@anuma/sdk 3.9.4 → 3.9.6

package/dist/server/{chunk-FJFPCJRI.mjs → chunk-FXS3EUKX.mjs}

@@ -18676,24 +18676,25 @@ function tokenize(text) {
   if (!m) return [];
   return m.filter((t) => !STOPWORDS.has(t));
 }
-function scoreBM25(query, items) {
-  const queryTerms = tokenize(query);
-  if (queryTerms.length === 0 || items.length === 0) return /* @__PURE__ */ new Map();
+function prepareBM25Corpus(items) {
+  const ids = items.map((i) => i.id);
   const docTokens = items.map((i) => tokenize(i.content));
   const docLens = docTokens.map((t) => t.length);
-  const avgdl = docLens.reduce((a, b) => a + b, 0) / docLens.length;
-  const queryTermSet = new Set(queryTerms);
+  const avgdl = docLens.length ? docLens.reduce((a, b) => a + b, 0) / docLens.length : 0;
   const df = /* @__PURE__ */ new Map();
   for (const tokens of docTokens) {
-    const seen = /* @__PURE__ */ new Set();
-    for (const t of tokens) {
-      if (queryTermSet.has(t)) seen.add(t);
-    }
+    const seen = new Set(tokens);
     for (const t of seen) df.set(t, (df.get(t) ?? 0) + 1);
   }
-  const N = items.length;
+  return { ids, docTokens, docLens, avgdl, df, n: items.length };
+}
+function scoreBM25Prepared(query, corpus) {
+  const queryTerms = tokenize(query);
   const scores = /* @__PURE__ */ new Map();
-  for (let i = 0; i < items.length; i++) {
+  if (queryTerms.length === 0 || corpus.n === 0) return scores;
+  const queryTermSet = new Set(queryTerms);
+  const { ids, docTokens, docLens, avgdl, df, n: N } = corpus;
+  for (let i = 0; i < N; i++) {
     const tokens = docTokens[i];
     const dl = docLens[i];
     if (dl === 0) continue;
@@ -18709,10 +18710,15 @@ function scoreBM25(query, items) {
       const tfNorm = f * (K1 + 1) / (f + K1 * (1 - B + B * dl / avgdl));
       score += idf * tfNorm;
     }
-    scores.set(items[i].id, score);
+    scores.set(ids[i], score);
   }
   return scores;
 }
+function scoreBM25(query, items) {
+  if (items.length === 0) return /* @__PURE__ */ new Map();
+  if (tokenize(query).length === 0) return /* @__PURE__ */ new Map();
+  return scoreBM25Prepared(query, prepareBM25Corpus(items));
+}
 
 // src/lib/memoryVault/decomposeQuery.ts
 var DEFAULT_MODEL = "inclusionai/ling-2.6-flash";
@@ -18967,7 +18973,7 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
     }
   });
   const baseIds = new Set(baseRanked.map((r) => r.uniqueId));
-  const bm25Scores = scoreBM25(
+  const bm25Scores = options?.preparedBM25Corpus ? scoreBM25Prepared(query, options.preparedBM25Corpus) : scoreBM25(
     query,
     items.map((i) => ({ id: i.id, content: i.content }))
   );
@@ -19164,7 +19170,11 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
   const limit = options?.limit ?? 5;
   const perFacetTopN = options?.perFacetTopN ?? 10;
   if (items.length === 0 || subQueries.length === 0) return [];
+  const preparedBM25Corpus = prepareBM25Corpus(
+    items.map((i) => ({ id: i.id, content: i.content }))
+  );
   const facetTuning = {
+    preparedBM25Corpus,
     ...options?.recencyAlpha !== void 0 && { recencyAlpha: options.recencyAlpha },
     ...options?.recency && { recency: options.recency },
     ...options?.supersessionBoost !== void 0 && {
@@ -19474,13 +19484,12 @@ async function buildProjectedCorpus(query, vaultCtx, embeddingOptions, cache, qu
   }));
   return { memories, embeddedItems, queryEmbedding, vaultSize };
 }
-async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, cache, searchOptions) {
-  const limit = searchOptions?.limit ?? 5;
-  const minSimilarity = searchOptions?.minSimilarity ?? 0.1;
-  const scopes = searchOptions?.scopes;
+async function prepareVaultCandidates(query, vaultCtx, embeddingOptions, cache, searchOptions) {
   if (!query || typeof query !== "string") {
-    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+    return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: 0 };
   }
+  const limit = searchOptions?.limit ?? 5;
