg` is an optional peer that need not be installed for this\n * package to type-check or for a cache-only consumer to run.\n */\nconst PG_MODULE = \"pg\";\n\n/**\n * Curated install string surfaced (at use time) when a `connectionString`\n * is configured but the optional `pg` driver is absent. Never thrown at\n * import — a cache-only consumer must be able to load this module.\n */\nconst PG_INSTALL_INSTRUCTIONS = `\nThe @warlock.js/ai Postgres vector store requires the pg package and a\nPostgres database with the pgvector extension. Install the driver with:\n\n  npm install pg\n\nOr with your preferred package manager:\n\n  pnpm add pg\n  yarn add pg\n`.trim();\n\n/**\n * Minimal structural view of a `pg.Pool` constructor — just enough of the\n * `pg` module surface for the store to build a client when handed a\n * `connectionString`. Declared locally (rather than `typeof import(\"pg\")`)\n * so this module type-checks even when `pg` is not installed.\n */\ninterface PgModuleLike {\n  Pool: new (config: { connectionString: string }) => PgClientLike;\n}\n\n/**\n * Lazily import `pg` and return a `Pool` built from `connectionString`. A\n * bare `catch` rethrows the curated install string — a missing optional\n * peer surfaces as actionable guidance, never a raw resolution error.\n */\nasync function buildPgClient(connectionString: string): Promise<PgClientLike> {\n  let sdk: PgModuleLike;\n\n  try {\n    sdk = (await import(PG_MODULE)) as unknown as PgModuleLike;\n  } catch {\n    throw new Error(PG_INSTALL_INSTRUCTIONS);\n  }\n\n  return new sdk.Pool({ connectionString });\n}\n\n/**\n * Serialize a JS `number[]` to the pgvector text literal: `[1,2,3]`.\n * pgvector accepts a vector either as this bracketed literal or via a\n * typed parameter; passing the literal string + an explicit `::vector`\n * cast keeps the store driver-agnostic (no dependency on a registered\n * `pg` type parser).\n *\n * Non-finite components (`NaN` / `±Infinity`) are rejected — pgvector\n * stores only finite floats, and silently coercing them would corrupt the\n * index. The check is cheap relative to the embed call that produced the\n * vector.\n *\n * @example\n * vectorLiteral([1, 0.5, -2]); // \"[1,0.5,-2]\"\n */\nexport function vectorLiteral(vector: number[]): string {\n  let literal = \"[\";\n\n  for (let index = 0; index < vector.length; index++) {\n    const component = vector[index];\n\n    if (!Number.isFinite(component)) {\n      throw new TypeError(\n        `pgVectorStore: embedding component at index ${index} is not finite (${component}); pgvector stores only finite floats.`,\n      );\n    }\n\n    if (index > 0) {\n      literal += \",\";\n    }\n\n    literal += String(component);\n  }\n\n  return literal + \"]\";\n}\n\n/**\n * Coerce a `value` JSONB column back into the stored payload. node-postgres\n * parses `JSONB` into a JS value already, but some pool wrappers hand back\n * the raw string — be defensive across both (mirrors the snapshot store's\n * `parsePayload`).\n */\nfunction parseValue<T>(value: unknown): T {\n  if (typeof value === \"string\") {\n    return JSON.parse(value) as T;\n  }\n\n  return value as T;\n}\n\n/**\n * Coerce a pgvector cosine **distance** (`<=>`, in `[0, 2]`, 0 = identical)\n * into the cosine **similarity** score the {@link VectorStore} contract\n * declares (`[0, 1]`, 1 = identical). `pg` returns the computed distance\n * column as a string for `double precision`; parse then map `1 - distance`,\n * clamped to `[0, 1]` so a tiny floating-point overshoot never yields a\n * score just outside the contract's range.\n */\nfunction distanceToScore(distance: unknown): number {\n  const value = typeof distance === \"string\" ? Number(distance) : (distance as number);\n  const score = 1 - value;\n\n  if (score < 0) {\n    return 0;\n  }\n\n  if (score > 1) {\n    return 1;\n  }\n\n  return score;\n}\n\n/**\n * Postgres + pgvector-backed {@link VectorStore} — one durable row per\n * indexed chunk, keyed by the RAG pipeline's dotted `key`\n * (`ai.rag.<name>.<sourceId>.<chunkIndex>`), with the chunk payload in a\n * `value` JSONB column and the embedding in a `vector` column.\n *\n * Owns: the three RAG vector operations against a pgvector index —\n * `upsert` (INSERT … ON CONFLICT DO UPDATE), `query` (cosine\n * `ORDER BY embedding <=> $vec` with a `threshold` floor + optional `tags`\n * overlap filter, capped at `topK`), and `removeNamespace` (prefix DELETE).\n * Also emits the reference DDL via {@link PgVectorStore.schema} (alias\n * {@link PgVectorStore.ensureSchema}).\n *\n * Does NOT own: the connection lifecycle (a dev-supplied `client` is never\n * closed; a store-built `Pool` from a `connectionString` is also left open\n * for the process to reuse) or schema migration — the dev runs the DDL\n * through their own tool; the framework never auto-migrates.\n *\n * Front it with the {@link pgVectorStore} factory — callers never `new` it.\n */\nclass PgVectorStore implements VectorStore {\n  /** Validated backing table name, safe to interpolate into SQL. */\n  private readonly table: string;\n\n  /** Embedding width baked into the DDL. */\n  private readonly dimensions: number;\n\n  /** ANN index strategy baked into the DDL. */\n  private readonly index: \"hnsw\" | \"ivfflat\" | \"none\";\n\n  /** `lists` parameter for an `ivfflat` index. */\n  private readonly ivfflatLists: number;\n\n  /**\n   * A ready client, or a promise resolving to one the store builds lazily\n   * from a `connectionString`. Resolved once and memoized so the optional\n   * `pg` import happens at most once.\n   */\n  private readonly clientPromise: Promise<PgClientLike>;\n\n  public constructor(options: PgVectorStoreOptions) {\n    const table = options.table ?? DEFAULT_TABLE;\n\n    if (!SAFE_IDENTIFIER.test(table)) {\n      throw new TypeError(\n        `pgVectorStore: invalid table name '${table}'. Allowed: [A-Za-z_][A-Za-z0-9_]*.`,\n      );\n    }\n\n    this.table = table;\n    this.dimensions = options.dimensions ?? DEFAULT_DIMENSIONS;\n    this.index = options.index ?? \"hnsw\";\n    this.ivfflatLists = options.ivfflatLists ?? DEFAULT_IVFFLAT_LISTS;\n\n    if (options.client) {\n      if (typeof options.client.query !== \"function\") {\n        throw new TypeError(\n          \"pgVectorStore requires a 'client' option implementing { query(text, params) } — pass a pg.Pool or pg.Client.\",\n        );\n      }\n\n      this.clientPromise = Promise.resolve(options.client);\n\n      return;\n    }\n\n    if (options.connectionString) {\n      // Defer the optional `pg` import to first use — a curated install\n      // string surfaces from `buildPgClient`, not at construction.\n      this.clientPromise = buildPgClient(options.connectionString);\n\n      return;\n    }\n\n    throw new TypeError(\n      \"pgVectorStore requires either a 'client' or a 'connectionString' option.\",\n    );\n  }\n\n  /**\n   * Resolve the backing client, surfacing the lazy `pg` import's curated\n   * install string on the first call that needs it.\n   */\n  private client(): Promise<PgClientLike> {\n    return this.clientPromise;\n  }\n\n  /**\n   * Index `value` under `key` with its embedding `vector`. Upserts on the\n   * `key` primary key — re-indexing the same chunk overwrites its payload,\n   * embedding, and tags rather than appending a duplicate row. Optional\n   * `tags` ride a `text[]` column so {@link query} can restrict the\n   * candidate set with an array-overlap filter.\n   *\n   * The embedding is sent as a pgvector text literal (`$3`) cast to\n   * `::vector`, so the store needs no registered `pg` type parser. `tags`\n   * defaults to an empty array (never `NULL`) to keep the overlap filter's\n   * `&&` semantics simple.\n   */\n  public async upsert(\n    key: string,\n    value: unknown,\n    vector: number[],\n    tags?: string[],\n  ): Promise<void> {\n    const client = await this.client();\n\n    await client.query(\n      `INSERT INTO ${this.table} (key, value, embedding, tags)\n       VALUES ($1, $2::jsonb, $3::vector, $4::text[])\n       ON CONFLICT (key) DO UPDATE\n         SET value = EXCLUDED.value,\n             embedding = EXCLUDED.embedding,\n             tags = EXCLUDED.tags`,\n      [key, JSON.stringify(value), vectorLiteral(vector), tags ?? []],\n    );\n  }\n\n  /**\n   * Return the cosine-nearest rows to `vector`, mapped to the contract's\n   * `{ key, value, score }` shape. The SQL:\n   *\n   * - computes `embedding <=> $1::vector` (cosine **distance**) once, aliased\n   *   `distance`, and `ORDER BY` it ascending (nearest first);\n   * - applies the `threshold` floor as `distance <= 1 - threshold`\n   *   (similarity `>=` threshold), so the default `0.5` floor maps to a\n   *   `<= 0.5` distance bound — the filter runs in SQL, not in JS, so a\n   *   below-floor row never crosses the wire;\n   * - when `tags` are given, restricts to rows whose `tags` array overlaps\n   *   the requested set via `tags && $tags::text[]` (one-of semantics,\n   *   matching the cache store);\n   * - caps the result at `topK` with `LIMIT`.\n   *\n   * The returned `score` is `1 - distance`, clamped to `[0, 1]`, so callers\n   * see the same cosine-similarity scale the cache store emits.\n   */\n  public async query<T>(\n    vector: number[],\n    options: { topK: number; threshold?: number; tags?: string[] },\n  ): Promise<{ key: string; value: T; score: number }[]> {\n    const client = await this.client();\n    const queryVector = vectorLiteral(vector);\n\n    // $1 = query vector, $2 = topK. Optional threshold + tags are appended\n    // as $3 / $4 only when present, so the prepared statement carries no\n    // unused placeholders.\n    const params: unknown[] = [queryVector, options.topK];\n    const conditions: string[] = [];\n\n    if (options.threshold !== undefined) {\n      params.push(1 - options.threshold);\n      conditions.push(`(embedding <=> $1::vector) <= $${params.length}`);\n    }\n\n    if (options.tags !== undefined && options.tags.length > 0) {\n      params.push(options.tags);\n      conditions.push(`tags && $${params.length}::text[]`);\n    }\n\n    const where = conditions.length > 0 ? `WHERE ${conditions.join(\" AND \")}` : \"\";\n\n    const { rows } = await client.query(\n      `SELECT key, value, (embedding <=> $1::vector) AS distance\n       FROM ${this.table}\n       ${where}\n       ORDER BY embedding <=> $1::vector\n       LIMIT $2`,\n      params,\n    );\n\n    return (rows as Record<string, unknown>[]).map((row) => ({\n      key: row.key as string,\n      value: parseValue<T>(row.value),\n      score: distanceToScore(row.distance),\n    }));\n  }\n\n  /**\n   * Drop every entry written under `namespace`. The RAG pipeline keys\n   * chunks as `<namespace>.<sourceId>.<chunkIndex>`, so a row belongs to\n   * the namespace when its `key` equals it OR begins with `<namespace>.`\n   * — deleting `ai.rag.docs` must not also catch `ai.rag.docs2`. The `_`\n   * and `%` LIKE wildcards in the namespace are escaped so a namespace\n   * that happens to contain them is matched literally.\n   */\n  public async removeNamespace(namespace: string): Promise<void> {\n    const client = await this.client();\n\n    const escaped = namespace\n      .replace(/\\\\/g, \"\\\\\\\\\")\n      .replace(/_/g, \"\\\\_\")\n      .replace(/%/g, \"\\\\%\");\n\n    await client.query(\n      `DELETE FROM ${this.table}\n       WHERE key = $1 OR key LIKE $2 ESCAPE '\\\\'`,\n      [namespace, `${escaped}.%`],\n    );\n  }\n\n  /**\n   * Return the reference migration DDL for this store's backing table,\n   * interpolating the configured table name, embedding width, and ANN\n   * index strategy. The dev runs it once through their migration tool —\n   * the framework never auto-migrates.\n   *\n   * The emitted statements:\n   * 1. `CREATE EXTENSION IF NOT EXISTS vector;` — enables pgvector (needs\n   *    a superuser or a role with `CREATE` on the database the first time).\n   * 2. `CREATE TABLE IF NOT EXISTS <table> (key TEXT PRIMARY KEY, value\n   *    JSONB NOT NULL, embedding vector(<dimensions>) NOT NULL, tags\n   *    text[] NOT NULL DEFAULT '{}');`\n   * 3. A GIN index on `tags` so the array-overlap filter stays sargable.\n   * 4. The chosen ANN index over `embedding` using `vector_cosine_ops`:\n   *    - `\"hnsw\"` → `USING hnsw (embedding vector_cosine_ops)`;\n   *    - `\"ivfflat\"` → `USING ivfflat (embedding vector_cosine_ops)\n   *      WITH (lists = <ivfflatLists>)`;\n   *    - `\"none\"` → emitted as a comment (exact scan, no ANN index).\n   *\n   * @example\n   * const store = pgVectorStore({ client: pool, dimensions: 1536 });\n   * await pool.query(store.ensureSchema());\n   */\n  public schema(): string {\n    const lines = [\n      `CREATE EXTENSION IF NOT EXISTS vector;`,\n      `CREATE TABLE IF NOT EXISTS ${this.table} (`,\n      `  key        TEXT PRIMARY KEY,`,\n      `  value      JSONB NOT NULL,`,\n      `  embedding  vector(${this.dimensions}) NOT NULL,`,\n      `  tags       TEXT[] NOT NULL DEFAULT '{}'`,\n      `);`,\n      `CREATE INDEX IF NOT EXISTS idx_${this.table}_tags`,\n      `  ON ${this.table} USING gin (tags);`,\n    ];\n\n    if (this.index === \"hnsw\") {\n      lines.push(\n        `CREATE INDEX IF NOT EXISTS idx_${this.table}_embedding`,\n        `  ON ${this.table} USING hnsw (embedding vector_cosine_ops);`,\n      );\n    } else if (this.index === \"ivfflat\") {\n      lines.push(\n        `CREATE INDEX IF NOT EXISTS idx_${this.table}_embedding`,\n        `  ON ${this.table} USING ivfflat (embedding vector_cosine_ops)`,\n        `  WITH (lists = ${this.ivfflatLists});`,\n      );\n    } else {\n      lines.push(\n        `-- No ANN index requested (index: \"none\"): cosine queries fall back`,\n        `-- to an exact sequential scan, which is correct but linear in rows.`,\n      );\n    }\n\n    return lines.join(\"\\n\");\n  }\n\n  /**\n   * Alias for {@link PgVectorStore.schema} — reads more naturally in a\n   * migration script (`await pool.query(store.ensureSchema())`). Returns\n   * the identical DDL string; it does NOT execute anything against the\n   * database (the store never auto-migrates).\n   */\n  public ensureSchema(): string {\n    return this.schema();\n  }\n}\n\n/**\n * The {@link VectorStore} surface plus the pg store's extra DDL helpers.