modelfusion 0.36.0 → 0.37.0

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "Build multimodal applications, chatbots, and agents with JavaScript and TypeScript.",
-  "version": "0.36.0",
+  "version": "0.37.0",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [
@@ -73,7 +73,7 @@
     "husky": "^8.0.3",
     "lint-staged": "14.0.1",
     "prettier": "3.0.3",
-    "rimraf": "5.0.1",
+    "rimraf": "5.0.5",
     "typescript": "5.2.2",
     "vitest": "^0.34.5"
   },

package/vector-index/VectorIndex.d.ts

@@ -1,5 +1,5 @@
 import { Vector } from "../core/Vector.js";
-export interface VectorIndex<DATA, INDEX> {
+export interface VectorIndex<DATA, INDEX, FILTER> {
     upsertMany(data: Array<{
         id: string;
         vector: Vector;
@@ -9,6 +9,7 @@ export interface VectorIndex<DATA, INDEX> {
         queryVector: Vector;
         maxResults: number;
         similarityThreshold?: number;
+        filter?: FILTER;
     }): Promise<Array<{
         id: string;
         data: DATA;

package/vector-index/VectorIndexRetriever.cjs

@@ -3,7 +3,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.VectorIndexRetriever = void 0;
 const embedText_js_1 = require("../model-function/embed-text/embedText.cjs");
 class VectorIndexRetriever {
-    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, }) {
+    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, filter, }) {
         Object.defineProperty(this, "vectorIndex", {
             enumerable: true,
             configurable: true,
@@ -27,6 +27,7 @@ class VectorIndexRetriever {
         this.settings = {
             maxResults,
             similarityThreshold,
+            filter,
         };
     }
     async retrieve(query, options) {
@@ -37,6 +38,11 @@ class VectorIndexRetriever {
                 run: options.run,
             });
         }
+        let filter = this.settings?.filter;
+        if (typeof query === "object") {
+            filter = query.filter ?? filter; // use filter from query if available
+            query = query.text;
+        }
         const embedding = await (0, embedText_js_1.embedText)(this.embeddingModel, query, {
             functionId: options?.functionId,
             run: options?.run,
@@ -45,6 +51,7 @@ class VectorIndexRetriever {
             queryVector: embedding,
             maxResults: this.settings.maxResults ?? 1,
             similarityThreshold: this.settings.similarityThreshold,
+            filter,
         });
         return queryResult.map((item) => item.data);
     }

package/vector-index/VectorIndexRetriever.d.ts

@@ -2,18 +2,24 @@ import { ModelFunctionOptions } from "../model-function/ModelFunctionOptions.js"
 import { TextEmbeddingModel, TextEmbeddingModelSettings } from "../model-function/embed-text/TextEmbeddingModel.js";
 import { Retriever, RetrieverSettings } from "../retriever/Retriever.js";
 import { VectorIndex } from "./VectorIndex.js";
-export interface VectorIndexRetrieverSettings {
+export interface VectorIndexRetrieverSettings<FILTER> {
     maxResults?: number;
     similarityThreshold?: number;
+    filter?: FILTER;
 }
-export declare class VectorIndexRetriever<OBJECT, INDEX> implements Retriever<OBJECT, string, VectorIndexRetrieverSettings> {
+type VectorIndexRetrieverQuery<FILTER> = string | {
+    text: string;
+    filter?: FILTER;
+};
+export declare class VectorIndexRetriever<OBJECT, INDEX, FILTER> implements Retriever<OBJECT, VectorIndexRetrieverQuery<FILTER>, VectorIndexRetrieverSettings<FILTER>> {
     private readonly vectorIndex;
     private readonly embeddingModel;
     private readonly settings;
-    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, }: {
-        vectorIndex: VectorIndex<OBJECT, INDEX>;
+    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, filter, }: {
+        vectorIndex: VectorIndex<OBJECT, INDEX, FILTER>;
         embeddingModel: TextEmbeddingModel<unknown, TextEmbeddingModelSettings>;
-    } & VectorIndexRetrieverSettings);
-    retrieve(query: string, options?: ModelFunctionOptions<RetrieverSettings>): Promise<OBJECT[]>;
-    withSettings(additionalSettings: Partial<VectorIndexRetrieverSettings>): this;
+    } & VectorIndexRetrieverSettings<FILTER>);
+    retrieve(query: VectorIndexRetrieverQuery<FILTER>, options?: ModelFunctionOptions<RetrieverSettings>): Promise<OBJECT[]>;
+    withSettings(additionalSettings: Partial<VectorIndexRetrieverSettings<FILTER>>): this;
 }
+export {};

