@anuma/sdk 1.1.3 → 1.1.5

package/dist/expo/{chunk-3RXNQBS3.mjs → chunk-IKPPX3GY.mjs}

@@ -3170,6 +3170,20 @@ async function withEmbeddingRetry(call) {
   if (threw) throw lastThrown;
   return last;
 }
+var EmbeddingHttpError = class extends Error {
+  constructor(message, status) {
+    super(message);
+    this.name = "EmbeddingHttpError";
+    this.status = status;
+  }
+};
+function isFatalEmbeddingError(err) {
+  return err instanceof EmbeddingHttpError && (err.status === 401 || err.status === 402 || err.status === 403);
+}
+function embeddingErrorFrom(response) {
+  const message = typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed";
+  return new EmbeddingHttpError(message, response.response?.status);
+}
 async function generateEmbedding(text, options) {
   const { baseUrl = BASE_URL, getToken, apiKey, model, cache } = options;
   if (cache) {
@@ -3201,9 +3215,7 @@ async function generateEmbedding(text, options) {
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data?.[0]?.embedding) {
     throw new Error("No embedding returned from API");
@@ -3232,9 +3244,7 @@ async function generateEmbeddingsBatch(texts, headers, baseUrl, model, onUsage,
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data) {
     throw new Error("No embeddings returned from API");
@@ -3370,6 +3380,7 @@ async function embedAllMessages(ctx, options, filter) {
         await updateMessageEmbeddingOp(ctx, message.uniqueId, embedding, embeddingModel);
         embeddedCount++;
       } catch (error) {
+        if (isFatalEmbeddingError(error)) throw error;
         getLogger().error(`Failed to embed message ${message.uniqueId}:`, error);
       }
     }
@@ -3457,6 +3468,7 @@ async function chunkAndEmbedAllMessages(ctx, options, filter) {
         }
       }
     } catch (error) {
+      if (isFatalEmbeddingError(error)) throw error;
       getLogger().error("Failed to batch-embed short messages:", error);
     }
   }
@@ -3473,6 +3485,7 @@ async function chunkAndEmbedAllMessages(ctx, options, filter) {
       await updateMessageChunksOp(ctx, msg.uniqueId, messageChunks, embeddingModel);
       embeddedCount++;
     } catch (error) {
+      if (isFatalEmbeddingError(error)) throw error;
       getLogger().error(`Failed to embed message ${msg.uniqueId}:`, error);
     }
   }
@@ -3969,6 +3982,7 @@ function rankVaultMemories(_query, queryEmbedding, items, options) {
     updatedAt: item.updatedAt,
     createdAt: item.createdAt,
     eventTimeStart: item.eventTimeStart,
+    sourceChunkIds: item.sourceChunkIds,
     eventTimeEnd: item.eventTimeEnd,
     eventTimeKind: item.eventTimeKind,
     similarity: cosineSimilarity(queryEmbedding, item.embedding)
@@ -4010,7 +4024,8 @@ function rankVaultMemories(_query, queryEmbedding, items, options) {
       content: r.content,
       similarity: r.similarity,
       createdAt: item?.createdAt ?? item?.updatedAt,
-      updatedAt: item?.updatedAt
+      updatedAt: item?.updatedAt,
+      sourceChunkIds: item?.sourceChunkIds
     };
   });
 }
@@ -4049,6 +4064,7 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
       createdAt: item.createdAt ?? item.updatedAt,
       updatedAt: item.updatedAt,
       eventTimeStart: item.eventTimeStart,
+      sourceChunkIds: item.sourceChunkIds,
       eventTimeEnd: item.eventTimeEnd,
       eventTimeKind: item.eventTimeKind
     });
@@ -4091,7 +4107,8 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
           content: it.content,
           similarity: (fused.get(id) ?? 0) * boostFor(id),
           createdAt: it.createdAt ?? it.updatedAt,
-          updatedAt: it.updatedAt
+          updatedAt: it.updatedAt,
+          sourceChunkIds: it.sourceChunkIds
         });
         seen.add(id);
       }
@@ -4129,13 +4146,7 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
       combined = headSlice.map((r) => {
         const v2 = v2ScoreById.get(r.uniqueId) ?? 0;
         const ce = ceScoreById.get(r.uniqueId) ?? 0;
-        return {
-          uniqueId: r.uniqueId,
-          content: r.content,
-          similarity: v2 * (1 + ceWeight * ce),
-          createdAt: r.createdAt,
-          updatedAt: r.updatedAt
-        };
+        return { ...r, similarity: v2 * (1 + ceWeight * ce) };
       });
       combined.sort((a, b) => b.similarity - a.similarity);
     } catch (err) {
