@ai-agent-forge/plugin-memory 0.88.11 → 0.88.13

package/README.md

@@ -60,6 +60,17 @@ vector-channel enabling path — the default `light` mode (vector channel
 stays disabled until `memory_enable` consent), mode `off`, and the pure
 store/ledger paths never touch them.
 
+本地装配失败和插件卸载会释放本实例的模型与索引。已进入的原生推理先完成再释放；
+清理错逐资源保留原始错误并输出 stderr，不调用宿主已经撤销的插件 API；
+某一资源或诊断输出失败不会阻断其他资源释放。
+transformers 4.3.0 尚不提供逐调用下载取消，卸载后只阻止新推理与后续重试，
+晚到加载会释放而不会发布通道。共享推理服务的模型由服务进程持有，会话卸载只断开客户端。
+成功缓存不删除，非空权重和手工导入文件保持原有检查契约。
+真实验收先经公开 `memory_enable({action:"download"})` 单独准备资源，随后用同一测试私有
+agentDir 验证跨会话写入与召回；冷资源耗时不计入模型任务时限，也不宣称因此变快。
+真实 fixture 同时按生产流程物化安装副本的依赖；失败时保留私有目录与 sidecar 日志。
+详见 `docs/design/冷资源准备与取消边界.md`。
+
 ## Identity
 
 Plugin manifest id `agent-forge.plugin.memory`; the embedded capability

package/agent-forge.json

@@ -1,7 +1,7 @@
 {
 	"packageFormatVersion": 1,
 	"id": "agent-forge.plugin.memory",
-	"pluginVersion": "0.88.11",
+	"pluginVersion": "0.88.13",
 	"apiVersion": "1",
 	"minHostVersion": "0.84.4",
 	"platform": {

package/dist/inference-bin.bundle.js

@@ -2074,6 +2074,9 @@ function createTransformersEmbeddingProvider(options) {
   const remoteHosts = options.remoteHosts ?? [DEFAULT_REMOTE_HOST];
   const load = options.loadImpl ?? createDefaultTransformersLoadImpl(options.moduleEntryPath);
   let backendPromise = null;
+  let disposed = false;
+  let disposePromise;
+  const activeCalls = /* @__PURE__ */ new Set();
   const loadBackend = () => {
     const existing = backendPromise;
     if (existing) {
@@ -2086,13 +2089,16 @@ function createTransformersEmbeddingProvider(options) {
           return await load(modelId, cacheDir, remoteHost);
         } catch (error) {
           lastError = error;
+          if (disposed) throw error;
           purgeZeroByteCacheArtifacts(join4(cacheDir, modelId));
         }
         try {
           await new Promise((resolve2) => setTimeout(resolve2, 2e3));
+          if (disposed) throw lastError;
           return await load(modelId, cacheDir, remoteHost);
         } catch (error) {
           lastError = error;
+          if (disposed) throw error;
           purgeZeroByteCacheArtifacts(join4(cacheDir, modelId));
         }
       }
@@ -2107,28 +2113,48 @@ function createTransformersEmbeddingProvider(options) {
     });
     return attempt;
   };
+  const embed = async (texts) => {
+    if (disposed) throw new Error("memory embedding provider is disposed");
+    if (texts.length === 0) return [];
+    const backend = await loadBackend();
+    if (disposed) throw new Error("memory embedding provider is disposed");
+    const vectors = [];
+    for (let offset = 0; offset < texts.length; offset += batchSize) {
+      if (disposed) throw new Error("memory embedding provider is disposed");
+      const batchTexts = texts.slice(offset, offset + batchSize);
+      const rows = (await backend.extract(batchTexts)).toList();
+      for (const row of rows) {
+        if (row.length !== dimensions) {
+          throw new Error(
+            `embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`
+          );
+        }
+        vectors.push(row);
+      }
+    }
+    return vectors;
+  };
   return {
     modelId,
     dimensions,
-    embed: async (texts) => {
-      if (texts.length === 0) {
-        return [];
-      }
-      const backend = await loadBackend();
-      const vectors = [];
-      for (let offset = 0; offset < texts.length; offset += batchSize) {
-        const batchTexts = texts.slice(offset, offset + batchSize);
-        const rows = (await backend.extract(batchTexts)).toList();
-        for (const row of rows) {
-          if (row.length !== dimensions) {
-            throw new Error(
-              `embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`
-            );
-          }
-          vectors.push(row);
-        }
-      }
-      return vectors;
+    embed: (texts) => {
+      const call = embed(texts);
+      activeCalls.add(call);
+      void call.then(
+        () => activeCalls.delete(call),
+        () => activeCalls.delete(call)
+      );
+      return call;
+    },
+    dispose: () => {
+      disposed = true;
+      disposePromise ??= (async () => {
+        const pendingBackend = backendPromise;
+        await Promise.allSettled([...activeCalls]);
+        const backend = await pendingBackend?.catch(() => void 0);
+        await backend?.dispose?.();
+      })();
+      return disposePromise;
     }
   };
 }
@@ -2151,6 +2177,7 @@ var init_embedding_provider = __esm({
         transformers.env.remoteHost = remoteHost;
         const extractor = await transformers.pipeline("feature-extraction", modelId, { dtype: "q8" });
         return {
+          dispose: () => extractor.dispose(),
           extract: async (texts) => {
             const output = await extractor(texts, { pooling: "cls", normalize: true });
             return { toList: () => output.tolist() };
@@ -2169,6 +2196,9 @@ function createTransformersReranker(options) {
   const remoteHosts = options.remoteHosts ?? [DEFAULT_REMOTE_HOST2];
   const load = options.loadImpl ?? createDefaultTransformersLoadImpl2(options.moduleEntryPath);