\n * `schema()` / `ensureSchema()` are not part of the base contract (the\n * cache store has no backing table), so the factory's return type widens\n * it for callers that want the migration SQL.\n */\nexport interface PgVectorStoreInstance extends VectorStore {\n  /** Reference migration DDL (extension + table + indexes). Never executed. */\n  schema(): string;\n  /** Alias for {@link PgVectorStoreInstance.schema} — reads better in a migration script. */\n  ensureSchema(): string;\n}\n\n/**\n * Create a Postgres + pgvector-backed {@link VectorStore} for the RAG\n * pipeline. Either pass a live `pg.Pool` / `pg.Client` (`{ client }`) —\n * `@warlock.js/ai` never imports `pg` in that case — or a\n * `{ connectionString }` and let the store lazily `import(\"pg\")` to build\n * its own pool. When `pg` is not installed, a curated install string\n * surfaces on first use, never at import.\n *\n * Run {@link PgVectorStoreInstance.ensureSchema} through your migration\n * tool once before use (it enables the `vector` extension, creates the\n * table, and builds the tag + ANN indexes); the store never auto-migrates.\n *\n * Index and query MUST use the same embedding model — the `vector(N)`\n * column width is fixed at table-creation time from `dimensions`.\n *\n * @example\n * import { Pool } from \"pg\";\n * import { ai } from \"@warlock.js/ai\";\n *\n * const pool = new Pool({ connectionString: process.env.DATABASE_URL });\n * const store = ai.rag.pgVectorStore({ client: pool, dimensions: 1536 });\n *\n * // Once, via your migration tooling:\n * // await pool.query(store.ensureSchema());\n *\n * const kb = ai.rag({\n *   name: \"docs\",\n *   embedder: openai.embedder({ name: \"text-embedding-3-small\" }),\n *   store,\n * });\n *\n * @example\n * // Let the store build its own pool from a connection string:\n * const store = ai.rag.pgVectorStore({\n *   connectionString: process.env.DATABASE_URL,\n *   index: \"ivfflat\",\n *   ivfflatLists: 200,\n * });\n */\nexport function pgVectorStore(options: PgVectorStoreOptions): PgVectorStoreInstance {\n  return new PgVectorStore(options);\n}\n"],"mappings":";;;;;;AA6EA,MAAM,gBAAgB;;AAGtB,MAAM,qBAAqB;;AAG3B,MAAM,wBAAwB;;;;;;;;AAS9B,MAAM,kBAAkB;;;;;;;AAQxB,MAAM,YAAY;;;;;;AAOlB,MAAM,0BAA0B;;;;;;;;;;EAU9B,KAAK;;;;;;AAiBP,eAAe,cAAc,kBAAiD;CAC5E,IAAI;CAEJ,IAAI;EACF,MAAO,MAAM,OAAO;CACtB,QAAQ;EACN,MAAM,IAAI,MAAM,uBAAuB;CACzC;CAEA,OAAO,IAAI,IAAI,KAAK,EAAE,iBAAiB,CAAC;AAC1C;;;;;;;;;;;;;;;;AAiBA,SAAgB,cAAc,QAA0B;CACtD,IAAI,UAAU;CAEd,KAAK,IAAI,QAAQ,GAAG,QAAQ,OAAO,QAAQ,SAAS;EAClD,MAAM,YAAY,OAAO;EAEzB,IAAI,CAAC,OAAO,SAAS,SAAS,GAC5B,MAAM,IAAI,UACR,+CAA+C,MAAM,kBAAkB,UAAU,uCACnF;EAGF,IAAI,QAAQ,GACV,WAAW;EAGb,WAAW,OAAO,SAAS;CAC7B;CAEA,OAAO,UAAU;AACnB;;;;;;;AAQA,SAAS,WAAc,OAAmB;CACxC,IAAI,OAAO,UAAU,UACnB,OAAO,KAAK,MAAM,KAAK;CAGzB,OAAO;AACT;;;;;;;;;AAUA,SAAS,gBAAgB,UAA2B;CAElD,MAAM,QAAQ,KADA,OAAO,aAAa,WAAW,OAAO,QAAQ,IAAK;CAGjE,IAAI,QAAQ,GACV,OAAO;CAGT,IAAI,QAAQ,GACV,OAAO;CAGT,OAAO;AACT;;;;;;;;;;;;;;;;;;;;;AAsBA,IAAM,gBAAN,MAA2C;CAoBzC,AAAO,YAAY,SAA+B;EAChD,MAAM,QAAQ,QAAQ,SAAS;EAE/B,IAAI,CAAC,gBAAgB,KAAK,KAAK,GAC7B,MAAM,IAAI,UACR,sCAAsC,MAAM,oCAC9C;EAGF,KAAK,QAAQ;EACb,KAAK,aAAa,QAAQ,cAAc;EACxC,KAAK,QAAQ,QAAQ,SAAS;EAC9B,KAAK,eAAe,QAAQ,gBAAgB;EAE5C,IAAI,QAAQ,QAAQ;GAClB,IAAI,OAAO,QAAQ,OAAO,UAAU,YAClC,MAAM,IAAI,UACR,8GACF;GAGF,KAAK,gBAAgB,QAAQ,QAAQ,QAAQ,MAAM;GAEnD;EACF;EAEA,IAAI,QAAQ,kBAAkB;GAG5B,KAAK,gBAAgB,cAAc,QAAQ,gBAAgB;GAE3D;EACF;EAEA,MAAM,IAAI,UACR,0EACF;CACF;;;;;CAMA,AAAQ,SAAgC;EACtC,OAAO,KAAK;CACd;;;;;;;;;;;;;CAcA,MAAa,OACX,KACA,OACA,QACA,MACe;EAGf,OAAM,MAFe,KAAK,OAAO,EAErB,CAAC,MACX,eAAe,KAAK,MAAM;;;;;oCAM1B;GAAC;GAAK,KAAK,UAAU,KAAK;GAAG,cAAc,MAAM;GAAG,QAAQ,CAAC;EAAC,CAChE;CACF;;;;;;;;;;;;;;;;;;;CAoBA,MAAa,MACX,QACA,SACqD;EACrD,MAAM,SAAS,MAAM,KAAK,OAAO;EAMjC,MAAM,SAAoB,CALN,cAAc,MAKG,GAAG,QAAQ,IAAI;EACpD,MAAM,aAAuB,CAAC;EAE9B,IAAI,QAAQ,cAAc,QAAW;GACnC,OAAO,KAAK,IAAI,QAAQ,SAAS;GACjC,WAAW,KAAK,kCAAkC,OAAO,QAAQ;EACnE;EAEA,IAAI,QAAQ,SAAS,UAAa,QAAQ,KAAK,SAAS,GAAG;GACzD,OAAO,KAAK,QAAQ,IAAI;GACxB,WAAW,KAAK,YAAY,OAAO,OAAO,SAAS;EACrD;EAEA,MAAM,QAAQ,WAAW,SAAS,IAAI,SAAS,WAAW,KAAK,OAAO,MAAM;EAE5E,MAAM,EAAE,SAAS,MAAM,OAAO,MAC5B;cACQ,KAAK,MAAM;SAChB,MAAM;;kBAGT,MACF;EAEA,OAAQ,KAAmC,KAAK,SAAS;GACvD,KAAK,IAAI;GACT,OAAO,WAAc,IAAI,KAAK;GAC9B,OAAO,gBAAgB,IAAI,QAAQ;EACrC,EAAE;CACJ;;;;;;;;;CAUA,MAAa,gBAAgB,WAAkC;EAC7D,MAAM,SAAS,MAAM,KAAK,OAAO;EAEjC,MAAM,UAAU,UACb,QAAQ,OAAO,MAAM,CAAC,CACtB,QAAQ,MAAM,KAAK,CAAC,CACpB,QAAQ,MAAM,KAAK;EAEtB,MAAM,OAAO,MACX,eAAe,KAAK,MAAM;mDAE1B,CAAC,WAAW,GAAG,QAAQ,GAAG,CAC5B;CACF;;;;;;;;;;;;;;;;;;;;;;;;CAyBA,AAAO,SAAiB;EACtB,MAAM,QAAQ;GACZ;GACA,8BAA8B,KAAK,MAAM;GACzC;GACA;GACA,uBAAuB,KAAK,WAAW;GACvC;GACA;GACA,kCAAkC,KAAK,MAAM;GAC7C,QAAQ,KAAK,MAAM;EACrB;EAEA,IAAI,KAAK,UAAU,QACjB,MAAM,KACJ,kCAAkC,KAAK,MAAM,aAC7C,QAAQ,KAAK,MAAM,2CACrB;OACK,IAAI,KAAK,UAAU,WACxB,MAAM,KACJ,kCAAkC,KAAK,MAAM,aAC7C,QAAQ,KAAK,MAAM,+CACnB,mBAAmB,KAAK,aAAa,GACvC;OAEA,MAAM,KACJ,uEACA,sEACF;EAGF,OAAO,MAAM,KAAK,IAAI;CACxB;;;;;;;CAQA,AAAO,eAAuB;EAC5B,OAAO,KAAK,OAAO;CACrB;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsDA,SAAgB,cAAc,SAAsD;CAClF,OAAO,IAAI,cAAc,OAAO;AAClC"}
+{"version":3,"file":"pg-vector-store.mjs","names":[],"sources":["../../../../../../../../ai/src/rag/store/pg-vector-store.ts"],"sourcesContent":["import type { VectorStore } from \"./vector-store.contract\";\n\n/**\n * Minimal `pg`-compatible client surface the Postgres {@link VectorStore}\n * depends on. Both `pg.Pool` and `pg.Client` satisfy it — the store only\n * ever calls `query`.\n *\n * `@warlock.js/ai` takes **no** hard dependency on `pg`; the dev installs\n * it (an optional peer) and passes the client in. Structurally identical\n * to the snapshot / human-interrupt stores' `PgClientLike`, so a single\n * pool can back the orchestrator checkpoint/snapshot tables, the\n * interrupt table, and this vectors table alike.\n */\nexport interface PgClientLike {\n  query(text: string, params?: unknown[]): Promise<{ rows: unknown[] }>;\n}\n\n/**\n * Options for the Postgres {@link VectorStore}.\n *\n * Two mutually-supportive ways to supply the connection (mirroring\n * `ai.human.interrupt.pg`):\n * - **`client`** — pass an already-built `pg.Pool` / `pg.Client` (anything\n *   satisfying {@link PgClientLike}). The store only ever calls `query`\n *   and never opens or closes it; one pool can back several stores.\n * - **`connectionString`** — let the store lazily `import(\"pg\")` and build\n *   its own `Pool`. `@warlock.js/ai` takes **no** hard dependency on\n *   `pg` (an optional peer); when it is absent the store throws a curated\n *   install string at first use, never a raw module-resolution stack trace\n *   at import.\n *\n * Exactly one of the two must be present.\n */\nexport interface PgVectorStoreOptions {\n  /** An already-built `pg.Pool` / `pg.Client` — anything matching {@link PgClientLike}. */\n  client?: PgClientLike;\n\n  /** Connection string the store passes to a lazily-imported `pg.Pool`. */\n  connectionString?: string;\n\n  /**\n   * Backing table name. Defaults to `warlock_ai_rag_vectors`. Must be a\n   * safe SQL identifier — it is interpolated into DDL/DML.\n   */\n  table?: string;\n\n  /**\n   * Embedding dimensionality used in the `CREATE TABLE` DDL emitted by\n   * {@link VectorStore.schema | ensureSchema}. Defaults to `1536`\n   * (OpenAI `text-embedding-3-small`). The column is declared\n   * `vector(N)`; queries and upserts never re-state it, so an existing\n   * table provisioned at a different size is unaffected — only the DDL\n   * helper reads this.\n   */\n  dimensions?: number;\n\n  /**\n   * Approximate-nearest-neighbour index strategy baked into the DDL\n   * emitted by {@link VectorStore.schema | ensureSchema}. Defaults to\n   * `\"hnsw\"` (better recall/latency on modern pgvector). Use `\"ivfflat\"`\n   * for the classic list-partitioned index, or `\"none\"` to emit no ANN\n   * index (exact scan — correct, but linear in row count).\n   */\n  index?: \"hnsw\" | \"ivfflat\" | \"none\";\n\n  /**\n   * `lists` parameter for an `ivfflat` index (ignored for `hnsw` / `none`).\n   * Defaults to `100`. Tune toward `rows / 1000` for large tables.\n   */\n  ivfflatLists?: number;\n}\n\n/**\n * Default backing table — provisions the store with no extra config when\n * the dev runs {@link VectorStore.schema | ensureSchema} through their\n * migration tool.\n */\nconst DEFAULT_TABLE = \"warlock_ai_rag_vectors\";\n\n/** Default embedding width baked into the DDL (OpenAI `text-embedding-3-small`). */\nconst DEFAULT_DIMENSIONS = 1536;\n\n/** Default `ivfflat` list count when that index strategy is chosen. */\nconst DEFAULT_IVFFLAT_LISTS = 100;\n\n/**\n * Allowed characters in a Postgres identifier (table name). The table name\n * is interpolated into DDL/DML, so anything outside this conservative\n * ASCII subset is rejected — interpolating an arbitrary string would be a\n * SQL-injection footgun (mirrors the snapshot / human-interrupt pg stores\n * and `@warlock.js/cache`'s `PgCacheDriver`).\n */\nconst SAFE_IDENTIFIER = /^[A-Za-z_][A-Za-z0-9_]*$/;\n\n/**\n * Module specifier for the optional `pg` driver. Held in a `string`\n * variable so the dynamic `import()` is not statically resolved at compile\n * time — `pg` is an optional peer that need not be installed for this\n * package to type-check or for a cache-only consumer to run.\n */\nconst PG_MODULE = \"pg\";\n\n/**\n * Curated install string surfaced (at use time) when a `connectionString`\n * is configured but the optional `pg` driver is absent. Never thrown at\n * import — a cache-only consumer must be able to load this module.\n */\nconst PG_INSTALL_INSTRUCTIONS = `\nThe @warlock.js/ai Postgres vector store requires the pg package and a\nPostgres database with the pgvector extension. Install the driver with:\n\n  npm install pg\n\nOr with your preferred package manager:\n\n  pnpm add pg\n  yarn add pg\n`.trim();\n\n/**\n * Minimal structural view of a `pg.Pool` constructor — just enough of the\n * `pg` module surface for the store to build a client when handed a\n * `connectionString`. Declared locally (rather than `typeof import(\"pg\")`)\n * so this module type-checks even when `pg` is not installed.\n */\ninterface PgModuleLike {\n  Pool: new (config: { connectionString: string }) => PgClientLike;\n}\n\n/**\n * Lazily import `pg` and return a `Pool` built from `connectionString`. A\n * bare `catch` rethrows the curated install string — a missing optional\n * peer surfaces as actionable guidance, never a raw resolution error.\n */\nasync function buildPgClient(connectionString: string): Promise<PgClientLike> {\n  let sdk: PgModuleLike;\n\n  try {\n    sdk = (await import(PG_MODULE)) as unknown as PgModuleLike;\n  } catch {\n    throw new Error(PG_INSTALL_INSTRUCTIONS);\n  }\n\n  return new sdk.Pool({ connectionString });\n}\n\n/**\n * Serialize a JS `number[]` to the pgvector text literal: `[1,2,3]`.\n * pgvector accepts a vector either as this bracketed literal or via a\n * typed parameter; passing the literal string + an explicit `::vector`\n * cast keeps the store driver-agnostic (no dependency on a registered\n * `pg` type parser).\n *\n * Non-finite components (`NaN` / `±Infinity`) are rejected — pgvector\n * stores only finite floats, and silently coercing them would corrupt the\n * index. The check is cheap relative to the embed call that produced the\n * vector.\n *\n * @example\n * vectorLiteral([1, 0.5, -2]); // \"[1,0.5,-2]\"\n */\nexport function vectorLiteral(vector: number[]): string {\n  let literal = \"[\";\n\n  for (let index = 0; index < vector.length; index++) {\n    const component = vector[index];\n\n    if (!Number.isFinite(component)) {\n      throw new TypeError(\n        `pgVectorStore: embedding component at index ${index} is not finite (${component}); pgvector stores only finite floats.`,\n      );\n    }\n\n    if (index > 0) {\n      literal += \",\";\n    }\n\n    literal += String(component);\n  }\n\n  return literal + \"]\";\n}\n\n/**\n * Coerce a `value` JSONB column back into the stored payload. node-postgres\n * parses `JSONB` into a JS value already, but some pool wrappers hand back\n * the raw string — be defensive across both (mirrors the snapshot store's\n * `parsePayload`).\n */\nfunction parseValue<T>(value: unknown): T {\n  if (typeof value === \"string\") {\n    return JSON.parse(value) as T;\n  }\n\n  return value as T;\n}\n\n/**\n * Coerce a pgvector cosine **distance** (`<=>`, in `[0, 2]`, 0 = identical)\n * into the cosine **similarity** score the {@link VectorStore} contract\n * declares (`[0, 1]`, 1 = identical). `pg` returns the computed distance\n * column as a string for `double precision`; parse then map `1 - distance`,\n * clamped to `[0, 1]` so a tiny floating-point overshoot never yields a\n * score just outside the contract's range.\n */\nfunction distanceToScore(distance: unknown): number {\n  const value = typeof distance === \"string\" ? Number(distance) : (distance as number);\n  const score = 1 - value;\n\n  if (score < 0) {\n    return 0;\n  }\n\n  if (score > 1) {\n    return 1;\n  }\n\n  return score;\n}\n\n/**\n * Postgres + pgvector-backed {@link VectorStore} — one durable row per\n * indexed chunk, keyed by the RAG pipeline's dotted `key`\n * (`ai.rag.<name>.<sourceId>.<chunkIndex>`), with the chunk payload in a\n * `value` JSONB column and the embedding in a `vector` column.\n *\n * Owns: the three RAG vector operations against a pgvector index —\n * `upsert` (INSERT … ON CONFLICT DO UPDATE), `query` (cosine\n * `ORDER BY embedding <=> $vec` with a `threshold` floor + optional `tags`\n * overlap filter, capped at `topK`), and `removeNamespace` (prefix DELETE).\n * Also emits the reference DDL via {@link PgVectorStore.schema} (alias\n * {@link PgVectorStore.ensureSchema}).\n *\n * Does NOT own: the connection lifecycle (a dev-supplied `client` is never\n * closed; a store-built `Pool` from a `connectionString` is also left open\n * for the process to reuse) or schema migration — the dev runs the DDL\n * through their own tool; the framework never auto-migrates.\n *\n * Front it with the {@link pgVectorStore} factory — callers never `new` it.