+  const scopes = searchOptions?.scopes;
   const folderId = searchOptions?.folderId;
   const queryOpts = {};
   if (scopes?.length) queryOpts.scopes = scopes;
@@ -19507,11 +19516,11 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       ...forceIncludeIds.length > 0 && { forceIncludeIds }
     });
     if (corpus.vaultSize === 0) {
-      return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: 0 };
     }
     ({ memories, embeddedItems, queryEmbedding, vaultSize } = corpus);
     if (memories.length === 0) {
-      return { results: [], vaultSize, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize };
     }
   } else {
     const loaded = await getAllVaultMemoriesOp(
@@ -19526,7 +19535,7 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       );
     }
     if (memories.length === 0) {
-      return { results: [], vaultSize: loaded.length, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: loaded.length };
     }
     queryEmbedding = await generateEmbedding(query, embeddingOptions);
     const currentModel = embeddingOptions.model ?? DEFAULT_API_EMBEDDING_MODEL;
@@ -19591,6 +19600,12 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       factType: m.factType
     }));
   }
+  return { memories, embeddedItems, queryEmbedding, vaultSize };
+}
+async function rankPreparedVaultCandidates(query, prepared, embeddingOptions, searchOptions) {
+  const limit = searchOptions?.limit ?? 5;
+  const minSimilarity = searchOptions?.minSimilarity ?? 0.1;
+  const { memories, embeddedItems, queryEmbedding, vaultSize } = prepared;
   if (queryEmbedding.length > 0) {
     const mismatched = embeddedItems.filter(
       (it) => it.embedding.length > 0 && it.embedding.length !== queryEmbedding.length
@@ -19728,6 +19743,25 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
   });
   return stampTimestamps({ results, vaultSize, hadV2Head: results.length > 0 });
 }
+async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, cache, searchOptions) {
+  if (!query || typeof query !== "string") {
+    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+  }
+  const prepared = await prepareVaultCandidates(
+    query,
+    vaultCtx,
+    embeddingOptions,
+    cache,
+    searchOptions
+  );
+  if (prepared.vaultSize === 0) {
+    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+  }
+  if (prepared.memories.length === 0) {
+    return { results: [], vaultSize: prepared.vaultSize, reranked: false, hadV2Head: false };
+  }
+  return rankPreparedVaultCandidates(query, prepared, embeddingOptions, searchOptions);
+}
 async function searchVaultMemories(query, vaultCtx, embeddingOptions, cache, searchOptions) {
   const { results } = await searchVaultMemoriesWithSize(
     query,
@@ -19815,7 +19849,7 @@ function createMemoryVaultSearchTool(vaultCtx, embeddingOptions, cache, searchOp
             legacy.map((r) => ({ id: r.uniqueId, content: r.content, score: r.similarity }))
           );
         }
-        const { recall: recall2 } = await import("./recall-23VTTAEG.mjs");
+        const { recall: recall2 } = await import("./recall-VZO4USXK.mjs");
         const result = await recall2(
           query,
           { vaultCtx, embeddingOptions, vaultCache: cache },
@@ -20846,6 +20880,8 @@ export {
   recall,
   preEmbedVaultMemories,
   eagerEmbedContent,
+  prepareVaultCandidates,
+  rankPreparedVaultCandidates,
   searchVaultMemories,
   createMemoryVaultSearchTool
 };

package/dist/server/{consolidate-BKLNVH5Y.mjs → consolidate-YCD45CXO.mjs}

@@ -8,7 +8,7 @@ import {
 import "./chunk-3RPGXXPZ.mjs";
 
 // src/lib/memory/consolidate.ts
-var DEFAULT_MODEL = "inclusionai/ling-2.6-flash";
+var DEFAULT_CONSOLIDATION_MODEL = "inclusionai/ling-2.6-flash";
 var DEFAULT_CONSOLIDATE_ATTEMPTS = 3;
 var DEFAULT_CONSOLIDATE_TOTAL_TIMEOUT_MS = 2e4;
 var SYSTEM_PROMPT = `You consolidate a new memory against existing memories from the same user.