   let backendPromise = null;
+  let disposed = false;
+  let disposePromise;
+  const activeCalls = /* @__PURE__ */ new Set();
   const loadBackend = () => {
     const existing = backendPromise;
     if (existing) {
@@ -2181,13 +2211,16 @@ function createTransformersReranker(options) {
           return await load(modelId, cacheDir, remoteHost);
         } catch (error) {
           lastError = error;
+          if (disposed) throw error;
           purgeZeroByteCacheArtifacts(join5(cacheDir, modelId));
         }
         try {
           await new Promise((resolve2) => setTimeout(resolve2, 2e3));
+          if (disposed) throw lastError;
           return await load(modelId, cacheDir, remoteHost);
         } catch (error) {
           lastError = error;
+          if (disposed) throw error;
           purgeZeroByteCacheArtifacts(join5(cacheDir, modelId));
         }
       }
@@ -2202,14 +2235,33 @@ function createTransformersReranker(options) {
     });
     return attempt;
   };
+  const rerank = async (query, docs) => {
+    if (disposed) throw new Error("memory reranker is disposed");
+    if (docs.length === 0) return [];
+    const backend = await loadBackend();
+    if (disposed) throw new Error("memory reranker is disposed");
+    return backend.score(query, [...docs]);
+  };
   return {
     modelId,
-    rerank: async (query, docs) => {
-      if (docs.length === 0) {
-        return [];
-      }
-      const backend = await loadBackend();
-      return backend.score(query, [...docs]);
+    rerank: (query, docs) => {
+      const call = rerank(query, docs);
+      activeCalls.add(call);
+      void call.then(
+        () => activeCalls.delete(call),
+        () => activeCalls.delete(call)
+      );
+      return call;
+    },
+    dispose: () => {
+      disposed = true;
+      disposePromise ??= (async () => {
+        const pendingBackend = backendPromise;
+        await Promise.allSettled([...activeCalls]);
+        const backend = await pendingBackend?.catch(() => void 0);
+        await backend?.dispose?.();
+      })();
+      return disposePromise;
     }
   };
 }
@@ -2233,6 +2285,9 @@ var init_embedding_reranker = __esm({
           dtype: "q8"
         });
         return {
+          dispose: async () => {
+            await model.dispose();
+          },
           score: async (query, docs) => {
             const inputs = tokenizer(
               docs.map(() => query),
@@ -2258,6 +2313,19 @@ async function createMemoryVectorIndex(options) {
   );
   const db = await lancedb.connect(options.dbPath);
   let table = null;
+  let closed = false;
+  let closePromise;
+  const activeCalls = /* @__PURE__ */ new Set();
+  const withOpenIndex = (operation) => {
+    if (closed) return Promise.reject(new Error("memory vector index is closed"));
+    const call = operation();
+    activeCalls.add(call);
+    void call.then(
+      () => activeCalls.delete(call),
+      () => activeCalls.delete(call)
+    );
+    return call;
+  };
   const ensureTable = async () => {
     if (table) {
       return table;
@@ -2277,7 +2345,7 @@ async function createMemoryVectorIndex(options) {
   };
   return {
     replicaId: "memory-vector-index",
-    upsert: async (entries) => {
+    upsert: (entries) => withOpenIndex(async () => {
       if (entries.length === 0) {
         return;
       }
@@ -2302,8 +2370,8 @@ async function createMemoryVectorIndex(options) {
         return;
       }
       await current.mergeInsert("memoryId").whenMatchedUpdateAll().whenNotMatchedInsertAll().execute(rows);
-    },
-    remove: async (memoryIds) => {
+    }),
+    remove: (memoryIds) => withOpenIndex(async () => {
       if (memoryIds.length === 0) {
         return;
       }
@@ -2313,8 +2381,8 @@ async function createMemoryVectorIndex(options) {
       }
       const predicate = `memoryId IN (${memoryIds.map((id) => `'${escapeSqlLiteral(id)}'`).join(",")})`;
       await current.delete(predicate);
-    },
-    queryKnn: async (vector, query) => {
+    }),
+    queryKnn: (vector, query) => withOpenIndex(async () => {
       if (vector.length !== dimensions) {
         throw new Error(`query vector dimension mismatch: got ${vector.length}, index declares ${dimensions}`);
       }
@@ -2324,40 +2392,45 @@ async function createMemoryVectorIndex(options) {
       }
       const rows = await current.search(Array.from(vector)).where(`owner = '${escapeSqlLiteral(query.owner)}'`).limit(query.limit).toArray();
       return rows.map((row) => ({ memoryId: row.memoryId, score: 1 / (1 + row._distance) }));
-    },
-    queryFts: async (queryText, query) => {
+    }),
+    queryFts: (queryText, query) => withOpenIndex(async () => {
       const current = await ensureTable();
       if (!current) {
         return [];
       }
       const rows = await current.search(queryText).where(`owner = '${escapeSqlLiteral(query.owner)}'`).limit(query.limit).toArray();
       return rows.map((row, rank) => ({ memoryId: row.memoryId, score: 1 / (RRF_K + rank + 1) }));
-    },
-    listMemoryIds: async () => {
+    }),
+    listMemoryIds: () => withOpenIndex(async () => {
       const current = await ensureTable();