\n */\nclass PgVectorStore implements VectorStore {\n  /** Validated backing table name, safe to interpolate into SQL. */\n  private readonly table: string;\n\n  /** Embedding width baked into the DDL. */\n  private readonly dimensions: number;\n\n  /** ANN index strategy baked into the DDL. */\n  private readonly index: \"hnsw\" | \"ivfflat\" | \"none\";\n\n  /** `lists` parameter for an `ivfflat` index. */\n  private readonly ivfflatLists: number;\n\n  /**\n   * A ready client, or a promise resolving to one the store builds lazily\n   * from a `connectionString`. Resolved once and memoized so the optional\n   * `pg` import happens at most once.\n   */\n  private readonly clientPromise: Promise<PgClientLike>;\n\n  public constructor(options: PgVectorStoreOptions) {\n    const table = options.table ?? DEFAULT_TABLE;\n\n    if (!SAFE_IDENTIFIER.test(table)) {\n      throw new TypeError(\n        `pgVectorStore: invalid table name '${table}'. Allowed: [A-Za-z_][A-Za-z0-9_]*.`,\n      );\n    }\n\n    this.table = table;\n    this.dimensions = options.dimensions ?? DEFAULT_DIMENSIONS;\n    this.index = options.index ?? \"hnsw\";\n    this.ivfflatLists = options.ivfflatLists ?? DEFAULT_IVFFLAT_LISTS;\n\n    if (options.client) {\n      if (typeof options.client.query !== \"function\") {\n        throw new TypeError(\n          \"pgVectorStore requires a 'client' option implementing { query(text, params) } — pass a pg.Pool or pg.Client.\",\n        );\n      }\n\n      this.clientPromise = Promise.resolve(options.client);\n\n      return;\n    }\n\n    if (options.connectionString) {\n      // Defer the optional `pg` import to first use — a curated install\n      // string surfaces from `buildPgClient`, not at construction.\n      this.clientPromise = buildPgClient(options.connectionString);\n\n      return;\n    }\n\n    throw new TypeError(\n      \"pgVectorStore requires either a 'client' or a 'connectionString' option.\",\n    );\n  }\n\n  /**\n   * Resolve the backing client, surfacing the lazy `pg` import's curated\n   * install string on the first call that needs it.\n   */\n  private client(): Promise<PgClientLike> {\n    return this.clientPromise;\n  }\n\n  /**\n   * Index `value` under `key` with its embedding `vector`. Upserts on the\n   * `key` primary key — re-indexing the same chunk overwrites its payload,\n   * embedding, and tags rather than appending a duplicate row. Optional\n   * `tags` ride a `text[]` column so {@link query} can restrict the\n   * candidate set with an array-overlap filter.\n   *\n   * The embedding is sent as a pgvector text literal (`$3`) cast to\n   * `::vector`, so the store needs no registered `pg` type parser. `tags`\n   * defaults to an empty array (never `NULL`) to keep the overlap filter's\n   * `&&` semantics simple.\n   */\n  public async upsert(\n    key: string,\n    value: unknown,\n    vector: number[],\n    tags?: string[],\n  ): Promise<void> {\n    const client = await this.client();\n\n    await client.query(\n      `INSERT INTO ${this.table} (key, value, embedding, tags)\n       VALUES ($1, $2::jsonb, $3::vector, $4::text[])\n       ON CONFLICT (key) DO UPDATE\n         SET value = EXCLUDED.value,\n             embedding = EXCLUDED.embedding,\n             tags = EXCLUDED.tags`,\n      [key, JSON.stringify(value), vectorLiteral(vector), tags ?? []],\n    );\n  }\n\n  /**\n   * Return the cosine-nearest rows to `vector`, mapped to the contract's\n   * `{ key, value, score }` shape. The SQL:\n   *\n   * - computes `embedding <=> $1::vector` (cosine **distance**) once, aliased\n   *   `distance`, and `ORDER BY` it ascending (nearest first);\n   * - applies the `threshold` floor as `distance <= 1 - threshold`\n   *   (similarity `>=` threshold), so the default `0.5` floor maps to a\n   *   `<= 0.5` distance bound — the filter runs in SQL, not in JS, so a\n   *   below-floor row never crosses the wire;\n   * - when `tags` are given, restricts to rows whose `tags` array overlaps\n   *   the requested set via `tags && $tags::text[]` (one-of semantics,\n   *   matching the cache store);\n   * - caps the result at `topK` with `LIMIT`.\n   *\n   * The returned `score` is `1 - distance`, clamped to `[0, 1]`, so callers\n   * see the same cosine-similarity scale the cache store emits.\n   */\n  public async query<T>(\n    vector: number[],\n    options: { topK: number; threshold?: number; tags?: string[] },\n  ): Promise<{ key: string; value: T; score: number }[]> {\n    const client = await this.client();\n    const queryVector = vectorLiteral(vector);\n\n    // $1 = query vector, $2 = topK. Optional threshold + tags are appended\n    // as $3 / $4 only when present, so the prepared statement carries no\n    // unused placeholders.\n    const params: unknown[] = [queryVector, options.topK];\n    const conditions: string[] = [];\n\n    if (options.threshold !== undefined) {\n      params.push(1 - options.threshold);\n      conditions.push(`(embedding <=> $1::vector) <= $${params.length}`);\n    }\n\n    if (options.tags !== undefined && options.tags.length > 0) {\n      params.push(options.tags);\n      conditions.push(`tags && $${params.length}::text[]`);\n    }\n\n    const where = conditions.length > 0 ? `WHERE ${conditions.join(\" AND \")}` : \"\";\n\n    const { rows } = await client.query(\n      `SELECT key, value, (embedding <=> $1::vector) AS distance\n       FROM ${this.table}\n       ${where}\n       ORDER BY embedding <=> $1::vector\n       LIMIT $2`,\n      params,\n    );\n\n    return (rows as Record<string, unknown>[]).map((row) => ({\n      key: row.key as string,\n      value: parseValue<T>(row.value),\n      score: distanceToScore(row.distance),\n    }));\n  }\n\n  /**\n   * Drop every entry written under `namespace`. The RAG pipeline keys\n   * chunks as `<namespace>.<sourceId>.<chunkIndex>`, so a row belongs to\n   * the namespace when its `key` equals it OR begins with `<namespace>.`\n   * — deleting `ai.rag.docs` must not also catch `ai.rag.docs2`. The `_`\n   * and `%` LIKE wildcards in the namespace are escaped so a namespace\n   * that happens to contain them is matched literally.\n   */\n  public async removeNamespace(namespace: string): Promise<void> {\n    const client = await this.client();\n\n    const escaped = namespace\n      .replace(/\\\\/g, \"\\\\\\\\\")\n      .replace(/_/g, \"\\\\_\")\n      .replace(/%/g, \"\\\\%\");\n\n    await client.query(\n      `DELETE FROM ${this.table}\n       WHERE key = $1 OR key LIKE $2 ESCAPE '\\\\'`,\n      [namespace, `${escaped}.%`],\n    );\n  }\n\n  /**\n   * Return the reference migration DDL for this store's backing table,\n   * interpolating the configured table name, embedding width, and ANN\n   * index strategy. The dev runs it once through their migration tool —\n   * the framework never auto-migrates.\n   *\n   * The emitted statements:\n   * 1. `CREATE EXTENSION IF NOT EXISTS vector;` — enables pgvector (needs\n   *    a superuser or a role with `CREATE` on the database the first time).\n   * 2. `CREATE TABLE IF NOT EXISTS <table> (key TEXT PRIMARY KEY, value\n   *    JSONB NOT NULL, embedding vector(<dimensions>) NOT NULL, tags\n   *    text[] NOT NULL DEFAULT '{}');`\n   * 3. A GIN index on `tags` so the array-overlap filter stays sargable.\n   * 4. The chosen ANN index over `embedding` using `vector_cosine_ops`:\n   *    - `\"hnsw\"` → `USING hnsw (embedding vector_cosine_ops)`;\n   *    - `\"ivfflat\"` → `USING ivfflat (embedding vector_cosine_ops)\n   *      WITH (lists = <ivfflatLists>)`;\n   *    - `\"none\"` → emitted as a comment (exact scan, no ANN index).\n   *\n   * @example\n   * const store = pgVectorStore({ client: pool, dimensions: 1536 });\n   * await pool.query(store.ensureSchema());\n   */\n  public schema(): string {\n    const lines = [\n      `CREATE EXTENSION IF NOT EXISTS vector;`,\n      `CREATE TABLE IF NOT EXISTS ${this.table} (`,\n      `  key        TEXT PRIMARY KEY,`,\n      `  value      JSONB NOT NULL,`,\n      `  embedding  vector(${this.dimensions}) NOT NULL,`,\n      `  tags       TEXT[] NOT NULL DEFAULT '{}'`,\n      `);`,\n      `CREATE INDEX IF NOT EXISTS idx_${this.table}_tags`,\n      `  ON ${this.table} USING gin (tags);`,\n    ];\n\n    if (this.index === \"hnsw\") {\n      lines.push(\n        `CREATE INDEX IF NOT EXISTS idx_${this.table}_embedding`,\n        `  ON ${this.table} USING hnsw (embedding vector_cosine_ops);`,\n      );\n    } else if (this.index === \"ivfflat\") {\n      lines.push(\n        `CREATE INDEX IF NOT EXISTS idx_${this.table}_embedding`,\n        `  ON ${this.table} USING ivfflat (embedding vector_cosine_ops)`,\n        `  WITH (lists = ${this.ivfflatLists});`,\n      );\n    } else {\n      lines.push(\n        `-- No ANN index requested (index: \"none\"): cosine queries fall back`,\n        `-- to an exact sequential scan, which is correct but linear in rows.`,\n      );\n    }\n\n    return lines.join(\"\\n\");\n  }\n\n  /**\n   * Alias for {@link PgVectorStore.schema} — reads more naturally in a\n   * migration script (`await pool.query(store.ensureSchema())`). Returns\n   * the identical DDL string; it does NOT execute anything against the\n   * database (the store never auto-migrates).\n   */\n  public ensureSchema(): string {\n    return this.schema();\n  }\n}\n\n/**\n * The {@link VectorStore} surface plus the pg store's extra DDL helpers.\n * `schema()` / `ensureSchema()` are not part of the base contract (the\n * cache store has no backing table), so the factory's return type widens\n * it for callers that want the migration SQL.\n */\nexport interface PgVectorStoreInstance extends VectorStore {\n  /** Reference migration DDL (extension + table + indexes). Never executed. */\n  schema(): string;\n  /** Alias for {@link PgVectorStoreInstance.schema} — reads better in a migration script. */\n  ensureSchema(): string;\n}\n\n/**\n * Create a Postgres + pgvector-backed {@link VectorStore} for the RAG\n * pipeline. Either pass a live `pg.Pool` / `pg.Client` (`{ client }`) —\n * `@warlock.js/ai` never imports `pg` in that case — or a\n * `{ connectionString }` and let the store lazily `import(\"pg\")` to build\n * its own pool. When `pg` is not installed, a curated install string\n * surfaces on first use, never at import.\n *\n * Run {@link PgVectorStoreInstance.ensureSchema} through your migration\n * tool once before use (it enables the `vector` extension, creates the\n * table, and builds the tag + ANN indexes); the store never auto-migrates.\n *\n * Index and query MUST use the same embedding model — the `vector(N)`\n * column width is fixed at table-creation time from `dimensions`.\n *\n * @example\n * import { Pool } from \"pg\";\n * import { ai } from \"@warlock.js/ai\";\n *\n * const pool = new Pool({ connectionString: process.env.DATABASE_URL });\n * const store = ai.rag.pgVectorStore({ client: pool, dimensions: 1536 });\n *\n * // Once, via your migration tooling:\n * // await pool.query(store.ensureSchema());\n *\n * const kb = ai.rag({\n *   name: \"docs\",\n *   embedder: openai.embedder({ name: \"text-embedding-3-small\" }),\n *   store,\n * });\n *\n * @example\n * // Let the store build its own pool from a connection string:\n * const store = ai.rag.pgVectorStore({\n *   connectionString: process.env.DATABASE_URL,\n *   index: \"ivfflat\",\n *   ivfflatLists: 200,\n * });\n */\nexport function pgVectorStore(options: PgVectorStoreOptions): PgVectorStoreInstance {\n  return new PgVectorStore(options);\n}\n"],"mappings":";;;;;;AA6EA,MAAM,gBAAgB;;AAGtB,MAAM,qBAAqB;;AAG3B,MAAM,wBAAwB;;;;;;;;AAS9B,MAAM,kBAAkB;;;;;;;AAQxB,MAAM,YAAY;;;;;;AAOlB,MAAM,0BAA0B;;;;;;;;;;EAU9B,KAAK;;;;;;AAiBP,eAAe,cAAc,kBAAiD;CAC5E,IAAI;CAEJ,IAAI;EACF,MAAO,MAAM,OAAO;CACtB,QAAQ;EACN,MAAM,IAAI,MAAM,uBAAuB;CACzC;CAEA,OAAO,IAAI,IAAI,KAAK,EAAE,iBAAiB,CAAC;AAC1C;;;;;;;;;;;;;;;;AAiBA,SAAgB,cAAc,QAA0B;CACtD,IAAI,UAAU;CAEd,KAAK,IAAI,QAAQ,GAAG,QAAQ,OAAO,QAAQ,SAAS;EAClD,MAAM,YAAY,OAAO;EAEzB,IAAI,CAAC,OAAO,SAAS,SAAS,GAC5B,MAAM,IAAI,UACR,+CAA+C,MAAM,kBAAkB,UAAU,uCACnF;EAGF,IAAI,QAAQ,GACV,WAAW;EAGb,WAAW,OAAO,SAAS;CAC7B;CAEA,OAAO,UAAU;AACnB;;;;;;;AAQA,SAAS,WAAc,OAAmB;CACxC,IAAI,OAAO,UAAU,UACnB,OAAO,KAAK,MAAM,KAAK;CAGzB,OAAO;AACT;;;;;;;;;AAUA,SAAS,gBAAgB,UAA2B;CAElD,MAAM,QAAQ,KADA,OAAO,aAAa,WAAW,OAAO,QAAQ,IAAK;CAGjE,IAAI,QAAQ,GACV,OAAO;CAGT,IAAI,QAAQ,GACV,OAAO;CAGT,OAAO;AACT;;;;;;;;;;;;;;;;;;;;;AAsBA,IAAM,gBAAN,MAA2C;CAoBzC,AAAO,YAAY,SAA+B;EAChD,MAAM,QAAQ,QAAQ,SAAS;EAE/B,IAAI,CAAC,gBAAgB,KAAK,KAAK,GAC7B,MAAM,IAAI,UACR,sCAAsC,MAAM,oCAC9C;EAGF,KAAK,QAAQ;EACb,KAAK,aAAa,QAAQ,cAAc;EACxC,KAAK,QAAQ,QAAQ,SAAS;EAC9B,KAAK,eAAe,QAAQ,gBAAgB;EAE5C,IAAI,QAAQ,QAAQ;GAClB,IAAI,OAAO,QAAQ,OAAO,UAAU,YAClC,MAAM,IAAI,UACR,8GACF;GAGF,KAAK,gBAAgB,QAAQ,QAAQ,QAAQ,MAAM;GAEnD;EACF;EAEA,IAAI,QAAQ,kBAAkB;GAG5B,KAAK,gBAAgB,cAAc,QAAQ,gBAAgB;GAE3D;EACF;EAEA,MAAM,IAAI,UACR,0EACF;CACF;;;;;CAMA,AAAQ,SAAgC;EACtC,OAAO,KAAK;CACd;;;;;;;;;;;;;CAcA,MAAa,OACX,KACA,OACA,QACA,MACe;EAGf,OAAM,MAFe,KAAK,OAAO,EAErB,CAAC,MACX,eAAe,KAAK,MAAM;;;;;oCAM1B;GAAC;GAAK,KAAK,UAAU,KAAK;GAAG,cAAc,MAAM;GAAG,QAAQ,CAAC;EAAC,CAChE;CACF;;;;;;;;;;;;;;;;;;;CAoBA,MAAa,MACX,QACA,SACqD;EACrD,MAAM,SAAS,MAAM,KAAK,OAAO;EAMjC,MAAM,SAAoB,CALN,cAAc,MAKG,GAAG,QAAQ,IAAI;EACpD,MAAM,aAAuB,CAAC;EAE9B,IAAI,QAAQ,cAAc,QAAW;GACnC,OAAO,KAAK,IAAI,QAAQ,SAAS;GACjC,WAAW,KAAK,kCAAkC,OAAO,QAAQ;EACnE;EAEA,IAAI,QAAQ,SAAS,UAAa,QAAQ,KAAK,SAAS,GAAG;GACzD,OAAO,KAAK,QAAQ,IAAI;GACxB,WAAW,KAAK,YAAY,OAAO,OAAO,SAAS;EACrD;EAEA,MAAM,QAAQ,WAAW,SAAS,IAAI,SAAS,WAAW,KAAK,OAAO,MAAM;EAE5E,MAAM,EAAE,SAAS,MAAM,OAAO,MAC5B;cACQ,KAAK,MAAM;SAChB,MAAM;;kBAGT,MACF;EAEA,OAAQ,KAAmC,KAAK,SAAS;GACvD,KAAK,IAAI;GACT,OAAO,WAAc,IAAI,KAAK;GAC9B,OAAO,gBAAgB,IAAI,QAAQ;EACrC,EAAE;CACJ;;;;;;;;;CAUA,MAAa,gBAAgB,WAAkC;EAC7D,MAAM,SAAS,MAAM,KAAK,OAAO;EAEjC,MAAM,UAAU,UACb,QAAQ,OAAO,MAAM,CAAC,CACtB,QAAQ,MAAM,KAAK,CAAC,CACpB,QAAQ,MAAM,KAAK;EAEtB,MAAM,OAAO,MACX,eAAe,KAAK,MAAM;mDAE1B,CAAC,WAAW,GAAG,QAAQ,GAAG,CAC5B;CACF;;;;;;;;;;;;;;;;;;;;;;;;CAyBA,AAAO,SAAiB;EACtB,MAAM,QAAQ;GACZ;GACA,8BAA8B,KAAK,MAAM;GACzC;GACA;GACA,uBAAuB,KAAK,WAAW;GACvC;GACA;GACA,kCAAkC,KAAK,MAAM;GAC7C,QAAQ,KAAK,MAAM;EACrB;EAEA,IAAI,KAAK,UAAU,QACjB,MAAM,KACJ,kCAAkC,KAAK,MAAM,aAC7C,QAAQ,KAAK,MAAM,2CACrB;OACK,IAAI,KAAK,UAAU,WACxB,MAAM,KACJ,kCAAkC,KAAK,MAAM,aAC7C,QAAQ,KAAK,MAAM,+CACnB,mBAAmB,KAAK,aAAa,GACvC;OAEA,MAAM,KACJ,uEACA,sEACF;EAGF,OAAO,MAAM,KAAK,IAAI;CACxB;;;;;;;CAQA,AAAO,eAAuB;EAC5B,OAAO,KAAK,OAAO;CACrB;AACF;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAsDA,SAAgB,cAAc,SAAsD;CAClF,OAAO,IAAI,cAAc,OAAO;AAClC"}

