@anuma/sdk 7.0.4 → 7.0.5

package/dist/server/index.mjs

@@ -18478,6 +18478,22 @@ var postApiV1Embeddings = (options) => (options.client ?? client).post({
   }
 });
 
+// src/lib/memoryEngine/embeddingCacheState.ts
+var EMBEDDING_CACHE_STATE_KEY = /* @__PURE__ */ Symbol.for("@anuma/sdk/embedding-cache/v1");
+var cacheGlobal = globalThis;
+var fallbackState = {
+  scopedViews: /* @__PURE__ */ new WeakMap(),
+  inFlightByCache: /* @__PURE__ */ new WeakMap(),
+  cachedEmbeddingModels: /* @__PURE__ */ new WeakMap()
+};
+function getEmbeddingCacheState() {
+  try {
+    return cacheGlobal[EMBEDDING_CACHE_STATE_KEY] ?? (cacheGlobal[EMBEDDING_CACHE_STATE_KEY] = fallbackState);
+  } catch {
+    return fallbackState;
+  }
+}
+
 // src/lib/memoryEngine/generate.ts
 var EMBED_MAX_ATTEMPTS = 4;
 var DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS = 15e3;
@@ -18569,15 +18585,18 @@ async function generateEmbedding(text12, options) {
   if (total === void 0) return embedOne(text12, options, void 0);
   return withDeadline((signal) => embedOne(text12, options, signal), total, "embedding");
 }
-var inFlightByCache = /* @__PURE__ */ new WeakMap();
+var { inFlightByCache, cachedEmbeddingModels } = getEmbeddingCacheState();
 async function embedOne(text12, options, outer) {
   const { cache } = options;
   if (!cache) return Array.from(await requestEmbedding(text12, options, outer));
+  const model = options.model ?? DEFAULT_API_EMBEDDING_MODEL;
   const cached = cache.get(text12);
-  if (cached) return Array.from(cached);
+  const cachedModel = cached && cachedEmbeddingModels.get(cached);
+  if (cached && (cachedModel === void 0 || cachedModel === model)) return Array.from(cached);
   const inFlight = inFlightByCache.get(cache) ?? /* @__PURE__ */ new Map();
   inFlightByCache.set(cache, inFlight);
-  const pending = inFlight.get(text12);
+  const requestKey = JSON.stringify([model, text12]);
+  const pending = inFlight.get(requestKey);
   if (pending) {
     try {
       return Array.from(await pending);
@@ -18586,13 +18605,14 @@ async function embedOne(text12, options, outer) {
     }
   }
   const request = requestEmbedding(text12, options, outer);
-  inFlight.set(text12, request);
+  inFlight.set(requestKey, request);
   try {
     const embedding = await request;
+    cachedEmbeddingModels.set(embedding, model);
     cache.set(text12, embedding);
     return Array.from(embedding);
   } finally {
-    if (inFlight.get(text12) === request) inFlight.delete(text12);
+    if (inFlight.get(requestKey) === request) inFlight.delete(requestKey);
   }
 }
 async function requestEmbedding(text12, options, outer) {
@@ -18675,13 +18695,15 @@ async function embedMany(texts, options, outer) {
   const { baseUrl = BASE_URL, model, batchSize, cache } = options;
   const timeoutMs = options.timeoutMs ?? DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS;
   const chunkSize = batchSize ?? DEFAULT_EMBEDDING_BATCH_SIZE;
+  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   const results = new Array(texts.length).fill(null);
   const uncachedIndices = [];
   const uncachedTexts = [];
