@langchain/langgraph-checkpoint 0.0.12 → 0.0.13

package/dist/store/memory.cjs

@@ -2,56 +2,150 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.MemoryStore = exports.InMemoryStore = void 0;
 const base_js_1 = require("./base.cjs");
+const utils_js_1 = require("./utils.cjs");
+/**
+ * In-memory key-value store with optional vector search.
+ *
+ * A lightweight store implementation using JavaScript Maps. Supports basic
+ * key-value operations and vector search when configured with embeddings.
+ *
+ * @example
+ * ```typescript
+ * // Basic key-value storage
+ * const store = new InMemoryStore();
+ * await store.put(["users", "123"], "prefs", { theme: "dark" });
+ * const item = await store.get(["users", "123"], "prefs");
+ *
+ * // Vector search with embeddings
+ * import { OpenAIEmbeddings } from "@langchain/openai";
+ * const store = new InMemoryStore({
+ *   index: {
+ *     dims: 1536,
+ *     embeddings: new OpenAIEmbeddings({ modelName: "text-embedding-3-small" }),
+ *   }
+ * });
+ *
+ * // Store documents
+ * await store.put(["docs"], "doc1", { text: "Python tutorial" });
+ * await store.put(["docs"], "doc2", { text: "TypeScript guide" });
+ *
+ * // Search by similarity
+ * const results = await store.search(["docs"], { query: "python programming" });
+ * ```
+ *
+ * @warning This store keeps all data in memory. Data is lost when the process exits.
+ * For persistence, use a database-backed store.
+ */
 class InMemoryStore extends base_js_1.BaseStore {
-    constructor() {
-        super(...arguments);
+    constructor(options) {
+        super();
         Object.defineProperty(this, "data", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: new Map()
         });
+        // Namespace -> Key -> Path/field -> Vector
+        Object.defineProperty(this, "vectors", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: new Map()
+        });
+        Object.defineProperty(this, "_indexConfig", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: void 0
+        });
+        if (options?.index) {
+            this._indexConfig = {
+                ...options.index,
+                __tokenizedFields: (options.index.fields ?? ["$"]).map((p) => [
+                    p,
+                    p === "$" ? [p] : (0, utils_js_1.tokenizePath)(p),
+                ]),
+            };
+        }
     }
     async batch(operations) {
         const results = [];
-        for (const op of operations) {
+        const putOps = new Map();
+        const searchOps = new Map();
+        // First pass - handle gets and prepare search/put operations
+        for (let i = 0; i < operations.length; i += 1) {
+            const op = operations[i];
             if ("key" in op && "namespace" in op && !("value" in op)) {
                 // GetOperation
                 results.push(this.getOperation(op));
             }
             else if ("namespacePrefix" in op) {
                 // SearchOperation
-                results.push(this.searchOperation(op));
+                const candidates = this.filterItems(op);
+                searchOps.set(i, [op, candidates]);
+                results.push(null);
             }
             else if ("value" in op) {
                 // PutOperation
-                results.push(this.putOperation(op));
+                const key = `${op.namespace.join(":")}:${op.key}`;
+                putOps.set(key, op);
+                results.push(null);
             }
             else if ("matchConditions" in op) {
                 // ListNamespacesOperation
                 results.push(this.listNamespacesOperation(op));
             }
         }
-        return Promise.resolve(results);
+        // Handle search operations with embeddings
+        if (searchOps.size > 0) {
+            if (this._indexConfig?.embeddings) {
+                const queries = new Set();
+                for (const [op] of searchOps.values()) {
+                    if (op.query)
+                        queries.add(op.query);
+                }
+                // Get embeddings for all queries
+                const queryEmbeddings = queries.size > 0
+                    ? await Promise.all(Array.from(queries).map((q) => this._indexConfig.embeddings.embedQuery(q)))
+                    : [];
+                const queryVectors = Object.fromEntries(Array.from(queries).map((q, i) => [q, queryEmbeddings[i]]));
+                // Process each search operation
+                for (const [i, [op, candidates]] of searchOps.entries()) {