@@ -4168,7 +4179,8 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
           content: it.content,
           similarity: fused.get(id) ?? 0,
           createdAt: it.createdAt ?? it.updatedAt,
-          updatedAt: it.updatedAt
+          updatedAt: it.updatedAt,
+          sourceChunkIds: it.sourceChunkIds
         });
         seen.add(id);
       }
@@ -4198,6 +4210,7 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
         createdAt: orig?.createdAt,
         updatedAt: orig?.updatedAt,
         eventTimeStart: orig?.eventTimeStart,
+        sourceChunkIds: orig?.sourceChunkIds,
         eventTimeEnd: orig?.eventTimeEnd,
         eventTimeKind: orig?.eventTimeKind
       };
@@ -4261,6 +4274,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
         createdAt: item.createdAt ?? item.updatedAt,
         updatedAt: item.updatedAt,
         eventTimeStart: item.eventTimeStart,
+        sourceChunkIds: item.sourceChunkIds,
         eventTimeEnd: item.eventTimeEnd,
         eventTimeKind: item.eventTimeKind
       };
@@ -4281,6 +4295,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
           createdAt: item.createdAt ?? item.updatedAt,
           updatedAt: item.updatedAt,
           eventTimeStart: item.eventTimeStart,
+          sourceChunkIds: item.sourceChunkIds,
           eventTimeEnd: item.eventTimeEnd,
           eventTimeKind: item.eventTimeKind
         });
@@ -4332,6 +4347,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
         createdAt: orig?.createdAt,
         updatedAt: orig?.updatedAt,
         eventTimeStart: orig?.eventTimeStart,
+        sourceChunkIds: orig?.sourceChunkIds,
         eventTimeEnd: orig?.eventTimeEnd,
         eventTimeKind: orig?.eventTimeKind
       };
@@ -4473,6 +4489,7 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
     updatedAt: m.updatedAt,
     proofCount: m.proofCount,
     eventTimeStart: m.eventTimeStart,
+    sourceChunkIds: m.sourceChunkIds,
     eventTimeEnd: m.eventTimeEnd,
     eventTimeKind: normalizeEventTimeKind(m.eventTimeKind)
   }));
@@ -4663,7 +4680,7 @@ function createMemoryVaultSearchTool(vaultCtx, embeddingOptions, cache, searchOp
             legacy.map((r) => ({ id: r.uniqueId, content: r.content, score: r.similarity }))
           );
         }
-        const { recall: recall2 } = await import("./recall-GJ6RNZMF.mjs");
+        const { recall: recall2 } = await import("./recall-4JTNZBAW.mjs");
         const result = await recall2(
           query,
           { vaultCtx, embeddingOptions, vaultCache: cache },
@@ -5131,7 +5148,31 @@ async function recall(query, ctx, options = {}) {
     m.scoreBreakdown.fused = r.similarity;
     byId.set(key, m);
   }
-  const memories = [...byId.values()].sort((a, b) => b.score - a.score).slice(0, limit);
+  const ordered = [...byId.values()].sort((a, b) => b.score - a.score);
+  const selectWith = (suppressed2) => {
+    const out = [];
+    for (const m of ordered) {
+      if (out.length >= limit) break;
+      if (m.kind === "chunk" && m.messageId && suppressed2.has(m.messageId)) continue;
+      out.push(m);
+    }
+    return out;
+  };
+  const provenanceOf = (selected) => {
+    const s = /* @__PURE__ */ new Set();
+    for (const m of selected) {
+      if (m.kind === "fact" && m.sourceChunkIds) for (const id of m.sourceChunkIds) s.add(id);
+    }
+    return s;
+  };
+  let suppressed = /* @__PURE__ */ new Set();
+  let memories = selectWith(suppressed);
+  for (let i = 0; i < ordered.length; i++) {
+    const next = provenanceOf(memories);
+    if (next.size === suppressed.size && [...next].every((id) => suppressed.has(id))) break;
+    suppressed = next;
+    memories = selectWith(suppressed);
+  }
   return {
     memories,
     usedBudget,
@@ -5152,6 +5193,7 @@ function toFactMemory(r) {
     ...r.eventTimeStart !== void 0 && { eventTimeStart: r.eventTimeStart },
     ...r.eventTimeEnd !== void 0 && { eventTimeEnd: r.eventTimeEnd },
     ...r.eventTimeKind !== void 0 && { eventTimeKind: r.eventTimeKind },
+    ...r.sourceChunkIds !== void 0 && r.sourceChunkIds !== null && { sourceChunkIds: r.sourceChunkIds },
     // r.similarity from searchVaultMemoriesWithSize is the fused score
     // (cosine + BM25 + RRF + recency + proof) when useFusion=true (the
     // default) and pure cosine when useFusion=false. The breakdown