   for (let i = 0; i < texts.length; i++) {
     if (cache) {
       const cached = cache.get(texts[i]);
-      if (cached) {
+      const cachedModel = cached && cachedEmbeddingModels.get(cached);
+      if (cached && (cachedModel === void 0 || cachedModel === embeddingModel)) {
         results[i] = Array.from(cached);
         continue;
       }
@@ -18697,7 +18719,6 @@ async function embedMany(texts, options, outer) {
     ...taskTypeHeader(options.taskType)
   };
   if (outer?.aborted) throw outer.reason;
-  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   let newEmbeddings;
   if (uncachedTexts.length <= chunkSize) {
     newEmbeddings = await generateEmbeddingsBatch(
@@ -18745,6 +18766,7 @@ async function embedMany(texts, options, outer) {
     const f32Embedding = Float32Array.from(newEmbeddings[i]);
     results[uncachedIndices[i]] = Array.from(f32Embedding);
     if (cache) {
+      cachedEmbeddingModels.set(f32Embedding, embeddingModel);
       cache.set(uncachedTexts[i], f32Embedding);
     }
   }
@@ -32532,7 +32554,7 @@ var MaskScopedEmbeddingCache = class extends Map {
     return this;
   }
 };
-var scopedViews = /* @__PURE__ */ new WeakMap();
+var { scopedViews } = getEmbeddingCacheState();
 function maskScopedEmbeddingCache(cache, masked) {
   const views = scopedViews.get(cache) ?? {};
   scopedViews.set(cache, views);
@@ -32954,44 +32976,48 @@ async function recall(query, ctx, options = {}) {
       embeddingOptions,
       ctx.vaultCache,
       searchOptions
-    )))(query, {
-      limit: types.includes("chunk") ? Math.max(limit * 2, 16) : limit,
-      minSimilarity: vaultMinScore,
-      useFusion: true,
-      rerank: flags.rerank,
-      ...options.rerankTopN !== void 0 && { rerankTopN: options.rerankTopN },
-      ...options.ceWeight !== void 0 && { ceWeight: options.ceWeight },
-      ...options.rerankLoadTimeoutMs !== void 0 && {
-        rerankLoadTimeoutMs: options.rerankLoadTimeoutMs
-      },
-      ...options.recencyAlpha !== void 0 && { recencyAlpha: options.recencyAlpha },
-      ...options.recency && { recency: options.recency },
-      ...options.mmr !== void 0 && { mmr: options.mmr },
-      ...options.supersessionBoost !== void 0 && {
-        supersessionBoost: options.supersessionBoost
-      },
-      ...options.supersessionWindow !== void 0 && {
-        supersessionWindow: options.supersessionWindow
-      },
-      ...options.proofCountAlpha !== void 0 && {
-        proofCountAlpha: options.proofCountAlpha
-      },
-      ...options.bm25AdmissionDivisor !== void 0 && {
-        bm25AdmissionDivisor: options.bm25AdmissionDivisor
+    )))(
+      query,
+      {
+        limit: types.includes("chunk") ? Math.max(limit * 2, 16) : limit,
+        minSimilarity: vaultMinScore,
+        useFusion: true,
+        rerank: flags.rerank,
+        ...options.rerankTopN !== void 0 && { rerankTopN: options.rerankTopN },
+        ...options.ceWeight !== void 0 && { ceWeight: options.ceWeight },
+        ...options.rerankLoadTimeoutMs !== void 0 && {
+          rerankLoadTimeoutMs: options.rerankLoadTimeoutMs
+        },