package/esm/rag/store/vector-store.contract.d.mts

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/store/vector-store.contract.d.ts
+//#region ../ai/src/rag/store/vector-store.contract.d.ts
 /**
  * The three vector operations the RAG pipeline needs, expressed as a thin
  * structural narrowing of the `@warlock.js/cache` `CacheDriver` surface

package/esm/rag/store/vector-store.contract.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"vector-store.contract.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/store/vector-store.contract.ts"],"mappings":";;AAWA;;;;;;;;;;UAAiB,WAAA;EAMK;;;;;EAApB,MAAA,CAAO,GAAA,UAAa,KAAA,WAAgB,MAAA,YAAkB,IAAA,cAAkB,OAAA;EAMtE;;;;EADF,KAAA,IACE,MAAA,YACA,OAAA;IAAW,IAAA;IAAc,SAAA;IAAoB,IAAA;EAAA,IAC5C,OAAA;IAAU,GAAA;IAAa,KAAA,EAAO,CAAA;IAAG,KAAA;EAAA;EAEO;EAA3C,eAAA,CAAgB,SAAA,WAAoB,OAAA;AAAA"}
+{"version":3,"file":"vector-store.contract.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/store/vector-store.contract.ts"],"mappings":";;AAWA;;;;;;;;;;UAAiB,WAAA;EAMK;;;;;EAApB,MAAA,CAAO,GAAA,UAAa,KAAA,WAAgB,MAAA,YAAkB,IAAA,cAAkB,OAAA;EAMtE;;;;EADF,KAAA,IACE,MAAA,YACA,OAAA;IAAW,IAAA;IAAc,SAAA;IAAoB,IAAA;EAAA,IAC5C,OAAA;IAAU,GAAA;IAAa,KAAA,EAAO,CAAA;IAAG,KAAA;EAAA;EAEO;EAA3C,eAAA,CAAgB,SAAA,WAAoB,OAAA;AAAA"}

package/esm/rag/transforms/multi-query.d.mts

@@ -1,6 +1,6 @@
 import { ModelContract } from "../../contracts/model.contract.mjs";
 
-//#region ../@warlock.js/ai/src/rag/transforms/multi-query.d.ts
+//#region ../ai/src/rag/transforms/multi-query.d.ts
 /** Options for {@link multiQuery}. */
 type MultiQueryOptions = {
   /** How many alternative phrasings to request. Default 3. */n?: number; /** Include the original query in the returned list. Default true. */

package/esm/rag/transforms/multi-query.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"multi-query.d.mts","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/transforms/multi-query.ts"],"mappings":";;;;KAGY,iBAAA;EAAA,4DAEV,CAAA;EAEA,eAAe;AAAA;AAkBjB;;;;;;;;;;;;;;;AAAA,iBAAsB,UAAA,CACpB,KAAA,EAAO,aAAA,EACP,KAAA,UACA,OAAA,GAAS,iBAAA,GACR,OAAA"}
+{"version":3,"file":"multi-query.d.mts","names":[],"sources":["../../../../../../../../ai/src/rag/transforms/multi-query.ts"],"mappings":";;;;KAGY,iBAAA;EAAA,4DAEV,CAAA;EAEA,eAAe;AAAA;AAkBjB;;;;;;;;;;;;;;;AAAA,iBAAsB,UAAA,CACpB,KAAA,EAAO,aAAA,EACP,KAAA,UACA,OAAA,GAAS,iBAAA,GACR,OAAA"}

package/esm/rag/transforms/multi-query.mjs

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/rag/transforms/multi-query.ts
+//#region ../ai/src/rag/transforms/multi-query.ts
 /**
 * Multi-query expansion (A4) — ask a model for several alternative
 * phrasings of `query`, so retrieval covers vocabulary the original

package/esm/rag/transforms/multi-query.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"multi-query.mjs","names":[],"sources":["../../../../../../../../@warlock.js/ai/src/rag/transforms/multi-query.ts"],"sourcesContent":["import type { ModelContract } from \"../../contracts/model.contract\";\n\n/** Options for {@link multiQuery}. */\nexport type MultiQueryOptions = {\n  /** How many alternative phrasings to request. Default 3. */\n  n?: number;\n  /** Include the original query in the returned list. Default true. */\n  includeOriginal?: boolean;\n};\n\n/**\n * Multi-query expansion (A4) — ask a model for several alternative\n * phrasings of `query`, so retrieval covers vocabulary the original\n * wording missed (synonyms, specificity, rephrasings). Pair the variants\n * with {@link hybridRank} / a vector search and fuse the per-variant hits.\n *\n * Deterministic, dependency-light parsing: the model is asked for one\n * query per line; bullets / numbering are stripped, blanks dropped, and\n * the set is de-duplicated. Returns the original (unless opted out) plus\n * up to `n` variants.\n *\n * @example\n * const queries = await multiQuery(model, \"how do I cancel?\", { n: 3 });\n * // → [\"how do I cancel?\", \"cancel my subscription\", \"end my plan\", ...]\n */\nexport async function multiQuery(\n  model: ModelContract,\n  query: string,\n  options: MultiQueryOptions = {},\n): Promise<string[]> {\n  const n = options.n ?? 3;\n  const includeOriginal = options.includeOriginal ?? true;\n\n  const prompt =\n    `Rewrite the following search query into ${n} alternative phrasings that would ` +\n    `retrieve relevant documents. Output ONE query per line, no numbering or commentary.\\n\\n` +\n    `Query: ${query}`;\n\n  const response = await model.complete([{ role: \"user\", content: prompt }]);\n\n  const variants = response.content\n    .split(\"\\n\")\n    .map(line => line.replace(/^\\s*(?:[-*•]|\\d+[.)])\\s*/, \"\").trim())\n    .filter(Boolean);\n\n  const seen = new Set<string>();\n  const out: string[] = [];\n\n  const add = (q: string) => {\n    const key = q.toLowerCase();\n    if (!seen.has(key)) {\n      seen.add(key);\n      out.push(q);\n    }\n  };\n\n  if (includeOriginal) add(query);\n  for (const variant of variants.slice(0, n)) add(variant);\n\n  return out;\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAyBA,eAAsB,WACpB,OACA,OACA,UAA6B,CAAC,GACX;CACnB,MAAM,IAAI,QAAQ,KAAK;CACvB,MAAM,kBAAkB,QAAQ,mBAAmB;CAEnD,MAAM,SACJ,2CAA2C,EAAE,kIAEnC;CAIZ,MAAM,YAAW,MAFM,MAAM,SAAS,CAAC;EAAE,MAAM;EAAQ,SAAS;CAAO,CAAC,CAAC,EAEhD,CAAC,QACvB,MAAM,IAAI,CAAC,CACX,KAAI,SAAQ,KAAK,QAAQ,4BAA4B,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAChE,OAAO,OAAO;CAEjB,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,MAAgB,CAAC;CAEvB,MAAM,OAAO,MAAc;EACzB,MAAM,MAAM,EAAE,YAAY;EAC1B,IAAI,CAAC,KAAK,IAAI,GAAG,GAAG;GAClB,KAAK,IAAI,GAAG;GACZ,IAAI,KAAK,CAAC;EACZ;CACF;CAEA,IAAI,iBAAiB,IAAI,KAAK;CAC9B,KAAK,MAAM,WAAW,SAAS,MAAM,GAAG,CAAC,GAAG,IAAI,OAAO;CAEvD,OAAO;AACT"}
+{"version":3,"file":"multi-query.mjs","names":[],"sources":["../../../../../../../../ai/src/rag/transforms/multi-query.ts"],"sourcesContent":["import type { ModelContract } from \"../../contracts/model.contract\";\n\n/** Options for {@link multiQuery}. */\nexport type MultiQueryOptions = {\n  /** How many alternative phrasings to request. Default 3. */\n  n?: number;\n  /** Include the original query in the returned list. Default true. */\n  includeOriginal?: boolean;\n};\n\n/**\n * Multi-query expansion (A4) — ask a model for several alternative\n * phrasings of `query`, so retrieval covers vocabulary the original\n * wording missed (synonyms, specificity, rephrasings). Pair the variants\n * with {@link hybridRank} / a vector search and fuse the per-variant hits.\n *\n * Deterministic, dependency-light parsing: the model is asked for one\n * query per line; bullets / numbering are stripped, blanks dropped, and\n * the set is de-duplicated. Returns the original (unless opted out) plus\n * up to `n` variants.\n *\n * @example\n * const queries = await multiQuery(model, \"how do I cancel?\", { n: 3 });\n * // → [\"how do I cancel?\", \"cancel my subscription\", \"end my plan\", ...]\n */\nexport async function multiQuery(\n  model: ModelContract,\n  query: string,\n  options: MultiQueryOptions = {},\n): Promise<string[]> {\n  const n = options.n ?? 3;\n  const includeOriginal = options.includeOriginal ?? true;\n\n  const prompt =\n    `Rewrite the following search query into ${n} alternative phrasings that would ` +\n    `retrieve relevant documents. Output ONE query per line, no numbering or commentary.\\n\\n` +\n    `Query: ${query}`;\n\n  const response = await model.complete([{ role: \"user\", content: prompt }]);\n\n  const variants = response.content\n    .split(\"\\n\")\n    .map(line => line.replace(/^\\s*(?:[-*•]|\\d+[.)])\\s*/, \"\").trim())\n    .filter(Boolean);\n\n  const seen = new Set<string>();\n  const out: string[] = [];\n\n  const add = (q: string) => {\n    const key = q.toLowerCase();\n    if (!seen.has(key)) {\n      seen.add(key);\n      out.push(q);\n    }\n  };\n\n  if (includeOriginal) add(query);\n  for (const variant of variants.slice(0, n)) add(variant);\n\n  return out;\n}\n"],"mappings":";;;;;;;;;;;;;;;;AAyBA,eAAsB,WACpB,OACA,OACA,UAA6B,CAAC,GACX;CACnB,MAAM,IAAI,QAAQ,KAAK;CACvB,MAAM,kBAAkB,QAAQ,mBAAmB;CAEnD,MAAM,SACJ,2CAA2C,EAAE,kIAEnC;CAIZ,MAAM,YAAW,MAFM,MAAM,SAAS,CAAC;EAAE,MAAM;EAAQ,SAAS;CAAO,CAAC,CAAC,EAEhD,CAAC,QACvB,MAAM,IAAI,CAAC,CACX,KAAI,SAAQ,KAAK,QAAQ,4BAA4B,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAChE,OAAO,OAAO;CAEjB,MAAM,uBAAO,IAAI,IAAY;CAC7B,MAAM,MAAgB,CAAC;CAEvB,MAAM,OAAO,MAAc;EACzB,MAAM,MAAM,EAAE,YAAY;EAC1B,IAAI,CAAC,KAAK,IAAI,GAAG,GAAG;GAClB,KAAK,IAAI,GAAG;GACZ,IAAI,KAAK,CAAC;EACZ;CACF;CAEA,IAAI,iBAAiB,IAAI,KAAK;CAC9B,KAAK,MAAM,WAAW,SAAS,MAAM,GAAG,CAAC,GAAG,IAAI,OAAO;CAEvD,OAAO;AACT"}

package/esm/security/outbound-policy.d.mts

@@ -1,6 +1,6 @@
 import { OutboundPolicy, ResolvedOutboundPolicy } from "./outbound-policy.type.mjs";
 