       if (!current) {
         return /* @__PURE__ */ new Set();
       }
       const rows = await current.query().select(["memoryId"]).toArray();
       return new Set(rows.map((row) => row.memoryId));
-    },
-    count: async () => {
+    }),
+    count: () => withOpenIndex(async () => {
       const current = await ensureTable();
       if (!current) {
         return 0;
       }
       return current.countRows();
-    },
-    getStoredModelId: async () => {
+    }),
+    getStoredModelId: () => withOpenIndex(async () => {
       const current = await ensureTable();
       if (!current) {
         return void 0;
       }
       const rows = await current.query().select(["modelId"]).limit(1).toArray();
       return rows[0]?.modelId;
-    },
-    close: async () => {
-      db.close();
+    }),
+    close: () => {
+      closed = true;
+      closePromise ??= (async () => {
+        await Promise.allSettled([...activeCalls]);
+        db.close();
+      })();
+      return closePromise;
     }
   };
 }
@@ -10835,7 +10908,7 @@ var FrameDecoder = class {
 };
 
 // ../../packages/protocol/src/schemas.ts
-var PROTOCOL_VERSION = 1;
+var PROTOCOL_VERSION = 2;
 var IdSchema = build_default.String({ minLength: 1 });
 var TimestampSchema = build_default.Integer({ minimum: 0 });
 var Base64Schema = build_default.String({ pattern: "^(?:[A-Za-z0-9+/]{4})*(?:[A-Za-z0-9+/]{2}==|[A-Za-z0-9+/]{3}=)?$" });
@@ -10874,12 +10947,6 @@ var ModelRefSchema = StrictObject({
   provider: IdSchema,
   id: IdSchema
 });
-var ModelCostSchema = StrictObject({
-  input: build_default.Number({ minimum: 0 }),
-  output: build_default.Number({ minimum: 0 }),
-  cacheRead: build_default.Number({ minimum: 0 }),
-  cacheWrite: build_default.Number({ minimum: 0 })
-});
 var ModelMetadataSchema = StrictObject({
   provider: IdSchema,
   id: IdSchema,
@@ -10889,7 +10956,6 @@ var ModelMetadataSchema = StrictObject({
   input: build_default.Array(build_default.Union([build_default.Literal("text"), build_default.Literal("image")])),
   contextWindow: build_default.Integer({ minimum: 1 }),
   maxTokens: build_default.Integer({ minimum: 1 }),
-  cost: ModelCostSchema,
   supportedThinkingLevels: build_default.Array(ThinkingLevelSchema, { minItems: 1 }),
   authenticated: build_default.Boolean()
 });
@@ -10922,14 +10988,7 @@ var UsageSchema = StrictObject({
   cacheRead: build_default.Integer({ minimum: 0 }),
   cacheWrite: build_default.Integer({ minimum: 0 }),
   reasoning: build_default.Optional(build_default.Integer({ minimum: 0 })),
-  totalTokens: build_default.Integer({ minimum: 0 }),
-  cost: StrictObject({
-    input: build_default.Number({ minimum: 0 }),
-    output: build_default.Number({ minimum: 0 }),
-    cacheRead: build_default.Number({ minimum: 0 }),
-    cacheWrite: build_default.Number({ minimum: 0 }),
-    total: build_default.Number({ minimum: 0 })
-  })
+  totalTokens: build_default.Integer({ minimum: 0 })
 });
 var UserTranscriptItemSchema = StrictObject({
   id: IdSchema,
@@ -11173,6 +11232,27 @@ var ArtifactTokenIssueCommandSchema = StrictObject({
   sessionId: IdSchema,
   artifactId: IdSchema
 });
+var ArtifactListCommandSchema = StrictObject({
+  command: build_default.Literal("artifact_list"),
+  sessionId: IdSchema
+});
+var FileSearchCommandSchema = StrictObject({
+  command: build_default.Literal("file_search"),
+  sessionId: IdSchema,
+  // 空串合法=预览模式（@ 触发即展示前 limit 条，mock 与组件契约同语义）。
+  query: build_default.String(),
+  limit: build_default.Optional(build_default.Integer({ minimum: 1, maximum: 50 }))
+});
+var FileSearchResultSchema = StrictObject({
+  command: build_default.Literal("file_search"),
+  results: build_default.Array(
+    StrictObject({
+      name: build_default.String(),
+      path: build_default.String()
+    })
+  ),
+  truncated: build_default.Boolean()
+});
 var LeaseRenewCommandSchema = StrictObject({
   command: build_default.Literal("lease_renew"),
   sessionId: IdSchema
@@ -11215,6 +11295,8 @@ var CommandSchema = build_default.Union([
   ArtifactStatCommandSchema,
   ArtifactReadCommandSchema,
   ArtifactTokenIssueCommandSchema,
+  ArtifactListCommandSchema,
+  FileSearchCommandSchema,
   LeaseRenewCommandSchema,
   OperationSubmitCommandSchema,
   OperationCancelCommandSchema,
@@ -11281,6 +11363,10 @@ var ArtifactTokenIssueResultSchema = StrictObject({
   sessionId: IdSchema,
   artifactId: IdSchema
 });
+var ArtifactListResultSchema = StrictObject({
+  command: build_default.Literal("artifact_list"),
+  artifacts: build_default.Array(ArtifactRefSchema)
+});
 var ReplayEventSchema = StrictObject({
   cursor: build_default.Integer({ minimum: 1 }),
   event: build_default.Union([
@@ -11329,6 +11415,8 @@ var CommandResultSchema = build_default.Union([
   ArtifactStatResultSchema,
   ArtifactReadResultSchema,
   ArtifactTokenIssueResultSchema,
+  ArtifactListResultSchema,
+  FileSearchResultSchema,
   OperationAckResultSchema
 ]);
 var ClientHelloSchema = StrictObject({

package/dist/memory/embedding-provider.d.ts

@@ -22,6 +22,8 @@ export interface MemoryEmbeddingProviderV1 {
      * responsibility (低档 bge-small-zh 系查询侧指令由调用方负责添加).