+        ...options.recencyAlpha !== void 0 && { recencyAlpha: options.recencyAlpha },
+        ...options.recency && { recency: options.recency },
+        ...options.mmr !== void 0 && { mmr: options.mmr },
+        ...options.supersessionBoost !== void 0 && {
+          supersessionBoost: options.supersessionBoost
+        },
+        ...options.supersessionWindow !== void 0 && {
+          supersessionWindow: options.supersessionWindow
+        },
+        ...options.proofCountAlpha !== void 0 && {
+          proofCountAlpha: options.proofCountAlpha
+        },
+        ...options.bm25AdmissionDivisor !== void 0 && {
+          bm25AdmissionDivisor: options.bm25AdmissionDivisor
+        },
+        ...options.rrfK !== void 0 && { rrfK: options.rrfK },
+        ...options.decryptLast !== void 0 && { decryptLast: options.decryptLast },
+        ...subQueries && { subQueries },
+        ...options.scopes && { scopes: options.scopes },
+        ...options.folderId !== void 0 && { folderId: options.folderId },
+        ...options.factTypes?.length && { factTypes: options.factTypes },
+        ...options.memoryIds !== void 0 && { memoryIds: options.memoryIds },
+        ...options.factTypeWeights && { factTypeWeights: options.factTypeWeights },
+        ...entityRanking.length > 0 && { entityRanking },
+        ...temporalRanking.length > 0 && { temporalRanking },
+        ...needsChunkEmbedding && { queryEmbedding: queryEmbedding ?? [] },
+        queryEmbedTotalTimeoutMs
       },
-      ...options.rrfK !== void 0 && { rrfK: options.rrfK },
-      ...options.decryptLast !== void 0 && { decryptLast: options.decryptLast },
-      ...subQueries && { subQueries },
-      ...options.scopes && { scopes: options.scopes },
-      ...options.folderId !== void 0 && { folderId: options.folderId },
-      ...options.factTypes?.length && { factTypes: options.factTypes },
-      ...options.memoryIds !== void 0 && { memoryIds: options.memoryIds },
-      ...options.factTypeWeights && { factTypeWeights: options.factTypeWeights },
-      ...entityRanking.length > 0 && { entityRanking },
-      ...temporalRanking.length > 0 && { temporalRanking },
-      ...needsChunkEmbedding && { queryEmbedding: queryEmbedding ?? [] },
-      queryEmbedTotalTimeoutMs
-    });
+      options.embeddingCache
+    );
     factResults.push(
       ...dedupeBy(
         results,
@@ -34130,10 +34156,19 @@ function createLocalMemoryStore(options) {
       retainOptions
     ),
     factSource: {
-      search: (query, searchOptions) => searchVaultMemoriesWithSize(
+      search: (query, searchOptions, embeddingCache) => searchVaultMemoriesWithSize(
         query,
         vaultCtx,
-        { ...options.embeddingOptions, taskType: "memory_recall_embed" },
+        {
+          ...options.embeddingOptions,
+          taskType: "memory_recall_embed",
+          ...embeddingCache && !options.embeddingOptions.cache && {
+            cache: maskScopedEmbeddingCache(
+              embeddingCache,
+              Boolean(options.embeddingOptions.maskInput)
+            )
+          }
+        },
         vaultCache,
         searchOptions
       ),