@@ -70,7 +70,7 @@ ${candidateText}`;
       ...options.apiKey !== void 0 && { apiKey: options.apiKey },
       ...options.getToken !== void 0 && { getToken: options.getToken },
       ...options.baseUrl !== void 0 && { baseUrl: options.baseUrl },
-      model: options.model ?? DEFAULT_MODEL,
+      model: options.model ?? DEFAULT_CONSOLIDATION_MODEL,
       systemPrompt: SYSTEM_PROMPT,
       userMessage,
       tag: "memory/consolidate",
@@ -147,5 +147,6 @@ function validate(parsed, newContent, validIds) {
   return { action: "update", targetId, content: c };
 }
 export {
+  DEFAULT_CONSOLIDATION_MODEL,
   consolidateMemory
 };

package/dist/server/index.cjs

@@ -19494,24 +19494,25 @@ function tokenize(text12) {
   if (!m) return [];
   return m.filter((t) => !STOPWORDS.has(t));
 }
-function scoreBM25(query, items) {
-  const queryTerms = tokenize(query);
-  if (queryTerms.length === 0 || items.length === 0) return /* @__PURE__ */ new Map();
+function prepareBM25Corpus(items) {
+  const ids = items.map((i) => i.id);
   const docTokens = items.map((i) => tokenize(i.content));
   const docLens = docTokens.map((t) => t.length);
-  const avgdl = docLens.reduce((a, b) => a + b, 0) / docLens.length;
-  const queryTermSet = new Set(queryTerms);
+  const avgdl = docLens.length ? docLens.reduce((a, b) => a + b, 0) / docLens.length : 0;
   const df = /* @__PURE__ */ new Map();
   for (const tokens of docTokens) {
-    const seen = /* @__PURE__ */ new Set();
-    for (const t of tokens) {
-      if (queryTermSet.has(t)) seen.add(t);
-    }
+    const seen = new Set(tokens);
     for (const t of seen) df.set(t, (df.get(t) ?? 0) + 1);
   }
-  const N = items.length;
+  return { ids, docTokens, docLens, avgdl, df, n: items.length };
+}
+function scoreBM25Prepared(query, corpus) {
+  const queryTerms = tokenize(query);
   const scores = /* @__PURE__ */ new Map();
-  for (let i = 0; i < items.length; i++) {
+  if (queryTerms.length === 0 || corpus.n === 0) return scores;
+  const queryTermSet = new Set(queryTerms);
+  const { ids, docTokens, docLens, avgdl, df, n: N } = corpus;
+  for (let i = 0; i < N; i++) {
     const tokens = docTokens[i];
     const dl = docLens[i];
     if (dl === 0) continue;
@@ -19527,10 +19528,15 @@ function scoreBM25(query, items) {
       const tfNorm = f * (K1 + 1) / (f + K1 * (1 - B + B * dl / avgdl));
       score += idf * tfNorm;
     }
-    scores.set(items[i].id, score);
+    scores.set(ids[i], score);
   }
   return scores;
 }
+function scoreBM25(query, items) {
+  if (items.length === 0) return /* @__PURE__ */ new Map();
+  if (tokenize(query).length === 0) return /* @__PURE__ */ new Map();
+  return scoreBM25Prepared(query, prepareBM25Corpus(items));
+}
 var K1, B, TOKEN_RE, STOPWORDS;
 var init_bm25 = __esm({
   "src/lib/memoryVault/bm25.ts"() {
@@ -21098,7 +21104,7 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
     }
   });
   const baseIds = new Set(baseRanked.map((r) => r.uniqueId));
-  const bm25Scores = scoreBM25(
+  const bm25Scores = options?.preparedBM25Corpus ? scoreBM25Prepared(query, options.preparedBM25Corpus) : scoreBM25(
     query,
     items.map((i) => ({ id: i.id, content: i.content }))
   );