      */
     embed(texts: readonly string[]): Promise<readonly (readonly number[])[]>;
+    /** 仅释放本实例拥有的模型；等待已进入的推理完成，不删除缓存文件。 */
+    dispose?(): Promise<void>;
 }
 /**
  * Minimal structural seam around a transformers.js feature-extraction
@@ -32,6 +34,7 @@ export interface TransformersEmbeddingBackendV1 {
     extract: (texts: string[]) => Promise<{
         toList: () => number[][];
     }>;
+    dispose?(): Promise<void>;
 }
 export type TransformersEmbeddingLoadImpl = (modelId: string, cacheDir: string, remoteHost: string) => Promise<TransformersEmbeddingBackendV1>;
 /**

package/dist/memory/embedding-provider.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"embedding-provider.d.ts","sourceRoot":"","sources":["../../src/memory/embedding-provider.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAUH,MAAM,WAAW,yBAAyB;IACzC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B;;;;;OAKG;IACH,KAAK,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,CAAC,SAAS,CAAC,SAAS,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC;CACzE;AAED;;;GAGG;AACH,MAAM,WAAW,8BAA8B;IAC9C,gFAAgF;IAChF,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,MAAM,EAAE,EAAE,CAAA;KAAE,CAAC,CAAC;CACpE;AAED,MAAM,MAAM,6BAA6B,GAAG,CAC3C,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,KACd,OAAO,CAAC,8BAA8B,CAAC,CAAC;AA2C7C;;;;;GAKG;AACH,wBAAgB,wCAAwC,CAAC,OAAO,CAAC,EAAE;IAClE,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;CAC7B,GAAG,yBAAyB,CAO5B;AAmBD;;;;;;;;;;;;;GAaG;AACH,wBAAgB,mCAAmC,CAAC,OAAO,EAAE;IAC5D,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,WAAW,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,CAAC,EAAE,6BAA6B,CAAC;IAClD,iGAAiG;IACjG,QAAQ,CAAC,eAAe,CAAC,EAAE,MAAM,CAAC;CAClC,GAAG,yBAAyB,CA0E5B","sourcesContent":["/**\n * Memory embedding providers for hybrid retrieval (混合检索 · 向量侧).\n *\n * POC-verified (2026-09, 100k corpus): high tier (Xenova/bge-m3, 1024d) and\n * low tier (Xenova/bge-small-zh-v1.5, 512d) inference is deterministic\n * bit-exact across runs; high-tier drift queries reach re@10 100% after\n * reranking, low tier 83%. Tier presets are caller configuration — this module\n * only exposes a modelId-parameterized interface (宪法: 不绑死单一 provider).\n *\n * Runtime dep @huggingface/transformers is lazy-loaded via `await import` at\n * first embed (惯例参考 packages/agent-forge/src/utils/photon.ts); type\n * positions use type-only imports. The deterministic mock provider keeps tests\n * offline (门禁: deterministic mock AI; 真实模型不进 CI).\n */\n\n/**\n * Batch embedding provider. Implementations must be deterministic for\n * identical input and must not silently degrade: any failure propagates.\n */\nimport { join } from \"node:path\";\nimport { importHostOrBareModule } from \"./host-module-import.ts\";\nimport { annotateUpstreamListingPoison, purgeZeroByteCacheArtifacts } from \"./model-cache-hygiene.ts\";\n\nexport interface MemoryEmbeddingProviderV1 {\n\treadonly modelId: string;\n\treadonly dimensions: number;\n\t/**\n\t * Batch embed. Vectors are L2-normalized float arrays. Deterministic for\n\t * identical input. Plain text mapping — no instruction prefix is added\n\t * here; query-side instructions for bge-small-zh family are the caller's\n\t * responsibility (低档 bge-small-zh 系查询侧指令由调用方负责添加).\n\t */\n\tembed(texts: readonly string[]): Promise<readonly (readonly number[])[]>;\n}\n\n/**\n * Minimal structural seam around a transformers.js feature-extraction\n * pipeline, so tests can inject a fake without touching the real loader.\n */\nexport interface TransformersEmbeddingBackendV1 {\n\t/** Extract features for a batch of texts; one vector per input, input order. */\n\textract: (texts: string[]) => Promise<{ toList: () => number[][] }>;\n}\n\nexport type TransformersEmbeddingLoadImpl = (\n\tmodelId: string,\n\tcacheDir: string,\n\tremoteHost: string,\n) => Promise<TransformersEmbeddingBackendV1>;\n\nconst DEFAULT_TRANSFORMERS_MODEL_ID = \"Xenova/bge-m3\";\nconst DEFAULT_TRANSFORMERS_DIMENSIONS = 1024;\nconst DEFAULT_TRANSFORMERS_BATCH_SIZE = 16;\nconst DEFAULT_REMOTE_HOST = \"https://huggingface.co\";\n\nfunction fnv1a32(input: string): number {\n\tlet hash = 0x811c9dc5;\n\tfor (let offset = 0; offset < input.length; offset += 1) {\n\t\thash ^= input.charCodeAt(offset);\n\t\thash = Math.imul(hash, 0x01000193);\n\t}\n\treturn hash >>> 0;\n}\n\nfunction embedDeterministicMock(text: string, dimensions: number): number[] {\n\tconst values = new Array<number>(dimensions).fill(0);\n\tconst tokens = text.toLowerCase().split(/[^\\p{L}\\p{N}]+/u);\n\tfor (const token of tokens) {\n\t\tif (token.length === 0) {\n\t\t\tcontinue;\n\t\t}\n\t\t// 多轮混淆累加: 每个 token 经 FNV-1a 后再派生 3 个维度沉点,\n\t\t// 放大不同文本间的向量差异; 全程无随机、无时钟。\n\t\tlet mixed = fnv1a32(token);\n\t\tfor (let round = 0; round < 3; round += 1) {\n\t\t\tmixed = Math.imul(mixed ^ 0x9e3779b9, 0x85ebca6b) >>> 0;\n\t\t\tvalues[mixed % dimensions] += 1;\n\t\t}\n\t}\n\tlet sumSquares = 0;\n\tfor (const value of values) {\n\t\tsumSquares += value * value;\n\t}\n\tif (sumSquares === 0) {\n\t\t// 空文本/全标点: 确定性零向量; L2 范数为 0, 不归一化 (避免除零 NaN)。