-//#region ../@warlock.js/ai/src/security/outbound-policy.d.ts
+//#region ../ai/src/security/outbound-policy.d.ts
 /**
  * Fill an {@link OutboundPolicy} with strict defaults: https-only,
  * private-IP deny on, 10s timeout, 5 MiB cap, global `fetch`. Idempotent

package/esm/security/outbound-policy.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"outbound-policy.d.mts","names":[],"sources":["../../../../../../../@warlock.js/ai/src/security/outbound-policy.ts"],"mappings":";;;;;AAmBA;;;iBAAgB,qBAAA,CACd,MAAA,GAAQ,cAAA,GACP,sBAAsB;;;;;AAAA;AAiCzB;;iBAAsB,gBAAA,CACpB,MAAA,UACA,MAAA,EAAQ,sBAAA,GACP,OAAA,CAAQ,GAAA;;;;;;;;iBAiGW,YAAA,CACpB,MAAA,UACA,WAAA,EAAa,cAAA,EACb,IAAA,GAAO,WAAA,GACN,OAAA,CAAQ,QAAA;;;;AArGG;AAiGd;;iBAkCsB,cAAA,CACpB,QAAA,EAAU,QAAA,EACV,QAAA,WACC,OAAO;;;;;;iBAgDY,mBAAA,CACpB,MAAA,UACA,WAAA,EAAa,cAAA,EACb,IAAA,GAAO,WAAA,GACN,OAAA;EAAU,EAAA;EAAa,MAAA;EAAgB,UAAA;EAAoB,IAAA;AAAA"}
+{"version":3,"file":"outbound-policy.d.mts","names":[],"sources":["../../../../../../../ai/src/security/outbound-policy.ts"],"mappings":";;;;;AAmBA;;;iBAAgB,qBAAA,CACd,MAAA,GAAQ,cAAA,GACP,sBAAsB;;;;;AAAA;AAiCzB;;iBAAsB,gBAAA,CACpB,MAAA,UACA,MAAA,EAAQ,sBAAA,GACP,OAAA,CAAQ,GAAA;;;;;;;;iBAiGW,YAAA,CACpB,MAAA,UACA,WAAA,EAAa,cAAA,EACb,IAAA,GAAO,WAAA,GACN,OAAA,CAAQ,QAAA;;;;AArGG;AAiGd;;iBAkCsB,cAAA,CACpB,QAAA,EAAU,QAAA,EACV,QAAA,WACC,OAAO;;;;;;iBAgDY,mBAAA,CACpB,MAAA,UACA,WAAA,EAAa,cAAA,EACb,IAAA,GAAO,WAAA,GACN,OAAA;EAAU,EAAA;EAAa,MAAA;EAAgB,UAAA;EAAoB,IAAA;AAAA"}

package/esm/security/outbound-policy.mjs

@@ -4,7 +4,7 @@ import { isPrivateOrReservedIp } from "./private-ip.mjs";
 import { lookup } from "node:dns/promises";
 import { isIP } from "node:net";
 
-//#region ../@warlock.js/ai/src/security/outbound-policy.ts
+//#region ../ai/src/security/outbound-policy.ts
 /** 5 MiB — default cap on an outbound response body. */
 const DEFAULT_MAX_BYTES = 5 * 1024 * 1024;
 /** 10s — default per-request timeout. */