@@ -21295,7 +21301,11 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
   const limit = options?.limit ?? 5;
   const perFacetTopN = options?.perFacetTopN ?? 10;
   if (items.length === 0 || subQueries.length === 0) return [];
+  const preparedBM25Corpus = prepareBM25Corpus(
+    items.map((i) => ({ id: i.id, content: i.content }))
+  );
   const facetTuning = {
+    preparedBM25Corpus,
     ...options?.recencyAlpha !== void 0 && { recencyAlpha: options.recencyAlpha },
     ...options?.recency && { recency: options.recency },
     ...options?.supersessionBoost !== void 0 && {
@@ -21605,13 +21615,12 @@ async function buildProjectedCorpus(query, vaultCtx, embeddingOptions, cache, qu
   }));
   return { memories, embeddedItems, queryEmbedding, vaultSize };
 }
-async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, cache, searchOptions) {
-  const limit = searchOptions?.limit ?? 5;
-  const minSimilarity = searchOptions?.minSimilarity ?? 0.1;
-  const scopes = searchOptions?.scopes;
+async function prepareVaultCandidates(query, vaultCtx, embeddingOptions, cache, searchOptions) {
   if (!query || typeof query !== "string") {
-    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+    return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: 0 };
   }
+  const limit = searchOptions?.limit ?? 5;
+  const scopes = searchOptions?.scopes;
   const folderId = searchOptions?.folderId;
   const queryOpts = {};
   if (scopes?.length) queryOpts.scopes = scopes;
@@ -21638,11 +21647,11 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       ...forceIncludeIds.length > 0 && { forceIncludeIds }
     });
     if (corpus.vaultSize === 0) {
-      return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: 0 };
     }
     ({ memories, embeddedItems, queryEmbedding, vaultSize } = corpus);
     if (memories.length === 0) {
-      return { results: [], vaultSize, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize };
     }
   } else {
     const loaded = await getAllVaultMemoriesOp(
@@ -21657,7 +21666,7 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       );
     }
     if (memories.length === 0) {
-      return { results: [], vaultSize: loaded.length, reranked: false, hadV2Head: false };
+      return { memories: [], embeddedItems: [], queryEmbedding: [], vaultSize: loaded.length };
     }
     queryEmbedding = await generateEmbedding(query, embeddingOptions);
     const currentModel = embeddingOptions.model ?? DEFAULT_API_EMBEDDING_MODEL;
@@ -21722,6 +21731,12 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
       factType: m.factType
     }));
   }
+  return { memories, embeddedItems, queryEmbedding, vaultSize };
+}
+async function rankPreparedVaultCandidates(query, prepared, embeddingOptions, searchOptions) {
+  const limit = searchOptions?.limit ?? 5;
+  const minSimilarity = searchOptions?.minSimilarity ?? 0.1;
+  const { memories, embeddedItems, queryEmbedding, vaultSize } = prepared;
   if (queryEmbedding.length > 0) {
     const mismatched = embeddedItems.filter(
       (it) => it.embedding.length > 0 && it.embedding.length !== queryEmbedding.length
@@ -21859,6 +21874,25 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
   });
   return stampTimestamps({ results, vaultSize, hadV2Head: results.length > 0 });