package/vector-index/VectorIndexRetriever.js

@@ -1,6 +1,6 @@
 import { embedText } from "../model-function/embed-text/embedText.js";
 export class VectorIndexRetriever {
-    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, }) {
+    constructor({ vectorIndex, embeddingModel, maxResults, similarityThreshold, filter, }) {
         Object.defineProperty(this, "vectorIndex", {
             enumerable: true,
             configurable: true,
@@ -24,6 +24,7 @@ export class VectorIndexRetriever {
         this.settings = {
             maxResults,
             similarityThreshold,
+            filter,
         };
     }
     async retrieve(query, options) {
@@ -34,6 +35,11 @@ export class VectorIndexRetriever {
                 run: options.run,
             });
         }
+        let filter = this.settings?.filter;
+        if (typeof query === "object") {
+            filter = query.filter ?? filter; // use filter from query if available
+            query = query.text;
+        }
         const embedding = await embedText(this.embeddingModel, query, {
             functionId: options?.functionId,
             run: options?.run,
@@ -42,6 +48,7 @@ export class VectorIndexRetriever {
             queryVector: embedding,
             maxResults: this.settings.maxResults ?? 1,
             similarityThreshold: this.settings.similarityThreshold,
+            filter,
         });
         return queryResult.map((item) => item.data);
     }

package/vector-index/memory/MemoryVectorIndex.cjs

@@ -40,8 +40,9 @@ class MemoryVectorIndex {
             this.entries.set(entry.id, entry);
         }
     }
-    async queryByVector({ queryVector, similarityThreshold, maxResults, }) {
+    async queryByVector({ queryVector, similarityThreshold, maxResults, filter, }) {
         const results = [...this.entries.values()]
+            .filter((value) => filter?.(value.data) ?? true)
             .map((entry) => ({
             id: entry.id,
             similarity: (0, cosineSimilarity_js_1.cosineSimilarity)(entry.vector, queryVector),

package/vector-index/memory/MemoryVectorIndex.d.ts

@@ -6,7 +6,7 @@ import { VectorIndex } from "../VectorIndex.js";
  * a small number of entries and don't want to set up a real database, e.g. for conversational memory
  * that does not need to be persisted.
  */
-export declare class MemoryVectorIndex<DATA> implements VectorIndex<DATA, MemoryVectorIndex<DATA>> {
+export declare class MemoryVectorIndex<DATA> implements VectorIndex<DATA, MemoryVectorIndex<DATA>, (value: DATA) => boolean> {
     static deserialize<DATA>({ serializedData, schema, }: {
         serializedData: string;
         schema?: z.ZodSchema<DATA>;
@@ -17,10 +17,11 @@ export declare class MemoryVectorIndex<DATA> implements VectorIndex<DATA, Memory
         vector: Vector;
         data: DATA;
     }>): Promise<void>;
-    queryByVector({ queryVector, similarityThreshold, maxResults, }: {
+    queryByVector({ queryVector, similarityThreshold, maxResults, filter, }: {
         queryVector: Vector;
         maxResults: number;
         similarityThreshold?: number;
+        filter?: (value: DATA) => boolean;
     }): Promise<Array<{
         id: string;
         data: DATA;

package/vector-index/memory/MemoryVectorIndex.js

@@ -34,8 +34,9 @@ export class MemoryVectorIndex {
             this.entries.set(entry.id, entry);
         }
     }
-    async queryByVector({ queryVector, similarityThreshold, maxResults, }) {
+    async queryByVector({ queryVector, similarityThreshold, maxResults, filter, }) {
         const results = [...this.entries.values()]
+            .filter((value) => filter?.(value.data) ?? true)
             .map((entry) => ({
             id: entry.id,
             similarity: cosineSimilarity(entry.vector, queryVector),

package/vector-index/pinecone/PineconeVectorIndex.d.ts

@@ -2,7 +2,7 @@ import { VectorOperationsApi } from "@pinecone-database/pinecone/dist/pinecone-g
 import { z } from "zod";
 import { Vector } from "../../core/Vector.js";
 import { VectorIndex } from "../VectorIndex.js";
-export declare class PineconeVectorIndex<DATA extends object | undefined> implements VectorIndex<DATA, PineconeVectorIndex<DATA>> {
+export declare class PineconeVectorIndex<DATA extends object | undefined> implements VectorIndex<DATA, PineconeVectorIndex<DATA>, null> {
     readonly index: VectorOperationsApi;
     readonly namespace?: string;
     readonly schema: z.ZodSchema<DATA>;

package/vector-index/upsertIntoVectorIndex.d.ts

@@ -2,7 +2,7 @@ import { ModelFunctionOptions } from "../model-function/ModelFunctionOptions.js"
 import { TextEmbeddingModel, TextEmbeddingModelSettings } from "../model-function/embed-text/TextEmbeddingModel.js";
 import { VectorIndex } from "./VectorIndex.js";
 export declare function upsertIntoVectorIndex<OBJECT, SETTINGS extends TextEmbeddingModelSettings>({ vectorIndex, embeddingModel, generateId, objects, getValueToEmbed, getId, }: {
-    vectorIndex: VectorIndex<OBJECT, unknown>;
+    vectorIndex: VectorIndex<OBJECT, unknown, unknown>;
     embeddingModel: TextEmbeddingModel<unknown, SETTINGS>;
     generateId?: () => string;
     objects: OBJECT[];