package/esm/security/outbound-policy.mjs.map

@@ -1 +1 @@
-{"version":3,"file":"outbound-policy.mjs","names":[],"sources":["../../../../../../../@warlock.js/ai/src/security/outbound-policy.ts"],"sourcesContent":["import { lookup } from \"node:dns/promises\";\nimport { isIP } from \"node:net\";\nimport { OutboundPolicyError } from \"../errors\";\nimport { isPrivateOrReservedIp } from \"./private-ip\";\nimport type {\n  OutboundPolicy,\n  ResolvedOutboundPolicy,\n} from \"./outbound-policy.type\";\n\n/** 5 MiB — default cap on an outbound response body. */\nconst DEFAULT_MAX_BYTES = 5 * 1024 * 1024;\n/** 10s — default per-request timeout. */\nconst DEFAULT_TIMEOUT_MS = 10_000;\n\n/**\n * Fill an {@link OutboundPolicy} with strict defaults: https-only,\n * private-IP deny on, 10s timeout, 5 MiB cap, global `fetch`. Idempotent\n * — resolving an already-resolved policy yields the same shape.\n */\nexport function resolveOutboundPolicy(\n  policy: OutboundPolicy = {},\n): ResolvedOutboundPolicy {\n  return {\n    allowedSchemes: policy.allowedSchemes ?? [\"https\"],\n    hostAllowlist: policy.hostAllowlist,\n    denyPrivateIPsAfterDNS: policy.denyPrivateIPsAfterDNS ?? true,\n    maxBytes: policy.maxBytes ?? DEFAULT_MAX_BYTES,\n    timeoutMs: policy.timeoutMs ?? DEFAULT_TIMEOUT_MS,\n    signal: policy.signal,\n    fetch: policy.fetch ?? globalThis.fetch,\n  };\n}\n\n/** Strip the `[ ]` IPv6 brackets `URL.hostname` keeps. */\nfunction stripBrackets(host: string): string {\n  return host.startsWith(\"[\") && host.endsWith(\"]\") ? host.slice(1, -1) : host;\n}\n\n/** Whether `host` equals or is a subdomain of any allowlist entry. */\nfunction hostAllowed(host: string, allowlist: string[]): boolean {\n  const lower = host.toLowerCase();\n  return allowlist.some(entry => {\n    const e = entry.toLowerCase();\n    return lower === e || lower.endsWith(`.${e}`);\n  });\n}\n\n/**\n * Validate a URL against the policy BEFORE any network call: scheme\n * allowlist, host allowlist, and (when enabled) a DNS resolution that\n * rejects private / loopback / link-local / metadata addresses — the SSRF\n * guard. Returns the parsed `URL` on success; throws\n * {@link OutboundPolicyError} otherwise.\n */\nexport async function assertUrlAllowed(\n  rawUrl: string,\n  policy: ResolvedOutboundPolicy,\n): Promise<URL> {\n  let url: URL;\n  try {\n    url = new URL(rawUrl);\n  } catch {\n    throw new OutboundPolicyError(`outbound request blocked — invalid URL: ${rawUrl}`, {\n      context: { url: rawUrl },\n    });\n  }\n\n  const scheme = url.protocol.replace(/:$/, \"\").toLowerCase();\n  if (!policy.allowedSchemes.some(s => s.toLowerCase() === scheme)) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — scheme \"${scheme}\" is not allowed (allowed: ${policy.allowedSchemes.join(\", \")})`,\n      { context: { url: rawUrl, scheme } },\n    );\n  }\n\n  const host = stripBrackets(url.hostname);\n\n  if (policy.hostAllowlist && !hostAllowed(host, policy.hostAllowlist)) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — host \"${host}\" is not in the allowlist`,\n      { context: { url: rawUrl, host } },\n    );\n  }\n\n  if (policy.denyPrivateIPsAfterDNS) {\n    await assertHostNotPrivate(host, rawUrl);\n  }\n\n  return url;\n}\n\n/**\n * Reject when `host` is — or resolves to — a private / reserved address.\n * IP literals are checked directly; hostnames are resolved via DNS and\n * every returned address is checked (a public name pointing inward is\n * caught). A resolution failure fails closed.\n */\nasync function assertHostNotPrivate(host: string, rawUrl: string): Promise<void> {\n  if (isIP(host) !== 0) {\n    if (isPrivateOrReservedIp(host)) {\n      throw new OutboundPolicyError(\n        `outbound request blocked — \"${host}\" is a private/reserved address`,\n        { context: { url: rawUrl, address: host } },\n      );\n    }\n    return;\n  }\n\n  let addresses: Array<{ address: string }>;\n  try {\n    addresses = await lookup(host, { all: true });\n  } catch (cause) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — could not resolve host \"${host}\" to verify it is public`,\n      { cause, context: { url: rawUrl, host } },\n    );\n  }\n\n  for (const { address } of addresses) {\n    if (isPrivateOrReservedIp(address)) {\n      throw new OutboundPolicyError(\n        `outbound request blocked — host \"${host}\" resolves to a private/reserved address (${address})`,\n        { context: { url: rawUrl, host, address } },\n      );\n    }\n  }\n}\n\n/** Merge the internal timeout signal with an optional caller signal. */\nfunction mergeSignals(\n  timeout: AbortSignal,\n  external?: AbortSignal,\n): AbortSignal {\n  if (!external) return timeout;\n\n  const controller = new AbortController();\n  const abort = (from: AbortSignal) => controller.abort(from.reason);\n\n  if (timeout.aborted) abort(timeout);\n  else timeout.addEventListener(\"abort\", () => abort(timeout), { once: true });\n\n  if (external.aborted) abort(external);\n  else external.addEventListener(\"abort\", () => abort(external), { once: true });\n\n  return controller.signal;\n}\n\n/**\n * Policy-guarded `fetch`: validates the URL ({@link assertUrlAllowed}),\n * then performs the request with the policy's timeout and (optional)\n * caller signal merged. Returns the raw `Response` — read its body via\n * {@link readTextCapped} to enforce `maxBytes`. Throws\n * {@link OutboundPolicyError} on a policy violation or timeout.\n */\nexport async function guardedFetch(\n  rawUrl: string,\n  policyInput: OutboundPolicy,\n  init?: RequestInit,\n): Promise<Response> {\n  const policy = resolveOutboundPolicy(policyInput);\n  const url = await assertUrlAllowed(rawUrl, policy);\n\n  const timeoutController = new AbortController();\n  const timer = setTimeout(() => {\n    timeoutController.abort(\n      new OutboundPolicyError(\n        `outbound request timed out after ${policy.timeoutMs}ms`,\n        { context: { url: rawUrl, timeoutMs: policy.timeoutMs } },\n      ),\n    );\n  }, policy.timeoutMs);\n\n  try {\n    return await policy.fetch(url, {\n      ...init,\n      signal: mergeSignals(timeoutController.signal, policy.signal),\n    });\n  } finally {\n    clearTimeout(timer);\n  }\n}\n\n/**\n * Read a response body as UTF-8 text with a hard byte cap. A declared\n * `content-length` over the cap fails fast; otherwise the stream is read\n * chunk-by-chunk and aborted the moment the running total exceeds\n * `maxBytes`. Throws {@link OutboundPolicyError} on overflow.\n */\nexport async function readTextCapped(\n  response: Response,\n  maxBytes: number,\n): Promise<string> {\n  const declared = Number(response.headers.get(\"content-length\"));\n  if (Number.isFinite(declared) && declared > maxBytes) {\n    throw new OutboundPolicyError(\n      `outbound response body too large — declared ${declared} bytes exceeds the ${maxBytes}-byte cap`,\n      { context: { declared, maxBytes } },\n    );\n  }\n\n  if (!response.body) {\n    const text = await response.text();\n    if (Buffer.byteLength(text) > maxBytes) {\n      throw new OutboundPolicyError(\n        `outbound response body exceeded the ${maxBytes}-byte cap`,\n        { context: { maxBytes } },\n      );\n    }\n    return text;\n  }\n\n  const reader = response.body.getReader();\n  const chunks: Uint8Array[] = [];\n  let total = 0;\n\n  for (;;) {\n    const { done, value } = await reader.read();\n    if (done) break;\n    if (!value) continue;\n\n    total += value.byteLength;\n    if (total > maxBytes) {\n      await reader.cancel();\n      throw new OutboundPolicyError(\n        `outbound response body exceeded the ${maxBytes}-byte cap`,\n        { context: { maxBytes } },\n      );\n    }\n    chunks.push(value);\n  }\n\n  return Buffer.concat(chunks).toString(\"utf8\");\n}\n\n/**\n * Convenience: {@link guardedFetch} + {@link readTextCapped}. Returns the\n * response status alongside the (capped) body text so callers can shape\n * their own not-OK error. The body is only read when the response is OK.\n */\nexport async function fetchTextWithPolicy(\n  rawUrl: string,\n  policyInput: OutboundPolicy,\n  init?: RequestInit,\n): Promise<{ ok: boolean; status: number; statusText: string; text: string }> {\n  const policy = resolveOutboundPolicy(policyInput);\n  const response = await guardedFetch(rawUrl, policy, init);\n\n  return {\n    ok: response.ok,\n    status: response.status,\n    statusText: response.statusText,\n    text: response.ok ? await readTextCapped(response, policy.maxBytes) : \"\",\n  };\n}\n"],"mappings":";;;;;;;;AAUA,MAAM,oBAAoB,IAAI,OAAO;;AAErC,MAAM,qBAAqB;;;;;;AAO3B,SAAgB,sBACd,SAAyB,CAAC,GACF;CACxB,OAAO;EACL,gBAAgB,OAAO,kBAAkB,CAAC,OAAO;EACjD,eAAe,OAAO;EACtB,wBAAwB,OAAO,0BAA0B;EACzD,UAAU,OAAO,YAAY;EAC7B,WAAW,OAAO,aAAa;EAC/B,QAAQ,OAAO;EACf,OAAO,OAAO,SAAS,WAAW;CACpC;AACF;;AAGA,SAAS,cAAc,MAAsB;CAC3C,OAAO,KAAK,WAAW,GAAG,KAAK,KAAK,SAAS,GAAG,IAAI,KAAK,MAAM,GAAG,EAAE,IAAI;AAC1E;;AAGA,SAAS,YAAY,MAAc,WAA8B;CAC/D,MAAM,QAAQ,KAAK,YAAY;CAC/B,OAAO,UAAU,MAAK,UAAS;EAC7B,MAAM,IAAI,MAAM,YAAY;EAC5B,OAAO,UAAU,KAAK,MAAM,SAAS,IAAI,GAAG;CAC9C,CAAC;AACH;;;;;;;;AASA,eAAsB,iBACpB,QACA,QACc;CACd,IAAI;CACJ,IAAI;EACF,MAAM,IAAI,IAAI,MAAM;CACtB,QAAQ;EACN,MAAM,IAAI,oBAAoB,2CAA2C,UAAU,EACjF,SAAS,EAAE,KAAK,OAAO,EACzB,CAAC;CACH;CAEA,MAAM,SAAS,IAAI,SAAS,QAAQ,MAAM,EAAE,CAAC,CAAC,YAAY;CAC1D,IAAI,CAAC,OAAO,eAAe,MAAK,MAAK,EAAE,YAAY,MAAM,MAAM,GAC7D,MAAM,IAAI,oBACR,sCAAsC,OAAO,6BAA6B,OAAO,eAAe,KAAK,IAAI,EAAE,IAC3G,EAAE,SAAS;EAAE,KAAK;EAAQ;CAAO,EAAE,CACrC;CAGF,MAAM,OAAO,cAAc,IAAI,QAAQ;CAEvC,IAAI,OAAO,iBAAiB,CAAC,YAAY,MAAM,OAAO,aAAa,GACjE,MAAM,IAAI,oBACR,oCAAoC,KAAK,4BACzC,EAAE,SAAS;EAAE,KAAK;EAAQ;CAAK,EAAE,CACnC;CAGF,IAAI,OAAO,wBACT,MAAM,qBAAqB,MAAM,MAAM;CAGzC,OAAO;AACT;;;;;;;AAQA,eAAe,qBAAqB,MAAc,QAA+B;CAC/E,IAAI,KAAK,IAAI,MAAM,GAAG;EACpB,IAAI,sBAAsB,IAAI,GAC5B,MAAM,IAAI,oBACR,+BAA+B,KAAK,kCACpC,EAAE,SAAS;GAAE,KAAK;GAAQ,SAAS;EAAK,EAAE,CAC5C;EAEF;CACF;CAEA,IAAI;CACJ,IAAI;EACF,YAAY,MAAM,OAAO,MAAM,EAAE,KAAK,KAAK,CAAC;CAC9C,SAAS,OAAO;EACd,MAAM,IAAI,oBACR,sDAAsD,KAAK,2BAC3D;GAAE;GAAO,SAAS;IAAE,KAAK;IAAQ;GAAK;EAAE,CAC1C;CACF;CAEA,KAAK,MAAM,EAAE,aAAa,WACxB,IAAI,sBAAsB,OAAO,GAC/B,MAAM,IAAI,oBACR,oCAAoC,KAAK,4CAA4C,QAAQ,IAC7F,EAAE,SAAS;EAAE,KAAK;EAAQ;EAAM;CAAQ,EAAE,CAC5C;AAGN;;AAGA,SAAS,aACP,SACA,UACa;CACb,IAAI,CAAC,UAAU,OAAO;CAEtB,MAAM,aAAa,IAAI,gBAAgB;CACvC,MAAM,SAAS,SAAsB,WAAW,MAAM,KAAK,MAAM;CAEjE,IAAI,QAAQ,SAAS,MAAM,OAAO;MAC7B,QAAQ,iBAAiB,eAAe,MAAM,OAAO,GAAG,EAAE,MAAM,KAAK,CAAC;CAE3E,IAAI,SAAS,SAAS,MAAM,QAAQ;MAC/B,SAAS,iBAAiB,eAAe,MAAM,QAAQ,GAAG,EAAE,MAAM,KAAK,CAAC;CAE7E,OAAO,WAAW;AACpB;;;;;;;;AASA,eAAsB,aACpB,QACA,aACA,MACmB;CACnB,MAAM,SAAS,sBAAsB,WAAW;CAChD,MAAM,MAAM,MAAM,iBAAiB,QAAQ,MAAM;CAEjD,MAAM,oBAAoB,IAAI,gBAAgB;CAC9C,MAAM,QAAQ,iBAAiB;EAC7B,kBAAkB,MAChB,IAAI,oBACF,oCAAoC,OAAO,UAAU,KACrD,EAAE,SAAS;GAAE,KAAK;GAAQ,WAAW,OAAO;EAAU,EAAE,CAC1D,CACF;CACF,GAAG,OAAO,SAAS;CAEnB,IAAI;EACF,OAAO,MAAM,OAAO,MAAM,KAAK;GAC7B,GAAG;GACH,QAAQ,aAAa,kBAAkB,QAAQ,OAAO,MAAM;EAC9D,CAAC;CACH,UAAU;EACR,aAAa,KAAK;CACpB;AACF;;;;;;;AAQA,eAAsB,eACpB,UACA,UACiB;CACjB,MAAM,WAAW,OAAO,SAAS,QAAQ,IAAI,gBAAgB,CAAC;CAC9D,IAAI,OAAO,SAAS,QAAQ,KAAK,WAAW,UAC1C,MAAM,IAAI,oBACR,+CAA+C,SAAS,qBAAqB,SAAS,YACtF,EAAE,SAAS;EAAE;EAAU;CAAS,EAAE,CACpC;CAGF,IAAI,CAAC,SAAS,MAAM;EAClB,MAAM,OAAO,MAAM,SAAS,KAAK;EACjC,IAAI,OAAO,WAAW,IAAI,IAAI,UAC5B,MAAM,IAAI,oBACR,uCAAuC,SAAS,YAChD,EAAE,SAAS,EAAE,SAAS,EAAE,CAC1B;EAEF,OAAO;CACT;CAEA,MAAM,SAAS,SAAS,KAAK,UAAU;CACvC,MAAM,SAAuB,CAAC;CAC9B,IAAI,QAAQ;CAEZ,SAAS;EACP,MAAM,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK;EAC1C,IAAI,MAAM;EACV,IAAI,CAAC,OAAO;EAEZ,SAAS,MAAM;EACf,IAAI,QAAQ,UAAU;GACpB,MAAM,OAAO,OAAO;GACpB,MAAM,IAAI,oBACR,uCAAuC,SAAS,YAChD,EAAE,SAAS,EAAE,SAAS,EAAE,CAC1B;EACF;EACA,OAAO,KAAK,KAAK;CACnB;CAEA,OAAO,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM;AAC9C;;;;;;AAOA,eAAsB,oBACpB,QACA,aACA,MAC4E;CAC5E,MAAM,SAAS,sBAAsB,WAAW;CAChD,MAAM,WAAW,MAAM,aAAa,QAAQ,QAAQ,IAAI;CAExD,OAAO;EACL,IAAI,SAAS;EACb,QAAQ,SAAS;EACjB,YAAY,SAAS;EACrB,MAAM,SAAS,KAAK,MAAM,eAAe,UAAU,OAAO,QAAQ,IAAI;CACxE;AACF"}