 }
+async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, cache, searchOptions) {
+  if (!query || typeof query !== "string") {
+    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+  }
+  const prepared = await prepareVaultCandidates(
+    query,
+    vaultCtx,
+    embeddingOptions,
+    cache,
+    searchOptions
+  );
+  if (prepared.vaultSize === 0) {
+    return { results: [], vaultSize: 0, reranked: false, hadV2Head: false };
+  }
+  if (prepared.memories.length === 0) {
+    return { results: [], vaultSize: prepared.vaultSize, reranked: false, hadV2Head: false };
+  }
+  return rankPreparedVaultCandidates(query, prepared, embeddingOptions, searchOptions);
+}
 async function searchVaultMemories(query, vaultCtx, embeddingOptions, cache, searchOptions) {
   const { results } = await searchVaultMemoriesWithSize(
     query,
@@ -22669,6 +22703,7 @@ var init_redactor = __esm({
 // src/lib/memory/consolidate.ts
 var consolidate_exports = {};
 __export(consolidate_exports, {
+  DEFAULT_CONSOLIDATION_MODEL: () => DEFAULT_CONSOLIDATION_MODEL,
   consolidateMemory: () => consolidateMemory
 });
 async function consolidateMemory(newContent, candidates, options) {
@@ -22697,7 +22732,7 @@ ${candidateText}`;
       ...options.apiKey !== void 0 && { apiKey: options.apiKey },
       ...options.getToken !== void 0 && { getToken: options.getToken },
       ...options.baseUrl !== void 0 && { baseUrl: options.baseUrl },
-      model: options.model ?? DEFAULT_MODEL3,
+      model: options.model ?? DEFAULT_CONSOLIDATION_MODEL,
       systemPrompt: SYSTEM_PROMPT3,
       userMessage,
       tag: "memory/consolidate",
@@ -22773,14 +22808,14 @@ function validate2(parsed, newContent, validIds) {
   if (!targetId || !validIds.has(targetId)) return null;
   return { action: "update", targetId, content: c };
 }
-var DEFAULT_MODEL3, DEFAULT_CONSOLIDATE_ATTEMPTS, DEFAULT_CONSOLIDATE_TOTAL_TIMEOUT_MS, SYSTEM_PROMPT3;
+var DEFAULT_CONSOLIDATION_MODEL, DEFAULT_CONSOLIDATE_ATTEMPTS, DEFAULT_CONSOLIDATE_TOTAL_TIMEOUT_MS, SYSTEM_PROMPT3;
 var init_consolidate = __esm({
   "src/lib/memory/consolidate.ts"() {
     "use strict";
     init_logger();
     init_redactor();
     init_portalLlm();
-    DEFAULT_MODEL3 = "inclusionai/ling-2.6-flash";
+    DEFAULT_CONSOLIDATION_MODEL = "inclusionai/ling-2.6-flash";
     DEFAULT_CONSOLIDATE_ATTEMPTS = 3;
     DEFAULT_CONSOLIDATE_TOTAL_TIMEOUT_MS = 2e4;
     SYSTEM_PROMPT3 = `You consolidate a new memory against existing memories from the same user.
@@ -27639,9 +27674,25 @@ async function retain(content, ctx, options = {}) {
   const resolvedScope = options.scope ?? DEFAULT_SCOPE;
   let supersedeTargetIds = [];
   let supersedeContent;
+  let prepared;
   if (enableAutoMerge) {
+    prepared = await prepareVaultCandidates(
+      trimmed,
+      ctx.vaultCtx,
+      ctx.embeddingOptions,
+      ctx.vaultCache,
+      {
+        limit: Math.max(options.consolidateTopK ?? DEFAULT_CONSOLIDATE_TOP_K, 1),
+        useFusion: false,
+        scopes: [resolvedScope],
+        // PR5 — include archived rows as merge candidates so a re-observed fact
+        // resurrects (un-archives) the decayed row instead of duplicating it.