package/dist/expo/index.cjs

@@ -2837,6 +2837,13 @@ async function withEmbeddingRetry(call) {
   if (threw) throw lastThrown;
   return last;
 }
+function isFatalEmbeddingError(err) {
+  return err instanceof EmbeddingHttpError && (err.status === 401 || err.status === 402 || err.status === 403);
+}
+function embeddingErrorFrom(response) {
+  const message = typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed";
+  return new EmbeddingHttpError(message, response.response?.status);
+}
 async function generateEmbedding(text11, options) {
   const { baseUrl = BASE_URL, getToken, apiKey, model, cache } = options;
   if (cache) {
@@ -2868,9 +2875,7 @@ async function generateEmbedding(text11, options) {
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data?.[0]?.embedding) {
     throw new Error("No embedding returned from API");
@@ -2897,9 +2902,7 @@ async function generateEmbeddingsBatch(texts, headers, baseUrl, model, onUsage,
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data) {
     throw new Error("No embeddings returned from API");
@@ -3035,6 +3038,7 @@ async function embedAllMessages(ctx, options, filter) {
         await updateMessageEmbeddingOp(ctx, message.uniqueId, embedding, embeddingModel);
         embeddedCount++;
       } catch (error) {
+        if (isFatalEmbeddingError(error)) throw error;
         getLogger().error(`Failed to embed message ${message.uniqueId}:`, error);
       }
     }
@@ -3122,6 +3126,7 @@ async function chunkAndEmbedAllMessages(ctx, options, filter) {
         }
       }
     } catch (error) {
+      if (isFatalEmbeddingError(error)) throw error;
       getLogger().error("Failed to batch-embed short messages:", error);
     }
   }
@@ -3138,12 +3143,13 @@ async function chunkAndEmbedAllMessages(ctx, options, filter) {
       await updateMessageChunksOp(ctx, msg.uniqueId, messageChunks, embeddingModel);
       embeddedCount++;
     } catch (error) {
+      if (isFatalEmbeddingError(error)) throw error;
       getLogger().error(`Failed to embed message ${msg.uniqueId}:`, error);
     }
   }
   return embeddedCount;
 }
-var DEFAULT_MIN_CONTENT_LENGTH, EMBED_MAX_ATTEMPTS, DEFAULT_EMBEDDING_BATCH_SIZE, DEFAULT_EMBEDDING_BATCH_CONCURRENCY;
+var DEFAULT_MIN_CONTENT_LENGTH, EMBED_MAX_ATTEMPTS, EmbeddingHttpError, DEFAULT_EMBEDDING_BATCH_SIZE, DEFAULT_EMBEDDING_BATCH_CONCURRENCY;
 var init_embeddings = __esm({
   "src/lib/memoryEngine/embeddings.ts"() {
     "use strict";
@@ -3155,6 +3161,13 @@ var init_embeddings = __esm({
     init_constants();
     DEFAULT_MIN_CONTENT_LENGTH = 10;
     EMBED_MAX_ATTEMPTS = 4;
+    EmbeddingHttpError = class extends Error {
+      constructor(message, status) {
+        super(message);
+        this.name = "EmbeddingHttpError";
+        this.status = status;
+      }
+    };
     DEFAULT_EMBEDDING_BATCH_SIZE = 100;
     DEFAULT_EMBEDDING_BATCH_CONCURRENCY = 3;
   }
@@ -5546,7 +5559,31 @@ async function recall(query, ctx, options = {}) {