+{"version":3,"file":"outbound-policy.mjs","names":[],"sources":["../../../../../../../ai/src/security/outbound-policy.ts"],"sourcesContent":["import { lookup } from \"node:dns/promises\";\nimport { isIP } from \"node:net\";\nimport { OutboundPolicyError } from \"../errors\";\nimport { isPrivateOrReservedIp } from \"./private-ip\";\nimport type {\n  OutboundPolicy,\n  ResolvedOutboundPolicy,\n} from \"./outbound-policy.type\";\n\n/** 5 MiB — default cap on an outbound response body. */\nconst DEFAULT_MAX_BYTES = 5 * 1024 * 1024;\n/** 10s — default per-request timeout. */\nconst DEFAULT_TIMEOUT_MS = 10_000;\n\n/**\n * Fill an {@link OutboundPolicy} with strict defaults: https-only,\n * private-IP deny on, 10s timeout, 5 MiB cap, global `fetch`. Idempotent\n * — resolving an already-resolved policy yields the same shape.\n */\nexport function resolveOutboundPolicy(\n  policy: OutboundPolicy = {},\n): ResolvedOutboundPolicy {\n  return {\n    allowedSchemes: policy.allowedSchemes ?? [\"https\"],\n    hostAllowlist: policy.hostAllowlist,\n    denyPrivateIPsAfterDNS: policy.denyPrivateIPsAfterDNS ?? true,\n    maxBytes: policy.maxBytes ?? DEFAULT_MAX_BYTES,\n    timeoutMs: policy.timeoutMs ?? DEFAULT_TIMEOUT_MS,\n    signal: policy.signal,\n    fetch: policy.fetch ?? globalThis.fetch,\n  };\n}\n\n/** Strip the `[ ]` IPv6 brackets `URL.hostname` keeps. */\nfunction stripBrackets(host: string): string {\n  return host.startsWith(\"[\") && host.endsWith(\"]\") ? host.slice(1, -1) : host;\n}\n\n/** Whether `host` equals or is a subdomain of any allowlist entry. */\nfunction hostAllowed(host: string, allowlist: string[]): boolean {\n  const lower = host.toLowerCase();\n  return allowlist.some(entry => {\n    const e = entry.toLowerCase();\n    return lower === e || lower.endsWith(`.${e}`);\n  });\n}\n\n/**\n * Validate a URL against the policy BEFORE any network call: scheme\n * allowlist, host allowlist, and (when enabled) a DNS resolution that\n * rejects private / loopback / link-local / metadata addresses — the SSRF\n * guard. Returns the parsed `URL` on success; throws\n * {@link OutboundPolicyError} otherwise.\n */\nexport async function assertUrlAllowed(\n  rawUrl: string,\n  policy: ResolvedOutboundPolicy,\n): Promise<URL> {\n  let url: URL;\n  try {\n    url = new URL(rawUrl);\n  } catch {\n    throw new OutboundPolicyError(`outbound request blocked — invalid URL: ${rawUrl}`, {\n      context: { url: rawUrl },\n    });\n  }\n\n  const scheme = url.protocol.replace(/:$/, \"\").toLowerCase();\n  if (!policy.allowedSchemes.some(s => s.toLowerCase() === scheme)) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — scheme \"${scheme}\" is not allowed (allowed: ${policy.allowedSchemes.join(\", \")})`,\n      { context: { url: rawUrl, scheme } },\n    );\n  }\n\n  const host = stripBrackets(url.hostname);\n\n  if (policy.hostAllowlist && !hostAllowed(host, policy.hostAllowlist)) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — host \"${host}\" is not in the allowlist`,\n      { context: { url: rawUrl, host } },\n    );\n  }\n\n  if (policy.denyPrivateIPsAfterDNS) {\n    await assertHostNotPrivate(host, rawUrl);\n  }\n\n  return url;\n}\n\n/**\n * Reject when `host` is — or resolves to — a private / reserved address.\n * IP literals are checked directly; hostnames are resolved via DNS and\n * every returned address is checked (a public name pointing inward is\n * caught). A resolution failure fails closed.\n */\nasync function assertHostNotPrivate(host: string, rawUrl: string): Promise<void> {\n  if (isIP(host) !== 0) {\n    if (isPrivateOrReservedIp(host)) {\n      throw new OutboundPolicyError(\n        `outbound request blocked — \"${host}\" is a private/reserved address`,\n        { context: { url: rawUrl, address: host } },\n      );\n    }\n    return;\n  }\n\n  let addresses: Array<{ address: string }>;\n  try {\n    addresses = await lookup(host, { all: true });\n  } catch (cause) {\n    throw new OutboundPolicyError(\n      `outbound request blocked — could not resolve host \"${host}\" to verify it is public`,\n      { cause, context: { url: rawUrl, host } },\n    );\n  }\n\n  for (const { address } of addresses) {\n    if (isPrivateOrReservedIp(address)) {\n      throw new OutboundPolicyError(\n        `outbound request blocked — host \"${host}\" resolves to a private/reserved address (${address})`,\n        { context: { url: rawUrl, host, address } },\n      );\n    }\n  }\n}\n\n/** Merge the internal timeout signal with an optional caller signal. */\nfunction mergeSignals(\n  timeout: AbortSignal,\n  external?: AbortSignal,\n): AbortSignal {\n  if (!external) return timeout;\n\n  const controller = new AbortController();\n  const abort = (from: AbortSignal) => controller.abort(from.reason);\n\n  if (timeout.aborted) abort(timeout);\n  else timeout.addEventListener(\"abort\", () => abort(timeout), { once: true });\n\n  if (external.aborted) abort(external);\n  else external.addEventListener(\"abort\", () => abort(external), { once: true });\n\n  return controller.signal;\n}\n\n/**\n * Policy-guarded `fetch`: validates the URL ({@link assertUrlAllowed}),\n * then performs the request with the policy's timeout and (optional)\n * caller signal merged. Returns the raw `Response` — read its body via\n * {@link readTextCapped} to enforce `maxBytes`. Throws\n * {@link OutboundPolicyError} on a policy violation or timeout.\n */\nexport async function guardedFetch(\n  rawUrl: string,\n  policyInput: OutboundPolicy,\n  init?: RequestInit,\n): Promise<Response> {\n  const policy = resolveOutboundPolicy(policyInput);\n  const url = await assertUrlAllowed(rawUrl, policy);\n\n  const timeoutController = new AbortController();\n  const timer = setTimeout(() => {\n    timeoutController.abort(\n      new OutboundPolicyError(\n        `outbound request timed out after ${policy.timeoutMs}ms`,\n        { context: { url: rawUrl, timeoutMs: policy.timeoutMs } },\n      ),\n    );\n  }, policy.timeoutMs);\n\n  try {\n    return await policy.fetch(url, {\n      ...init,\n      signal: mergeSignals(timeoutController.signal, policy.signal),\n    });\n  } finally {\n    clearTimeout(timer);\n  }\n}\n\n/**\n * Read a response body as UTF-8 text with a hard byte cap. A declared\n * `content-length` over the cap fails fast; otherwise the stream is read\n * chunk-by-chunk and aborted the moment the running total exceeds\n * `maxBytes`. Throws {@link OutboundPolicyError} on overflow.\n */\nexport async function readTextCapped(\n  response: Response,\n  maxBytes: number,\n): Promise<string> {\n  const declared = Number(response.headers.get(\"content-length\"));\n  if (Number.isFinite(declared) && declared > maxBytes) {\n    throw new OutboundPolicyError(\n      `outbound response body too large — declared ${declared} bytes exceeds the ${maxBytes}-byte cap`,\n      { context: { declared, maxBytes } },\n    );\n  }\n\n  if (!response.body) {\n    const text = await response.text();\n    if (Buffer.byteLength(text) > maxBytes) {\n      throw new OutboundPolicyError(\n        `outbound response body exceeded the ${maxBytes}-byte cap`,\n        { context: { maxBytes } },\n      );\n    }\n    return text;\n  }\n\n  const reader = response.body.getReader();\n  const chunks: Uint8Array[] = [];\n  let total = 0;\n\n  for (;;) {\n    const { done, value } = await reader.read();\n    if (done) break;\n    if (!value) continue;\n\n    total += value.byteLength;\n    if (total > maxBytes) {\n      await reader.cancel();\n      throw new OutboundPolicyError(\n        `outbound response body exceeded the ${maxBytes}-byte cap`,\n        { context: { maxBytes } },\n      );\n    }\n    chunks.push(value);\n  }\n\n  return Buffer.concat(chunks).toString(\"utf8\");\n}\n\n/**\n * Convenience: {@link guardedFetch} + {@link readTextCapped}. Returns the\n * response status alongside the (capped) body text so callers can shape\n * their own not-OK error. The body is only read when the response is OK.\n */\nexport async function fetchTextWithPolicy(\n  rawUrl: string,\n  policyInput: OutboundPolicy,\n  init?: RequestInit,\n): Promise<{ ok: boolean; status: number; statusText: string; text: string }> {\n  const policy = resolveOutboundPolicy(policyInput);\n  const response = await guardedFetch(rawUrl, policy, init);\n\n  return {\n    ok: response.ok,\n    status: response.status,\n    statusText: response.statusText,\n    text: response.ok ? await readTextCapped(response, policy.maxBytes) : \"\",\n  };\n}\n"],"mappings":";;;;;;;;AAUA,MAAM,oBAAoB,IAAI,OAAO;;AAErC,MAAM,qBAAqB;;;;;;AAO3B,SAAgB,sBACd,SAAyB,CAAC,GACF;CACxB,OAAO;EACL,gBAAgB,OAAO,kBAAkB,CAAC,OAAO;EACjD,eAAe,OAAO;EACtB,wBAAwB,OAAO,0BAA0B;EACzD,UAAU,OAAO,YAAY;EAC7B,WAAW,OAAO,aAAa;EAC/B,QAAQ,OAAO;EACf,OAAO,OAAO,SAAS,WAAW;CACpC;AACF;;AAGA,SAAS,cAAc,MAAsB;CAC3C,OAAO,KAAK,WAAW,GAAG,KAAK,KAAK,SAAS,GAAG,IAAI,KAAK,MAAM,GAAG,EAAE,IAAI;AAC1E;;AAGA,SAAS,YAAY,MAAc,WAA8B;CAC/D,MAAM,QAAQ,KAAK,YAAY;CAC/B,OAAO,UAAU,MAAK,UAAS;EAC7B,MAAM,IAAI,MAAM,YAAY;EAC5B,OAAO,UAAU,KAAK,MAAM,SAAS,IAAI,GAAG;CAC9C,CAAC;AACH;;;;;;;;AASA,eAAsB,iBACpB,QACA,QACc;CACd,IAAI;CACJ,IAAI;EACF,MAAM,IAAI,IAAI,MAAM;CACtB,QAAQ;EACN,MAAM,IAAI,oBAAoB,2CAA2C,UAAU,EACjF,SAAS,EAAE,KAAK,OAAO,EACzB,CAAC;CACH;CAEA,MAAM,SAAS,IAAI,SAAS,QAAQ,MAAM,EAAE,CAAC,CAAC,YAAY;CAC1D,IAAI,CAAC,OAAO,eAAe,MAAK,MAAK,EAAE,YAAY,MAAM,MAAM,GAC7D,MAAM,IAAI,oBACR,sCAAsC,OAAO,6BAA6B,OAAO,eAAe,KAAK,IAAI,EAAE,IAC3G,EAAE,SAAS;EAAE,KAAK;EAAQ;CAAO,EAAE,CACrC;CAGF,MAAM,OAAO,cAAc,IAAI,QAAQ;CAEvC,IAAI,OAAO,iBAAiB,CAAC,YAAY,MAAM,OAAO,aAAa,GACjE,MAAM,IAAI,oBACR,oCAAoC,KAAK,4BACzC,EAAE,SAAS;EAAE,KAAK;EAAQ;CAAK,EAAE,CACnC;CAGF,IAAI,OAAO,wBACT,MAAM,qBAAqB,MAAM,MAAM;CAGzC,OAAO;AACT;;;;;;;AAQA,eAAe,qBAAqB,MAAc,QAA+B;CAC/E,IAAI,KAAK,IAAI,MAAM,GAAG;EACpB,IAAI,sBAAsB,IAAI,GAC5B,MAAM,IAAI,oBACR,+BAA+B,KAAK,kCACpC,EAAE,SAAS;GAAE,KAAK;GAAQ,SAAS;EAAK,EAAE,CAC5C;EAEF;CACF;CAEA,IAAI;CACJ,IAAI;EACF,YAAY,MAAM,OAAO,MAAM,EAAE,KAAK,KAAK,CAAC;CAC9C,SAAS,OAAO;EACd,MAAM,IAAI,oBACR,sDAAsD,KAAK,2BAC3D;GAAE;GAAO,SAAS;IAAE,KAAK;IAAQ;GAAK;EAAE,CAC1C;CACF;CAEA,KAAK,MAAM,EAAE,aAAa,WACxB,IAAI,sBAAsB,OAAO,GAC/B,MAAM,IAAI,oBACR,oCAAoC,KAAK,4CAA4C,QAAQ,IAC7F,EAAE,SAAS;EAAE,KAAK;EAAQ;EAAM;CAAQ,EAAE,CAC5C;AAGN;;AAGA,SAAS,aACP,SACA,UACa;CACb,IAAI,CAAC,UAAU,OAAO;CAEtB,MAAM,aAAa,IAAI,gBAAgB;CACvC,MAAM,SAAS,SAAsB,WAAW,MAAM,KAAK,MAAM;CAEjE,IAAI,QAAQ,SAAS,MAAM,OAAO;MAC7B,QAAQ,iBAAiB,eAAe,MAAM,OAAO,GAAG,EAAE,MAAM,KAAK,CAAC;CAE3E,IAAI,SAAS,SAAS,MAAM,QAAQ;MAC/B,SAAS,iBAAiB,eAAe,MAAM,QAAQ,GAAG,EAAE,MAAM,KAAK,CAAC;CAE7E,OAAO,WAAW;AACpB;;;;;;;;AASA,eAAsB,aACpB,QACA,aACA,MACmB;CACnB,MAAM,SAAS,sBAAsB,WAAW;CAChD,MAAM,MAAM,MAAM,iBAAiB,QAAQ,MAAM;CAEjD,MAAM,oBAAoB,IAAI,gBAAgB;CAC9C,MAAM,QAAQ,iBAAiB;EAC7B,kBAAkB,MAChB,IAAI,oBACF,oCAAoC,OAAO,UAAU,KACrD,EAAE,SAAS;GAAE,KAAK;GAAQ,WAAW,OAAO;EAAU,EAAE,CAC1D,CACF;CACF,GAAG,OAAO,SAAS;CAEnB,IAAI;EACF,OAAO,MAAM,OAAO,MAAM,KAAK;GAC7B,GAAG;GACH,QAAQ,aAAa,kBAAkB,QAAQ,OAAO,MAAM;EAC9D,CAAC;CACH,UAAU;EACR,aAAa,KAAK;CACpB;AACF;;;;;;;AAQA,eAAsB,eACpB,UACA,UACiB;CACjB,MAAM,WAAW,OAAO,SAAS,QAAQ,IAAI,gBAAgB,CAAC;CAC9D,IAAI,OAAO,SAAS,QAAQ,KAAK,WAAW,UAC1C,MAAM,IAAI,oBACR,+CAA+C,SAAS,qBAAqB,SAAS,YACtF,EAAE,SAAS;EAAE;EAAU;CAAS,EAAE,CACpC;CAGF,IAAI,CAAC,SAAS,MAAM;EAClB,MAAM,OAAO,MAAM,SAAS,KAAK;EACjC,IAAI,OAAO,WAAW,IAAI,IAAI,UAC5B,MAAM,IAAI,oBACR,uCAAuC,SAAS,YAChD,EAAE,SAAS,EAAE,SAAS,EAAE,CAC1B;EAEF,OAAO;CACT;CAEA,MAAM,SAAS,SAAS,KAAK,UAAU;CACvC,MAAM,SAAuB,CAAC;CAC9B,IAAI,QAAQ;CAEZ,SAAS;EACP,MAAM,EAAE,MAAM,UAAU,MAAM,OAAO,KAAK;EAC1C,IAAI,MAAM;EACV,IAAI,CAAC,OAAO;EAEZ,SAAS,MAAM;EACf,IAAI,QAAQ,UAAU;GACpB,MAAM,OAAO,OAAO;GACpB,MAAM,IAAI,oBACR,uCAAuC,SAAS,YAChD,EAAE,SAAS,EAAE,SAAS,EAAE,CAC1B;EACF;EACA,OAAO,KAAK,KAAK;CACnB;CAEA,OAAO,OAAO,OAAO,MAAM,CAAC,CAAC,SAAS,MAAM;AAC9C;;;;;;AAOA,eAAsB,oBACpB,QACA,aACA,MAC4E;CAC5E,MAAM,SAAS,sBAAsB,WAAW;CAChD,MAAM,WAAW,MAAM,aAAa,QAAQ,QAAQ,IAAI;CAExD,OAAO;EACL,IAAI,SAAS;EACb,QAAQ,SAAS;EACjB,YAAY,SAAS;EACrB,MAAM,SAAS,KAAK,MAAM,eAAe,UAAU,OAAO,QAAQ,IAAI;CACxE;AACF"}