+        includeArchived: true,
+        ...options.folderId !== void 0 && { folderId: options.folderId }
+      }
+    );
     if (options.consolidateOptions) {
-      const outcome = await tryConsolidate(trimmed, ctx, options);
+      const outcome = await tryConsolidate(trimmed, ctx, options, prepared);
       if (outcome) {
         if ("done" in outcome) return outcome.done;
         supersedeTargetIds = outcome.supersede;
@@ -27649,23 +27700,14 @@ async function retain(content, ctx, options = {}) {
       }
     }
     if (supersedeTargetIds.length === 0) {
-      const matches = await searchVaultMemories(
+      const { results: matches } = await rankPreparedVaultCandidates(
         trimmed,
-        ctx.vaultCtx,
+        prepared,
         ctx.embeddingOptions,
-        ctx.vaultCache,
         {
           limit: 1,
           minSimilarity: threshold,
-          useFusion: false,
-          scopes: [resolvedScope],
-          // PR5 — include archived rows as merge candidates so a re-observed
-          // fact resurrects (un-archives) the decayed row instead of creating a
-          // fresh duplicate. The resurrection is applied on the merge write
-          // below, but ONLY for a row that is not superseded/deleted (the
-          // guards below preserve main's tombstone/supersession suppression).
-          includeArchived: true,
-          ...options.folderId !== void 0 && { folderId: options.folderId }
+          useFusion: false
         }
       );
       if (matches.length > 0) {
@@ -27713,7 +27755,8 @@ async function retain(content, ctx, options = {}) {
     }
   }
   const contentToWrite = supersedeContent ?? trimmed;
-  const embedding = await generateEmbedding(contentToWrite, ctx.embeddingOptions);
+  const reusableQueryEmbedding = supersedeContent === void 0 && prepared && prepared.queryEmbedding.length > 0 ? prepared.queryEmbedding : void 0;
+  const embedding = reusableQueryEmbedding ?? await generateEmbedding(contentToWrite, ctx.embeddingOptions);
   const embeddingModel = ctx.embeddingOptions.model ?? DEFAULT_API_EMBEDDING_MODEL;
   if (options.respectTombstones) {
     const tombstoneId = await findTombstoneMatch(embedding, embeddingModel, resolvedScope, ctx, {
@@ -27842,26 +27885,19 @@ function pickFactTypeUpdate(existing, incoming) {
 function resurrectFields(existing) {
   return typeof existing.archivedAt === "number" ? { restore: true } : { preserveUpdatedAt: true };
 }
-async function tryConsolidate(trimmed, ctx, options) {
+async function tryConsolidate(trimmed, ctx, options, prepared) {
   const consolidateOptions = options.consolidateOptions;
   if (!consolidateOptions) return null;
   const consolidateThreshold = options.consolidateThreshold ?? DEFAULT_CONSOLIDATE_THRESHOLD;
   const topK = options.consolidateTopK ?? DEFAULT_CONSOLIDATE_TOP_K;
-  const resolvedScope = options.scope ?? DEFAULT_SCOPE;
-  const matches = await searchVaultMemories(
+  const { results: matches } = await rankPreparedVaultCandidates(
     trimmed,
-    ctx.vaultCtx,
+    prepared,
     ctx.embeddingOptions,
-    ctx.vaultCache,
     {
       limit: topK,
       minSimilarity: consolidateThreshold,
-      useFusion: false,
-      scopes: [resolvedScope],
-      // PR5 — archived rows are consolidation candidates too, so a paraphrased
-      // re-observation resurrects a decayed row rather than duplicating it.