\n\t\treturn values;\n\t}\n\tconst norm = Math.sqrt(sumSquares);\n\treturn values.map((value) => value / norm);\n}\n\n/**\n * Deterministic mock embedding provider for offline tests: tokenizes by\n * `[^\\p{L}\\p{N}]+`, hashes each token with FNV-1a plus multi-round mixing into\n * dimension indices, then L2-normalizes. Same text → same vector, no random,\n * no clock. POC 已用真实模型验证 bit-exact; mock 只承担确定性测试门禁。\n */\nexport function createDeterministicMockEmbeddingProvider(options?: {\n\treadonly dimensions?: number;\n}): MemoryEmbeddingProviderV1 {\n\tconst dimensions = options?.dimensions ?? 64;\n\treturn {\n\t\tmodelId: \"mock-embedding@1\",\n\t\tdimensions,\n\t\tembed: async (texts) => texts.map((text) => embedDeterministicMock(text, dimensions)),\n\t};\n}\n\nconst createDefaultTransformersLoadImpl = (moduleEntryPath?: string): TransformersEmbeddingLoadImpl => {\n\treturn async (modelId, cacheDir, remoteHost) => {\n\t\tconst transformers = await importHostOrBareModule<typeof import(\"@huggingface/transformers\")>(\n\t\t\tmoduleEntryPath ?? \"@huggingface/transformers\",\n\t\t);\n\t\ttransformers.env.cacheDir = cacheDir;\n\t\ttransformers.env.remoteHost = remoteHost;\n\t\tconst extractor = await transformers.pipeline(\"feature-extraction\", modelId, { dtype: \"q8\" });\n\t\treturn {\n\t\t\textract: async (texts) => {\n\t\t\t\tconst output = await extractor(texts, { pooling: \"cls\", normalize: true });\n\t\t\t\treturn { toList: () => output.tolist() as number[][] };\n\t\t\t},\n\t\t};\n\t};\n};\n\n/**\n * Transformers.js-backed embedding provider (POC 档位: 高档默认\n * Xenova/bge-m3/1024d; 低档由调用方传 modelId \"Xenova/bge-small-zh-v1.5\" +\n * dimensions 512)。\n *\n * 镜像序列: 首次 embed 时按 `remoteHosts` 顺序逐个尝试加载 (调用方通常传\n * [\"https://huggingface.co\", \"https://hf-mirror.com\"]); 某 host 加载抛错则\n * 尝试下一个, 全部失败时抛最后一个错误。host 确定后模型句柄只加载一次\n * (并发 embed 共享同一 in-flight promise); 加载失败不缓存, 下次 embed 重新\n * 走完整序列, 失败原因原样保留, 不静默降级。\n *\n * `loadImpl` 是测试 seam: 注入后绕过真实加载, 签名\n * (modelId, cacheDir, remoteHost) => backend。\n */\nexport function createTransformersEmbeddingProvider(options: {\n\treadonly modelId?: string;\n\treadonly dimensions?: number;\n\treadonly cacheDir: string;\n\treadonly remoteHosts?: readonly string[];\n\treadonly batchSize?: number;\n\treadonly loadImpl?: TransformersEmbeddingLoadImpl;\n\t/** Host-resolved transformers entry path (install-store visibility); undefined = bare import. */\n\treadonly moduleEntryPath?: string;\n}): MemoryEmbeddingProviderV1 {\n\tconst modelId = options.modelId ?? DEFAULT_TRANSFORMERS_MODEL_ID;\n\tconst dimensions = options.dimensions ?? DEFAULT_TRANSFORMERS_DIMENSIONS;\n\tconst batchSize = options.batchSize ?? DEFAULT_TRANSFORMERS_BATCH_SIZE;\n\tconst cacheDir = options.cacheDir;\n\tconst remoteHosts = options.remoteHosts ?? [DEFAULT_REMOTE_HOST];\n\tconst load = options.loadImpl ?? createDefaultTransformersLoadImpl(options.moduleEntryPath);\n\n\tlet backendPromise: Promise<TransformersEmbeddingBackendV1> | null = null;\n\tconst loadBackend = (): Promise<TransformersEmbeddingBackendV1> => {\n\t\tconst existing = backendPromise;\n\t\tif (existing) {\n\t\t\treturn existing;\n\t\t}\n\t\tconst attempt = (async () => {\n\t\t\tlet lastError: unknown;\n\t\t\tfor (const remoteHost of remoteHosts) {\n\t\t\t\ttry {\n\t\t\t\t\treturn await load(modelId, cacheDir, remoteHost);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tlastError = error;\n\t\t\t\t\t// 镜像轮换前清 0 字节残骸: 失败下载留下的空文件会被 transformers\n\t\t\t\t\t// 当作缓存命中, 毒化下一个 host 的重试 (model-cache-hygiene)。\n\t\t\t\t\tpurgeZeroByteCacheArtifacts(join(cacheDir, modelId));\n\t\t\t\t}\n\t\t\t\t// 同 host 一次短退避重试 (2026-10-06 live E2E): 覆盖 rejection 型瞬断\n\t\t\t\t// (fetch 网络错/5xx)。对\"仓库 API 清单缺文件\"型不生效——该结果被\n\t\t\t\t// transformers 模块级 memoize (见 annotateUpstreamListingPoison),\n\t\t\t\t// 进程内换 host 也拿同一空清单。\n\t\t\t\ttry {\n\t\t\t\t\tawait new Promise((resolve) => setTimeout(resolve, 2_000));\n\t\t\t\t\treturn await load(modelId, cacheDir, remoteHost);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tlastError = error;\n\t\t\t\t\tpurgeZeroByteCacheArtifacts(join(cacheDir, modelId));\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (lastError === undefined) throw new Error(`no remote hosts configured for model ${modelId}`);\n\t\t\tthrow annotateUpstreamListingPoison(lastError);\n\t\t})();\n\t\tbackendPromise = attempt;\n\t\tattempt.catch(() => {\n\t\t\t// 失败不缓存: 下次 embed 重新走镜像序列 (显式失败, 不静默降级)。\n\t\t\tif (backendPromise === attempt) {\n\t\t\t\tbackendPromise = null;\n\t\t\t}\n\t\t});\n\t\treturn attempt;\n\t};\n\n\treturn {\n\t\tmodelId,\n\t\tdimensions,\n\t\tembed: async (texts) => {\n\t\t\tif (texts.length === 0) {\n\t\t\t\treturn [];\n\t\t\t}\n\t\t\tconst backend = await loadBackend();\n\t\t\tconst vectors: number[][] = [];\n\t\t\tfor (let offset = 0; offset < texts.length; offset += batchSize) {\n\t\t\t\tconst batchTexts = texts.slice(offset, offset + batchSize);\n\t\t\t\tconst rows = (await backend.extract(batchTexts)).toList();\n\t\t\t\tfor (const row of rows) {\n\t\t\t\t\tif (row.length !== dimensions) {\n\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t`embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t\tvectors.push(row);\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn vectors;\n\t\t},\n\t};\n}\n"]}