     m.scoreBreakdown.fused = r.similarity;
     byId.set(key, m);
   }
-  const memories = [...byId.values()].sort((a, b) => b.score - a.score).slice(0, limit);
+  const ordered = [...byId.values()].sort((a, b) => b.score - a.score);
+  const selectWith = (suppressed2) => {
+    const out = [];
+    for (const m of ordered) {
+      if (out.length >= limit) break;
+      if (m.kind === "chunk" && m.messageId && suppressed2.has(m.messageId)) continue;
+      out.push(m);
+    }
+    return out;
+  };
+  const provenanceOf = (selected) => {
+    const s = /* @__PURE__ */ new Set();
+    for (const m of selected) {
+      if (m.kind === "fact" && m.sourceChunkIds) for (const id of m.sourceChunkIds) s.add(id);
+    }
+    return s;
+  };
+  let suppressed = /* @__PURE__ */ new Set();
+  let memories = selectWith(suppressed);
+  for (let i = 0; i < ordered.length; i++) {
+    const next = provenanceOf(memories);
+    if (next.size === suppressed.size && [...next].every((id) => suppressed.has(id))) break;
+    suppressed = next;
+    memories = selectWith(suppressed);
+  }
   return {
     memories,
     usedBudget,
@@ -5567,6 +5604,7 @@ function toFactMemory(r) {
     ...r.eventTimeStart !== void 0 && { eventTimeStart: r.eventTimeStart },
     ...r.eventTimeEnd !== void 0 && { eventTimeEnd: r.eventTimeEnd },
     ...r.eventTimeKind !== void 0 && { eventTimeKind: r.eventTimeKind },
+    ...r.sourceChunkIds !== void 0 && r.sourceChunkIds !== null && { sourceChunkIds: r.sourceChunkIds },
     // r.similarity from searchVaultMemoriesWithSize is the fused score
     // (cosine + BM25 + RRF + recency + proof) when useFusion=true (the
     // default) and pure cosine when useFusion=false. The breakdown
@@ -5681,6 +5719,7 @@ function rankVaultMemories(_query, queryEmbedding, items, options) {
     updatedAt: item.updatedAt,
     createdAt: item.createdAt,
     eventTimeStart: item.eventTimeStart,
+    sourceChunkIds: item.sourceChunkIds,
     eventTimeEnd: item.eventTimeEnd,
     eventTimeKind: item.eventTimeKind,
     similarity: cosineSimilarity(queryEmbedding, item.embedding)
@@ -5722,7 +5761,8 @@ function rankVaultMemories(_query, queryEmbedding, items, options) {
       content: r.content,
       similarity: r.similarity,
       createdAt: item?.createdAt ?? item?.updatedAt,
-      updatedAt: item?.updatedAt
+      updatedAt: item?.updatedAt,
+      sourceChunkIds: item?.sourceChunkIds
     };
   });
 }
@@ -5761,6 +5801,7 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
       createdAt: item.createdAt ?? item.updatedAt,
       updatedAt: item.updatedAt,
       eventTimeStart: item.eventTimeStart,
+      sourceChunkIds: item.sourceChunkIds,
       eventTimeEnd: item.eventTimeEnd,
       eventTimeKind: item.eventTimeKind
     });
@@ -5803,7 +5844,8 @@ function rankFusedVaultMemories(query, queryEmbedding, items, options) {
           content: it.content,
           similarity: (fused.get(id) ?? 0) * boostFor(id),