-      includeArchived: true,
-      ...options.folderId !== void 0 && { folderId: options.folderId }
+      useFusion: false
     }
   );
   if (matches.length === 0) return null;
@@ -28644,7 +28680,7 @@ function classifyDecay(m, now, policy) {
 init_logger();
 init_redactor();
 init_portalLlm();
-var DEFAULT_MODEL4 = "inclusionai/ling-2.6-flash";
+var DEFAULT_MODEL3 = "inclusionai/ling-2.6-flash";
 var DEFAULT_ATTEMPTS2 = 2;
 var DEFAULT_TOTAL_TIMEOUT_MS2 = 12e3;
 var MAX_CONTENT_CHARS = 400;
@@ -28683,7 +28719,7 @@ function createLlmDecayClassifier(options) {
           ...options.apiKey !== void 0 && { apiKey: options.apiKey },
           ...options.getToken !== void 0 && { getToken: options.getToken },
           ...options.baseUrl !== void 0 && { baseUrl: options.baseUrl },
-          model: options.model ?? DEFAULT_MODEL4,
+          model: options.model ?? DEFAULT_MODEL3,
           systemPrompt: SYSTEM_PROMPT5,
           userMessage: `Memory content:
   ${safe}
@@ -29116,7 +29152,7 @@ init_logger();
 init_portalLlm();
 init_recall();
 var DEFAULT_BASE_URL2 = BASE_URL;
-var DEFAULT_MODEL5 = "anthropic/claude-sonnet-4-6";
+var DEFAULT_MODEL4 = "anthropic/claude-sonnet-4-6";
 var DEFAULT_MAX_TOKENS = 4096;
 var REQUEST_TIMEOUT_MS = 6e4;
 var DEFAULT_SYSTEM_PROMPT = `You are a personal assistant with access to the user's memory. Answer the user's question using the supplied memories as evidence.
@@ -29148,7 +29184,7 @@ async function reflect(query, ctx, options) {
   const evidence = recalledMemories.map((m, i) => `[${i + 1}] (id: ${m.id}, kind: ${m.kind})
 ${m.content}`).join("\n\n");
   const baseUrl = options.baseUrl ?? DEFAULT_BASE_URL2;
-  const model = options.llmModel ?? DEFAULT_MODEL5;
+  const model = options.llmModel ?? DEFAULT_MODEL4;
   const wantsStructured = !!options.responseSchema;
   const sendResponseFormat = wantsStructured && supportsResponseFormat(model, "json_schema");
   const basePrompt = options.systemPrompt ?? DEFAULT_SYSTEM_PROMPT;

package/dist/server/index.mjs

@@ -88,6 +88,8 @@ import {
   onClearAllEncryptionState,
   onKeyAvailable,
   preEmbedVaultMemories,
+  prepareVaultCandidates,
+  rankPreparedVaultCandidates,
   recall,
   replaceMemoryEntitiesGuardedOp,
   requestEncryptionKey,
@@ -125,7 +127,7 @@ import {
   updateVaultMemoryOp,
   upsertMessageOp,
   vaultMemoryToStored
-} from "./chunk-FJFPCJRI.mjs";
+} from "./chunk-FXS3EUKX.mjs";
 import {
   createStreamingDeAnonymizer,
   resolvePiiRedactor
@@ -5265,9 +5267,25 @@ async function retain(content, ctx, options = {}) {
   const resolvedScope = options.scope ?? DEFAULT_SCOPE;
   let supersedeTargetIds = [];
   let supersedeContent;
+  let prepared;
   if (enableAutoMerge) {
+    prepared = await prepareVaultCandidates(
+      trimmed,
+      ctx.vaultCtx,
+      ctx.embeddingOptions,
+      ctx.vaultCache,
+      {
+        limit: Math.max(options.consolidateTopK ?? DEFAULT_CONSOLIDATE_TOP_K, 1),
+        useFusion: false,
+        scopes: [resolvedScope],
+        // PR5 — include archived rows as merge candidates so a re-observed fact
+        // resurrects (un-archives) the decayed row instead of duplicating it.
+        includeArchived: true,
+        ...options.folderId !== void 0 && { folderId: options.folderId }
+      }
+    );
     if (options.consolidateOptions) {
-      const outcome = await tryConsolidate(trimmed, ctx, options);
+      const outcome = await tryConsolidate(trimmed, ctx, options, prepared);
       if (outcome) {
         if ("done" in outcome) return outcome.done;
         supersedeTargetIds = outcome.supersede;
@@ -5275,23 +5293,14 @@ async function retain(content, ctx, options = {}) {
       }
     }
     if (supersedeTargetIds.length === 0) {
-      const matches = await searchVaultMemories(
+      const { results: matches } = await rankPreparedVaultCandidates(
         trimmed,
-        ctx.vaultCtx,
+        prepared,
         ctx.embeddingOptions,
-        ctx.vaultCache,
         {
           limit: 1,
           minSimilarity: threshold,
-          useFusion: false,
-          scopes: [resolvedScope],
-          // PR5 — include archived rows as merge candidates so a re-observed
-          // fact resurrects (un-archives) the decayed row instead of creating a
-          // fresh duplicate. The resurrection is applied on the merge write
-          // below, but ONLY for a row that is not superseded/deleted (the
-          // guards below preserve main's tombstone/supersession suppression).