+{"version":3,"file":"embedding-provider.d.ts","sourceRoot":"","sources":["../../src/memory/embedding-provider.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAUH,MAAM,WAAW,yBAAyB;IACzC,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAC5B;;;;;OAKG;IACH,KAAK,CAAC,KAAK,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,CAAC,SAAS,CAAC,SAAS,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC;IACzE,oGAAsC;IACtC,OAAO,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CAC1B;AAED;;;GAGG;AACH,MAAM,WAAW,8BAA8B;IAC9C,gFAAgF;IAChF,OAAO,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,OAAO,CAAC;QAAE,MAAM,EAAE,MAAM,MAAM,EAAE,EAAE,CAAA;KAAE,CAAC,CAAC;IACpE,OAAO,CAAC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;CAC1B;AAED,MAAM,MAAM,6BAA6B,GAAG,CAC3C,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,MAAM,EAChB,UAAU,EAAE,MAAM,KACd,OAAO,CAAC,8BAA8B,CAAC,CAAC;AA2C7C;;;;;GAKG;AACH,wBAAgB,wCAAwC,CAAC,OAAO,CAAC,EAAE;IAClE,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;CAC7B,GAAG,yBAAyB,CAO5B;AAoBD;;;;;;;;;;;;;GAaG;AACH,wBAAgB,mCAAmC,CAAC,OAAO,EAAE;IAC5D,QAAQ,CAAC,OAAO,CAAC,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,UAAU,CAAC,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,WAAW,CAAC,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAC5B,QAAQ,CAAC,QAAQ,CAAC,EAAE,6BAA6B,CAAC;IAClD,iGAAiG;IACjG,QAAQ,CAAC,eAAe,CAAC,EAAE,MAAM,CAAC;CAClC,GAAG,yBAAyB,CAoG5B","sourcesContent":["/**\n * Memory embedding providers for hybrid retrieval (混合检索 · 向量侧).\n *\n * POC-verified (2026-09, 100k corpus): high tier (Xenova/bge-m3, 1024d) and\n * low tier (Xenova/bge-small-zh-v1.5, 512d) inference is deterministic\n * bit-exact across runs; high-tier drift queries reach re@10 100% after\n * reranking, low tier 83%. Tier presets are caller configuration — this module\n * only exposes a modelId-parameterized interface (宪法: 不绑死单一 provider).\n *\n * Runtime dep @huggingface/transformers is lazy-loaded via `await import` at\n * first embed (惯例参考 packages/agent-forge/src/utils/photon.ts); type\n * positions use type-only imports. The deterministic mock provider keeps tests\n * offline (门禁: deterministic mock AI; 真实模型不进 CI).\n */\n\n/**\n * Batch embedding provider. Implementations must be deterministic for\n * identical input and must not silently degrade: any failure propagates.\n */\nimport { join } from \"node:path\";\nimport { importHostOrBareModule } from \"./host-module-import.ts\";\nimport { annotateUpstreamListingPoison, purgeZeroByteCacheArtifacts } from \"./model-cache-hygiene.ts\";\n\nexport interface MemoryEmbeddingProviderV1 {\n\treadonly modelId: string;\n\treadonly dimensions: number;\n\t/**\n\t * Batch embed. Vectors are L2-normalized float arrays. Deterministic for\n\t * identical input. Plain text mapping — no instruction prefix is added\n\t * here; query-side instructions for bge-small-zh family are the caller's\n\t * responsibility (低档 bge-small-zh 系查询侧指令由调用方负责添加).\n\t */\n\tembed(texts: readonly string[]): Promise<readonly (readonly number[])[]>;\n\t/** 仅释放本实例拥有的模型；等待已进入的推理完成，不删除缓存文件。 */\n\tdispose?(): Promise<void>;\n}\n\n/**\n * Minimal structural seam around a transformers.js feature-extraction\n * pipeline, so tests can inject a fake without touching the real loader.\n */\nexport interface TransformersEmbeddingBackendV1 {\n\t/** Extract features for a batch of texts; one vector per input, input order. */\n\textract: (texts: string[]) => Promise<{ toList: () => number[][] }>;\n\tdispose?(): Promise<void>;\n}\n\nexport type TransformersEmbeddingLoadImpl = (\n\tmodelId: string,\n\tcacheDir: string,\n\tremoteHost: string,\n) => Promise<TransformersEmbeddingBackendV1>;\n\nconst DEFAULT_TRANSFORMERS_MODEL_ID = \"Xenova/bge-m3\";\nconst DEFAULT_TRANSFORMERS_DIMENSIONS = 1024;\nconst DEFAULT_TRANSFORMERS_BATCH_SIZE = 16;\nconst DEFAULT_REMOTE_HOST = \"https://huggingface.co\";\n\nfunction fnv1a32(input: string): number {\n\tlet hash = 0x811c9dc5;\n\tfor (let offset = 0; offset < input.length; offset += 1) {\n\t\thash ^= input.charCodeAt(offset);\n\t\thash = Math.imul(hash, 0x01000193);\n\t}\n\treturn hash >>> 0;\n}\n\nfunction embedDeterministicMock(text: string, dimensions: number): number[] {\n\tconst values = new Array<number>(dimensions).fill(0);\n\tconst tokens = text.toLowerCase().split(/[^\\p{L}\\p{N}]+/u);\n\tfor (const token of tokens) {\n\t\tif (token.length === 0) {\n\t\t\tcontinue;\n\t\t}\n\t\t// 多轮混淆累加: 每个 token 经 FNV-1a 后再派生 3 个维度沉点,\n\t\t// 放大不同文本间的向量差异; 全程无随机、无时钟。\n\t\tlet mixed = fnv1a32(token);\n\t\tfor (let round = 0; round < 3; round += 1) {\n\t\t\tmixed = Math.imul(mixed ^ 0x9e3779b9, 0x85ebca6b) >>> 0;\n\t\t\tvalues[mixed % dimensions] += 1;\n\t\t}\n\t}\n\tlet sumSquares = 0;\n\tfor (const value of values) {\n\t\tsumSquares += value * value;\n\t}\n\tif (sumSquares === 0) {\n\t\t// 空文本/全标点: 确定性零向量; L2 范数为 0, 不归一化 (避免除零 NaN)。\n\t\treturn values;\n\t}\n\tconst norm = Math.sqrt(sumSquares);\n\treturn values.map((value) => value / norm);\n}\n\n/**\n * Deterministic mock embedding provider for offline tests: tokenizes by\n * `[^\\p{L}\\p{N}]+`, hashes each token with FNV-1a plus multi-round mixing into\n * dimension indices, then L2-normalizes. Same text → same vector, no random,\n * no clock. POC 已用真实模型验证 bit-exact; mock 只承担确定性测试门禁。\n */\nexport function createDeterministicMockEmbeddingProvider(options?: {\n\treadonly dimensions?: number;\n}): MemoryEmbeddingProviderV1 {\n\tconst dimensions = options?.dimensions ?? 