           createdAt: it.createdAt ?? it.updatedAt,
-          updatedAt: it.updatedAt
+          updatedAt: it.updatedAt,
+          sourceChunkIds: it.sourceChunkIds
         });
         seen.add(id);
       }
@@ -5841,13 +5883,7 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
       combined = headSlice.map((r) => {
         const v2 = v2ScoreById.get(r.uniqueId) ?? 0;
         const ce = ceScoreById.get(r.uniqueId) ?? 0;
-        return {
-          uniqueId: r.uniqueId,
-          content: r.content,
-          similarity: v2 * (1 + ceWeight * ce),
-          createdAt: r.createdAt,
-          updatedAt: r.updatedAt
-        };
+        return { ...r, similarity: v2 * (1 + ceWeight * ce) };
       });
       combined.sort((a, b) => b.similarity - a.similarity);
     } catch (err) {
@@ -5880,7 +5916,8 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
           content: it.content,
           similarity: fused.get(id) ?? 0,
           createdAt: it.createdAt ?? it.updatedAt,
-          updatedAt: it.updatedAt
+          updatedAt: it.updatedAt,
+          sourceChunkIds: it.sourceChunkIds
         });
         seen.add(id);
       }
@@ -5910,6 +5947,7 @@ async function rankFusedVaultMemoriesAsync(query, queryEmbedding, items, options
         createdAt: orig?.createdAt,
         updatedAt: orig?.updatedAt,
         eventTimeStart: orig?.eventTimeStart,
+        sourceChunkIds: orig?.sourceChunkIds,
         eventTimeEnd: orig?.eventTimeEnd,
         eventTimeKind: orig?.eventTimeKind
       };
@@ -5973,6 +6011,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
         createdAt: item.createdAt ?? item.updatedAt,
         updatedAt: item.updatedAt,
         eventTimeStart: item.eventTimeStart,
+        sourceChunkIds: item.sourceChunkIds,
         eventTimeEnd: item.eventTimeEnd,
         eventTimeKind: item.eventTimeKind
       };
@@ -5993,6 +6032,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
           createdAt: item.createdAt ?? item.updatedAt,
           updatedAt: item.updatedAt,
           eventTimeStart: item.eventTimeStart,
+          sourceChunkIds: item.sourceChunkIds,
           eventTimeEnd: item.eventTimeEnd,
           eventTimeKind: item.eventTimeKind
         });
@@ -6044,6 +6084,7 @@ async function rankComposite(originalQuery, originalQueryEmbedding, subQueries,
         createdAt: orig?.createdAt,
         updatedAt: orig?.updatedAt,
         eventTimeStart: orig?.eventTimeStart,
+        sourceChunkIds: orig?.sourceChunkIds,
         eventTimeEnd: orig?.eventTimeEnd,
         eventTimeKind: orig?.eventTimeKind
       };
@@ -6185,6 +6226,7 @@ async function searchVaultMemoriesWithSize(query, vaultCtx, embeddingOptions, ca
     updatedAt: m.updatedAt,
     proofCount: m.proofCount,
     eventTimeStart: m.eventTimeStart,
+    sourceChunkIds: m.sourceChunkIds,
     eventTimeEnd: m.eventTimeEnd,
     eventTimeKind: normalizeEventTimeKind(m.eventTimeKind)
   }));