package/esm/security/outbound-policy.type.d.mts

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/security/outbound-policy.type.d.ts
+//#region ../ai/src/security/outbound-policy.type.d.ts
 /**
  * One shared policy for every server-side outbound request the framework
  * makes on a user's behalf — attachment fetches (S1), URL skill manifests

package/esm/security/outbound-policy.type.d.mts.map

@@ -1 +1 @@
-{"version":3,"file":"outbound-policy.type.d.mts","names":[],"sources":["../../../../../../../@warlock.js/ai/src/security/outbound-policy.type.ts"],"mappings":";;AAoBA;;;;;;;;;;;;;;AAuCsB;AAItB;;;;KA3CY,cAAA;EA6CV;;;;;EAvCA,cAAA;EA4CA;;;AAAmB;;;EArCnB,aAAA;;;;;;;EAOA,sBAAA;;;;;;EAMA,QAAA;EAEA,SAAA;;;;;EAKA,MAAA,GAAS,WAAA;;;;;;EAMT,KAAA,UAAe,KAAK;AAAA;;KAIV,sBAAA;EACV,cAAA;EACA,aAAA;EACA,sBAAA;EACA,QAAA;EACA,SAAA;EACA,MAAA,GAAS,WAAA;EACT,KAAA,SAAc,KAAK;AAAA"}
+{"version":3,"file":"outbound-policy.type.d.mts","names":[],"sources":["../../../../../../../ai/src/security/outbound-policy.type.ts"],"mappings":";;AAoBA;;;;;;;;;;;;;;AAuCsB;AAItB;;;;KA3CY,cAAA;EA6CV;;;;;EAvCA,cAAA;EA4CA;;;AAAmB;;;EArCnB,aAAA;;;;;;;EAOA,sBAAA;;;;;;EAMA,QAAA;EAEA,SAAA;;;;;EAKA,MAAA,GAAS,WAAA;;;;;;EAMT,KAAA,UAAe,KAAK;AAAA;;KAIV,sBAAA;EACV,cAAA;EACA,aAAA;EACA,sBAAA;EACA,QAAA;EACA,SAAA;EACA,MAAA,GAAS,WAAA;EACT,KAAA,SAAc,KAAK;AAAA"}

package/esm/security/private-ip.d.mts

@@ -1,4 +1,4 @@
-//#region ../@warlock.js/ai/src/security/private-ip.d.ts
+//#region ../ai/src/security/private-ip.d.ts
 /**
  * True when `ip` is a private, loopback, link-local, unique-local,
  * carrier-grade-NAT, unspecified, or otherwise non-public address — the