64;\n\treturn {\n\t\tmodelId: \"mock-embedding@1\",\n\t\tdimensions,\n\t\tembed: async (texts) => texts.map((text) => embedDeterministicMock(text, dimensions)),\n\t};\n}\n\nconst createDefaultTransformersLoadImpl = (moduleEntryPath?: string): TransformersEmbeddingLoadImpl => {\n\treturn async (modelId, cacheDir, remoteHost) => {\n\t\tconst transformers = await importHostOrBareModule<typeof import(\"@huggingface/transformers\")>(\n\t\t\tmoduleEntryPath ?? \"@huggingface/transformers\",\n\t\t);\n\t\ttransformers.env.cacheDir = cacheDir;\n\t\ttransformers.env.remoteHost = remoteHost;\n\t\tconst extractor = await transformers.pipeline(\"feature-extraction\", modelId, { dtype: \"q8\" });\n\t\treturn {\n\t\t\tdispose: () => extractor.dispose(),\n\t\t\textract: async (texts) => {\n\t\t\t\tconst output = await extractor(texts, { pooling: \"cls\", normalize: true });\n\t\t\t\treturn { toList: () => output.tolist() as number[][] };\n\t\t\t},\n\t\t};\n\t};\n};\n\n/**\n * Transformers.js-backed embedding provider (POC 档位: 高档默认\n * Xenova/bge-m3/1024d; 低档由调用方传 modelId \"Xenova/bge-small-zh-v1.5\" +\n * dimensions 512)。\n *\n * 镜像序列: 首次 embed 时按 `remoteHosts` 顺序逐个尝试加载 (调用方通常传\n * [\"https://huggingface.co\", \"https://hf-mirror.com\"]); 某 host 加载抛错则\n * 尝试下一个, 全部失败时抛最后一个错误。host 确定后模型句柄只加载一次\n * (并发 embed 共享同一 in-flight promise); 加载失败不缓存, 下次 embed 重新\n * 走完整序列, 失败原因原样保留, 不静默降级。\n *\n * `loadImpl` 是测试 seam: 注入后绕过真实加载, 签名\n * (modelId, cacheDir, remoteHost) => backend。\n */\nexport function createTransformersEmbeddingProvider(options: {\n\treadonly modelId?: string;\n\treadonly dimensions?: number;\n\treadonly cacheDir: string;\n\treadonly remoteHosts?: readonly string[];\n\treadonly batchSize?: number;\n\treadonly loadImpl?: TransformersEmbeddingLoadImpl;\n\t/** Host-resolved transformers entry path (install-store visibility); undefined = bare import. */\n\treadonly moduleEntryPath?: string;\n}): MemoryEmbeddingProviderV1 {\n\tconst modelId = options.modelId ?? DEFAULT_TRANSFORMERS_MODEL_ID;\n\tconst dimensions = options.dimensions ?? DEFAULT_TRANSFORMERS_DIMENSIONS;\n\tconst batchSize = options.batchSize ?? DEFAULT_TRANSFORMERS_BATCH_SIZE;\n\tconst cacheDir = options.cacheDir;\n\tconst remoteHosts = options.remoteHosts ?? [DEFAULT_REMOTE_HOST];\n\tconst load = options.loadImpl ?? createDefaultTransformersLoadImpl(options.moduleEntryPath);\n\n\tlet backendPromise: Promise<TransformersEmbeddingBackendV1> | null = null;\n\tlet disposed = false;\n\tlet disposePromise: Promise<void> | undefined;\n\tconst activeCalls = new Set<Promise<unknown>>();\n\tconst loadBackend = (): Promise<TransformersEmbeddingBackendV1> => {\n\t\tconst existing = backendPromise;\n\t\tif (existing) {\n\t\t\treturn existing;\n\t\t}\n\t\tconst attempt = (async () => {\n\t\t\tlet lastError: unknown;\n\t\t\tfor (const remoteHost of remoteHosts) {\n\t\t\t\ttry {\n\t\t\t\t\treturn await load(modelId, cacheDir, remoteHost);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tlastError = error;\n\t\t\t\t\tif (disposed) throw error;\n\t\t\t\t\t// 镜像轮换前清 0 字节残骸: 失败下载留下的空文件会被 transformers\n\t\t\t\t\t// 当作缓存命中, 毒化下一个 host 的重试 (model-cache-hygiene)。\n\t\t\t\t\tpurgeZeroByteCacheArtifacts(join(cacheDir, modelId));\n\t\t\t\t}\n\t\t\t\t// 同 host 一次短退避重试 (2026-10-06 live E2E): 覆盖 rejection 型瞬断\n\t\t\t\t// (fetch 网络错/5xx)。对\"仓库 API 清单缺文件\"型不生效——该结果被\n\t\t\t\t// transformers 模块级 memoize (见 annotateUpstreamListingPoison),\n\t\t\t\t// 进程内换 host 也拿同一空清单。\n\t\t\t\ttry {\n\t\t\t\t\tawait new Promise((resolve) => setTimeout(resolve, 2_000));\n\t\t\t\t\tif (disposed) throw lastError;\n\t\t\t\t\treturn await load(modelId, cacheDir, remoteHost);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tlastError = error;\n\t\t\t\t\tif (disposed) throw error;\n\t\t\t\t\tpurgeZeroByteCacheArtifacts(join(cacheDir, modelId));\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (lastError === undefined) throw new Error(`no remote hosts configured for model ${modelId}`);\n\t\t\tthrow annotateUpstreamListingPoison(lastError);\n\t\t})();\n\t\tbackendPromise = attempt;\n\t\tattempt.catch(() => {\n\t\t\t// 失败不缓存: 下次 embed 重新走镜像序列 (显式失败, 不静默降级)。\n\t\t\tif (backendPromise === attempt) {\n\t\t\t\tbackendPromise = null;\n\t\t\t}\n\t\t});\n\t\treturn attempt;\n\t};\n\n\tconst embed = async (texts: readonly string[]): Promise<readonly (readonly number[])[]> => {\n\t\tif (disposed) throw new Error(\"memory embedding provider is disposed\");\n\t\tif (texts.length === 0) return [];\n\t\tconst backend = await loadBackend();\n\t\tif (disposed) throw new Error(\"memory embedding provider is disposed\");\n\t\tconst vectors: number[][] = [];\n\t\tfor (let offset = 0; offset < texts.length; offset += batchSize) {\n\t\t\tif (disposed) throw new Error(\"memory embedding provider is disposed\");\n\t\t\tconst batchTexts = texts.slice(offset, offset + batchSize);\n\t\t\tconst rows = (await backend.extract(batchTexts)).toList();\n\t\t\tfor (const row of rows) {\n\t\t\t\tif (row.length !== dimensions) {\n\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t`embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\tvectors.push(row);\n\t\t\t}\n\t\t}\n\t\treturn vectors;\n\t};\n\treturn {\n\t\tmodelId,\n\t\tdimensions,\n\t\tembed: (texts) => {\n\t\t\tconst call = embed(texts);\n\t\t\tactiveCalls.add(call);\n\t\t\tvoid call.then(\n\t\t\t\t() => activeCalls.delete(call),\n\t\t\t\t() => activeCalls.delete(call),\n\t\t\t);\n\t\t\treturn call;\n\t\t},\n\t\tdispose: () => {\n\t\t\tdisposed = true;\n\t\t\tdisposePromise ??= (async () => {\n\t\t\t\tconst pendingBackend = backendPromise;\n\t\t\t\tawait Promise.allSettled([...activeCalls]);\n\t\t\t\tconst backend = await pendingBackend?.catch(() => undefined);\n\t\t\t\tawait backend?.dispose?.();\n\t\t\t})();\n\t\t\treturn disposePromise;\n\t\t},\n\t};\n}\n"]}