package/dist/expo/index.d.mts

@@ -4848,6 +4848,9 @@ interface VaultSearchResult {
     eventTimeStart?: number | null;
     eventTimeEnd?: number | null;
     eventTimeKind?: "point" | "range" | "ongoing" | null;
+    /** Message ids this fact was extracted from (provenance). recall() uses
+     * these to suppress the originating chunk in the chunk lane. */
+    sourceChunkIds?: string[] | null;
 }
 /**
  * Search vault memories by semantic similarity. Returns structured results

package/dist/expo/index.d.ts

@@ -4848,6 +4848,9 @@ interface VaultSearchResult {
     eventTimeStart?: number | null;
     eventTimeEnd?: number | null;
     eventTimeKind?: "point" | "range" | "ongoing" | null;
+    /** Message ids this fact was extracted from (provenance). recall() uses
+     * these to suppress the originating chunk in the chunk lane. */
+    sourceChunkIds?: string[] | null;
 }
 /**
  * Search vault memories by semantic similarity. Returns structured results

package/dist/expo/index.mjs

@@ -79,7 +79,7 @@ import {
   updateVaultMemoryOp,
   upsertMessageOp,
   useEncryption
-} from "./chunk-3RXNQBS3.mjs";
+} from "./chunk-IKPPX3GY.mjs";
 import {
   PII_PATTERNS,
   PiiRedactor,

package/dist/expo/{recall-GJ6RNZMF.mjs → recall-4JTNZBAW.mjs}

@@ -1,6 +1,6 @@
 import {
   recall
-} from "./chunk-3RXNQBS3.mjs";
+} from "./chunk-IKPPX3GY.mjs";
 import "./chunk-DFVCS3QQ.mjs";
 import "./chunk-GYKJOWEO.mjs";
 export {

package/dist/index.cjs

@@ -2095,6 +2095,17 @@ async function withEmbeddingRetry(call) {
   if (threw) throw lastThrown;
   return last;
 }
+var EmbeddingHttpError = class extends Error {
+  constructor(message, status) {
+    super(message);
+    this.name = "EmbeddingHttpError";
+    this.status = status;
+  }
+};
+function embeddingErrorFrom(response) {
+  const message = typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed";
+  return new EmbeddingHttpError(message, response.response?.status);
+}
 async function generateEmbedding(text, options) {
   const { baseUrl = BASE_URL, getToken, apiKey, model, cache } = options;
   if (cache) {
@@ -2126,9 +2137,7 @@ async function generateEmbedding(text, options) {
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data?.[0]?.embedding) {
     throw new Error("No embedding returned from API");
@@ -2157,9 +2166,7 @@ async function generateEmbeddingsBatch(texts, headers, baseUrl, model, onUsage,
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data) {
     throw new Error("No embeddings returned from API");

package/dist/index.mjs

@@ -1904,6 +1904,17 @@ async function withEmbeddingRetry(call) {
   if (threw) throw lastThrown;
   return last;
 }
+var EmbeddingHttpError = class extends Error {
+  constructor(message, status) {
+    super(message);
+    this.name = "EmbeddingHttpError";
+    this.status = status;
+  }
+};
+function embeddingErrorFrom(response) {
+  const message = typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed";
+  return new EmbeddingHttpError(message, response.response?.status);
+}
 async function generateEmbedding(text, options) {
   const { baseUrl = BASE_URL, getToken, apiKey, model, cache } = options;
   if (cache) {
@@ -1935,9 +1946,7 @@ async function generateEmbedding(text, options) {
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data?.[0]?.embedding) {
     throw new Error("No embedding returned from API");
@@ -1966,9 +1975,7 @@ async function generateEmbeddingsBatch(texts, headers, baseUrl, model, onUsage,
     })
   );
   if (response.error) {
-    throw new Error(
-      typeof response.error === "object" && response.error && "error" in response.error ? response.error.error : "API embedding failed"
-    );
+    throw embeddingErrorFrom(response);
   }
   if (!response.data?.data) {
     throw new Error("No embeddings returned from API");