package/dist/tools/selection/index.cjs

@@ -930,6 +930,22 @@ function taskTypeHeader(taskType) {
 // src/lib/memoryEngine/constants.ts
 var DEFAULT_API_EMBEDDING_MODEL = "qwen/qwen3-embedding-8b";
 
+// src/lib/memoryEngine/embeddingCacheState.ts
+var EMBEDDING_CACHE_STATE_KEY = /* @__PURE__ */ Symbol.for("@anuma/sdk/embedding-cache/v1");
+var cacheGlobal = globalThis;
+var fallbackState = {
+  scopedViews: /* @__PURE__ */ new WeakMap(),
+  inFlightByCache: /* @__PURE__ */ new WeakMap(),
+  cachedEmbeddingModels: /* @__PURE__ */ new WeakMap()
+};
+function getEmbeddingCacheState() {
+  try {
+    return cacheGlobal[EMBEDDING_CACHE_STATE_KEY] ?? (cacheGlobal[EMBEDDING_CACHE_STATE_KEY] = fallbackState);
+  } catch {
+    return fallbackState;
+  }
+}
+
 // src/lib/memoryEngine/generate.ts
 var EMBED_MAX_ATTEMPTS = 4;
 var DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS = 15e3;
@@ -1018,15 +1034,18 @@ async function generateEmbedding(text, options) {
   if (total === void 0) return embedOne(text, options, void 0);
   return withDeadline((signal) => embedOne(text, options, signal), total, "embedding");
 }
-var inFlightByCache = /* @__PURE__ */ new WeakMap();
+var { inFlightByCache, cachedEmbeddingModels } = getEmbeddingCacheState();
 async function embedOne(text, options, outer) {
   const { cache } = options;
   if (!cache) return Array.from(await requestEmbedding(text, options, outer));
+  const model = options.model ?? DEFAULT_API_EMBEDDING_MODEL;
   const cached = cache.get(text);
-  if (cached) return Array.from(cached);
+  const cachedModel = cached && cachedEmbeddingModels.get(cached);
+  if (cached && (cachedModel === void 0 || cachedModel === model)) return Array.from(cached);
   const inFlight = inFlightByCache.get(cache) ?? /* @__PURE__ */ new Map();
   inFlightByCache.set(cache, inFlight);
-  const pending = inFlight.get(text);
+  const requestKey = JSON.stringify([model, text]);
+  const pending = inFlight.get(requestKey);
   if (pending) {
     try {
       return Array.from(await pending);
@@ -1035,13 +1054,14 @@ async function embedOne(text, options, outer) {
     }
   }
   const request = requestEmbedding(text, options, outer);
-  inFlight.set(text, request);
+  inFlight.set(requestKey, request);
   try {
     const embedding = await request;
+    cachedEmbeddingModels.set(embedding, model);
     cache.set(text, embedding);
     return Array.from(embedding);
   } finally {
-    if (inFlight.get(text) === request) inFlight.delete(text);
+    if (inFlight.get(requestKey) === request) inFlight.delete(requestKey);
   }
 }
 async function requestEmbedding(text, options, outer) {
@@ -1124,13 +1144,15 @@ async function embedMany(texts, options, outer) {
   const { baseUrl = BASE_URL, model, batchSize, cache } = options;
   const timeoutMs = options.timeoutMs ?? DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS;
   const chunkSize = batchSize ?? DEFAULT_EMBEDDING_BATCH_SIZE;
+  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   const results = new Array(texts.length).fill(null);
   const uncachedIndices = [];
   const uncachedTexts = [];
   for (let i = 0; i < texts.length; i++) {
     if (cache) {
       const cached = cache.get(texts[i]);
-      if (cached) {
+      const cachedModel = cached && cachedEmbeddingModels.get(cached);
+      if (cached && (cachedModel === void 0 || cachedModel === embeddingModel)) {
         results[i] = Array.from(cached);
         continue;
       }
@@ -1146,7 +1168,6 @@ async function embedMany(texts, options, outer) {
     ...taskTypeHeader(options.taskType)
   };
   if (outer?.aborted) throw outer.reason;
-  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   let newEmbeddings;
   if (uncachedTexts.length <= chunkSize) {
     newEmbeddings = await generateEmbeddingsBatch(
@@ -1194,6 +1215,7 @@ async function embedMany(texts, options, outer) {
     const f32Embedding = Float32Array.from(newEmbeddings[i]);
     results[uncachedIndices[i]] = Array.from(f32Embedding);
     if (cache) {
+      cachedEmbeddingModels.set(f32Embedding, embeddingModel);
       cache.set(uncachedTexts[i], f32Embedding);
     }
   }

package/dist/tools/selection/index.mjs

@@ -808,6 +808,22 @@ function taskTypeHeader(taskType) {
 // src/lib/memoryEngine/constants.ts
 var DEFAULT_API_EMBEDDING_MODEL = "qwen/qwen3-embedding-8b";
 