package/dist/memory/embedding-provider.js

@@ -78,6 +78,7 @@ const createDefaultTransformersLoadImpl = (moduleEntryPath) => {
         transformers.env.remoteHost = remoteHost;
         const extractor = await transformers.pipeline("feature-extraction", modelId, { dtype: "q8" });
         return {
+            dispose: () => extractor.dispose(),
             extract: async (texts) => {
                 const output = await extractor(texts, { pooling: "cls", normalize: true });
                 return { toList: () => output.tolist() };
@@ -107,6 +108,9 @@ export function createTransformersEmbeddingProvider(options) {
     const remoteHosts = options.remoteHosts ?? [DEFAULT_REMOTE_HOST];
     const load = options.loadImpl ?? createDefaultTransformersLoadImpl(options.moduleEntryPath);
     let backendPromise = null;
+    let disposed = false;
+    let disposePromise;
+    const activeCalls = new Set();
     const loadBackend = () => {
         const existing = backendPromise;
         if (existing) {
@@ -120,6 +124,8 @@ export function createTransformersEmbeddingProvider(options) {
                 }
                 catch (error) {
                     lastError = error;
+                    if (disposed)
+                        throw error;
                     // 镜像轮换前清 0 字节残骸: 失败下载留下的空文件会被 transformers
                     // 当作缓存命中, 毒化下一个 host 的重试 (model-cache-hygiene)。
                     purgeZeroByteCacheArtifacts(join(cacheDir, modelId));
@@ -130,10 +136,14 @@ export function createTransformersEmbeddingProvider(options) {
                 // 进程内换 host 也拿同一空清单。
                 try {
                     await new Promise((resolve) => setTimeout(resolve, 2_000));
+                    if (disposed)
+                        throw lastError;
                     return await load(modelId, cacheDir, remoteHost);
                 }
                 catch (error) {
                     lastError = error;
+                    if (disposed)
+                        throw error;
                     purgeZeroByteCacheArtifacts(join(cacheDir, modelId));
                 }
             }
@@ -150,26 +160,47 @@ export function createTransformersEmbeddingProvider(options) {
         });
         return attempt;
     };
+    const embed = async (texts) => {
+        if (disposed)
+            throw new Error("memory embedding provider is disposed");
+        if (texts.length === 0)
+            return [];
+        const backend = await loadBackend();
+        if (disposed)
+            throw new Error("memory embedding provider is disposed");
+        const vectors = [];
+        for (let offset = 0; offset < texts.length; offset += batchSize) {
+            if (disposed)
+                throw new Error("memory embedding provider is disposed");
+            const batchTexts = texts.slice(offset, offset + batchSize);
+            const rows = (await backend.extract(batchTexts)).toList();
+            for (const row of rows) {
+                if (row.length !== dimensions) {
+                    throw new Error(`embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`);
+                }
+                vectors.push(row);
+            }
+        }
+        return vectors;
+    };
     return {
         modelId,
         dimensions,
-        embed: async (texts) => {
-            if (texts.length === 0) {
-                return [];
-            }
-            const backend = await loadBackend();
-            const vectors = [];
-            for (let offset = 0; offset < texts.length; offset += batchSize) {
-                const batchTexts = texts.slice(offset, offset + batchSize);
-                const rows = (await backend.extract(batchTexts)).toList();
-                for (const row of rows) {
-                    if (row.length !== dimensions) {
-                        throw new Error(`embedding dimension mismatch: model ${modelId} returned ${row.length} dims, provider declares ${dimensions}`);
-                    }
-                    vectors.push(row);
-                }
-            }
-            return vectors;
+        embed: (texts) => {
+            const call = embed(texts);
+            activeCalls.add(call);
+            void call.then(() => activeCalls.delete(call), () => activeCalls.delete(call));
+            return call;
+        },
+        dispose: () => {
+            disposed = true;
+            disposePromise ??= (async () => {
+                const pendingBackend = backendPromise;
+                await Promise.allSettled([...activeCalls]);
+                const backend = await pendingBackend?.catch(() => undefined);
+                await backend?.dispose?.();
+            })();
+            return disposePromise;
         },
     };
 }