-          includeArchived: true,
-          ...options.folderId !== void 0 && { folderId: options.folderId }
+          useFusion: false
         }
       );
       if (matches.length > 0) {
@@ -5339,7 +5348,8 @@ async function retain(content, ctx, options = {}) {
     }
   }
   const contentToWrite = supersedeContent ?? trimmed;
-  const embedding = await generateEmbedding(contentToWrite, ctx.embeddingOptions);
+  const reusableQueryEmbedding = supersedeContent === void 0 && prepared && prepared.queryEmbedding.length > 0 ? prepared.queryEmbedding : void 0;
+  const embedding = reusableQueryEmbedding ?? await generateEmbedding(contentToWrite, ctx.embeddingOptions);
   const embeddingModel = ctx.embeddingOptions.model ?? DEFAULT_API_EMBEDDING_MODEL;
   if (options.respectTombstones) {
     const tombstoneId = await findTombstoneMatch(embedding, embeddingModel, resolvedScope, ctx, {
@@ -5468,26 +5478,19 @@ function pickFactTypeUpdate(existing, incoming) {
 function resurrectFields(existing) {
   return typeof existing.archivedAt === "number" ? { restore: true } : { preserveUpdatedAt: true };
 }
-async function tryConsolidate(trimmed, ctx, options) {
+async function tryConsolidate(trimmed, ctx, options, prepared) {
   const consolidateOptions = options.consolidateOptions;
   if (!consolidateOptions) return null;
   const consolidateThreshold = options.consolidateThreshold ?? DEFAULT_CONSOLIDATE_THRESHOLD;
   const topK = options.consolidateTopK ?? DEFAULT_CONSOLIDATE_TOP_K;
-  const resolvedScope = options.scope ?? DEFAULT_SCOPE;
-  const matches = await searchVaultMemories(
+  const { results: matches } = await rankPreparedVaultCandidates(
     trimmed,
-    ctx.vaultCtx,
+    prepared,
     ctx.embeddingOptions,
-    ctx.vaultCache,
     {
       limit: topK,
       minSimilarity: consolidateThreshold,
-      useFusion: false,
-      scopes: [resolvedScope],
-      // PR5 — archived rows are consolidation candidates too, so a paraphrased
-      // re-observation resurrects a decayed row rather than duplicating it.
-      includeArchived: true,
-      ...options.folderId !== void 0 && { folderId: options.folderId }
+      useFusion: false
     }
   );
   if (matches.length === 0) return null;
@@ -5496,7 +5499,7 @@ async function tryConsolidate(trimmed, ctx, options) {
     content: m.content,
     similarity: m.similarity
   }));
-  const { consolidateMemory: doConsolidate } = await import("./consolidate-BKLNVH5Y.mjs");
+  const { consolidateMemory: doConsolidate } = await import("./consolidate-YCD45CXO.mjs");
   const decision = await doConsolidate(trimmed, candidates, consolidateOptions);
   if (decision.action === "create") return null;
   if (decision.action === "supersede" && decision.content) {

package/dist/server/{recall-23VTTAEG.mjs → recall-VZO4USXK.mjs}

@@ -1,6 +1,6 @@
 import {
   recall
-} from "./chunk-FJFPCJRI.mjs";
+} from "./chunk-FXS3EUKX.mjs";
 import "./chunk-MIXEASUP.mjs";
 import "./chunk-3RPGXXPZ.mjs";
 export {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anuma/sdk",
-  "version": "3.9.4",
+  "version": "3.9.6",
   "description": "",
   "main": "./dist/index.cjs",
   "module": "./dist/index.mjs",