+// src/lib/memoryEngine/embeddingCacheState.ts
+var EMBEDDING_CACHE_STATE_KEY = /* @__PURE__ */ Symbol.for("@anuma/sdk/embedding-cache/v1");
+var cacheGlobal = globalThis;
+var fallbackState = {
+  scopedViews: /* @__PURE__ */ new WeakMap(),
+  inFlightByCache: /* @__PURE__ */ new WeakMap(),
+  cachedEmbeddingModels: /* @__PURE__ */ new WeakMap()
+};
+function getEmbeddingCacheState() {
+  try {
+    return cacheGlobal[EMBEDDING_CACHE_STATE_KEY] ?? (cacheGlobal[EMBEDDING_CACHE_STATE_KEY] = fallbackState);
+  } catch {
+    return fallbackState;
+  }
+}
+
 // src/lib/memoryEngine/generate.ts
 var EMBED_MAX_ATTEMPTS = 4;
 var DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS = 15e3;
@@ -896,15 +912,18 @@ async function generateEmbedding(text, options) {
   if (total === void 0) return embedOne(text, options, void 0);
   return withDeadline((signal) => embedOne(text, options, signal), total, "embedding");
 }
-var inFlightByCache = /* @__PURE__ */ new WeakMap();
+var { inFlightByCache, cachedEmbeddingModels } = getEmbeddingCacheState();
 async function embedOne(text, options, outer) {
   const { cache } = options;
   if (!cache) return Array.from(await requestEmbedding(text, options, outer));
+  const model = options.model ?? DEFAULT_API_EMBEDDING_MODEL;
   const cached = cache.get(text);
-  if (cached) return Array.from(cached);
+  const cachedModel = cached && cachedEmbeddingModels.get(cached);
+  if (cached && (cachedModel === void 0 || cachedModel === model)) return Array.from(cached);
   const inFlight = inFlightByCache.get(cache) ?? /* @__PURE__ */ new Map();
   inFlightByCache.set(cache, inFlight);
-  const pending = inFlight.get(text);
+  const requestKey = JSON.stringify([model, text]);
+  const pending = inFlight.get(requestKey);
   if (pending) {
     try {
       return Array.from(await pending);
@@ -913,13 +932,14 @@ async function embedOne(text, options, outer) {
     }
   }
   const request = requestEmbedding(text, options, outer);
-  inFlight.set(text, request);
+  inFlight.set(requestKey, request);
   try {
     const embedding = await request;
+    cachedEmbeddingModels.set(embedding, model);
     cache.set(text, embedding);
     return Array.from(embedding);
   } finally {
-    if (inFlight.get(text) === request) inFlight.delete(text);
+    if (inFlight.get(requestKey) === request) inFlight.delete(requestKey);
   }
 }
 async function requestEmbedding(text, options, outer) {
@@ -1002,13 +1022,15 @@ async function embedMany(texts, options, outer) {
   const { baseUrl = BASE_URL, model, batchSize, cache } = options;
   const timeoutMs = options.timeoutMs ?? DEFAULT_EMBEDDING_REQUEST_TIMEOUT_MS;
   const chunkSize = batchSize ?? DEFAULT_EMBEDDING_BATCH_SIZE;
+  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   const results = new Array(texts.length).fill(null);
   const uncachedIndices = [];
   const uncachedTexts = [];
   for (let i = 0; i < texts.length; i++) {
     if (cache) {
       const cached = cache.get(texts[i]);
-      if (cached) {
+      const cachedModel = cached && cachedEmbeddingModels.get(cached);
+      if (cached && (cachedModel === void 0 || cachedModel === embeddingModel)) {
         results[i] = Array.from(cached);
         continue;
       }
@@ -1024,7 +1046,6 @@ async function embedMany(texts, options, outer) {
     ...taskTypeHeader(options.taskType)
   };
   if (outer?.aborted) throw outer.reason;
-  const embeddingModel = model ?? DEFAULT_API_EMBEDDING_MODEL;
   let newEmbeddings;
   if (uncachedTexts.length <= chunkSize) {
     newEmbeddings = await generateEmbeddingsBatch(
@@ -1072,6 +1093,7 @@ async function embedMany(texts, options, outer) {
     const f32Embedding = Float32Array.from(newEmbeddings[i]);
     results[uncachedIndices[i]] = Array.from(f32Embedding);
     if (cache) {
+      cachedEmbeddingModels.set(f32Embedding, embeddingModel);
       cache.set(uncachedTexts[i], f32Embedding);
     }
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anuma/sdk",
-  "version": "7.0.4",
+  "version": "7.0.5",
   "description": "",
   "main": "./dist/index.cjs",
   "module": "./dist/index.mjs",
@@ -154,7 +154,7 @@
     "check:excluded-suites": "node scripts/assert-excluded-suites-still-fail.mjs",
     "knip": "knip",
     "check:export-consumers": "node scripts/assert-export-consumers.mjs",
-    "check-exports": "publint && attw --pack --ignore-rules no-resolution false-cjs && node scripts/assert-no-client-self-import.mjs && node scripts/assert-shared-logger.mjs",
+    "check-exports": "publint && attw --pack --ignore-rules no-resolution false-cjs && node scripts/assert-no-client-self-import.mjs && node scripts/assert-shared-logger.mjs && node scripts/assert-embedding-cross-entrypoints.mjs",
     "check": "pnpm run lint && pnpm run knip && pnpm run check:export-consumers && pnpm run format && pnpm run generate",
     "eval:extraction": "tsx test/memory/src/extraction/benchmark.test.ts",
     "eval:topic": "tsx test/memory/src/topic/benchmark.ts",