+                    if (op.query && queryVectors[op.query]) {
+                        const queryVector = queryVectors[op.query];
+                        const scoredResults = this.scoreResults(candidates, queryVector, op.offset ?? 0, op.limit ?? 10);
+                        results[i] = scoredResults;
+                    }
+                    else {
+                        results[i] = this.paginateResults(candidates.map((item) => ({ ...item, score: undefined })), op.offset ?? 0, op.limit ?? 10);
+                    }
+                }
+            }
+            else {
+                // No embeddings - just paginate the filtered results
+                for (const [i, [op, candidates]] of searchOps.entries()) {
+                    results[i] = this.paginateResults(candidates.map((item) => ({ ...item, score: undefined })), op.offset ?? 0, op.limit ?? 10);
+                }
+            }
+        }
+        // Handle put operations with embeddings
+        if (putOps.size > 0 && this._indexConfig?.embeddings) {
+            const toEmbed = this.extractTexts(Array.from(putOps.values()));
+            if (Object.keys(toEmbed).length > 0) {
+                const embeddings = await this._indexConfig.embeddings.embedDocuments(Object.keys(toEmbed));
+                this.insertVectors(toEmbed, embeddings);
+            }
+        }
+        // Apply all put operations
+        for (const op of putOps.values()) {
+            this.putOperation(op);
+        }
+        return results;
     }
     getOperation(op) {
         const namespaceKey = op.namespace.join(":");
         const item = this.data.get(namespaceKey)?.get(op.key);
-        return item || null;
-    }
-    searchOperation(op) {
-        const candidates = [];
-        for (const [namespace, items] of this.data.entries()) {
-            if (namespace.startsWith(op.namespacePrefix.join(":"))) {
-                candidates.push(...items.values());
-            }
-        }
-        let filteredCandidates = candidates;
-        if (op.filter) {
-            filteredCandidates = candidates.filter((item) => Object.entries(op.filter).every(([key, value]) => item.value[key] === value));
-        }
-        const searchResults = filteredCandidates.slice(op.offset || 0, (op.offset || 0) + (op.limit || 10));
-        return searchResults;
+        return item ?? null;
     }
     putOperation(op) {
         const namespaceKey = op.namespace.join(":");
@@ -90,28 +184,179 @@ class InMemoryStore extends base_js_1.BaseStore {
             namespaces = Array.from(new Set(namespaces.map((ns) => ns.slice(0, op.maxDepth).join(":")))).map((ns) => ns.split(":"));
         }
         namespaces.sort((a, b) => a.join(":").localeCompare(b.join(":")));
-        const paginatedNamespaces = namespaces.slice(op.offset, op.offset + op.limit);
-        return paginatedNamespaces;
+        return namespaces.slice(op.offset ?? 0, (op.offset ?? 0) + (op.limit ?? namespaces.length));
     }
     doesMatch(matchCondition, key) {
         const { matchType, path } = matchCondition;
-        if (key.length < path.length) {
-            return false;
-        }
         if (matchType === "prefix") {
+            if (path.length > key.length)
+                return false;
             return path.every((pElem, index) => {
                 const kElem = key[index];
                 return pElem === "*" || kElem === pElem;
             });
         }
         else if (matchType === "suffix") {
-            return [...path].reverse().every((pElem, index) => {
-                const kElem = key[key.length - 1 - index];
+            if (path.length > key.length)
+                return false;
+            return path.every((pElem, index) => {
+                const kElem = key[key.length - path.length + index];
                 return pElem === "*" || kElem === pElem;
             });
         }
         throw new Error(`Unsupported match type: ${matchType}`);
     }
+    filterItems(op) {
+        const candidates = [];
+        for (const [namespace, items] of this.data.entries()) {
+            if (namespace.startsWith(op.namespacePrefix.join(":"))) {
+                candidates.push(...items.values());
+            }
+        }
+        let filteredCandidates = candidates;
+        if (op.filter) {
+            filteredCandidates = candidates.filter((item) => Object.entries(op.filter).every(([key, value]) => (0, utils_js_1.compareValues)(item.value[key], value)));
+        }
+        return filteredCandidates;
+    }
+    scoreResults(candidates, queryVector, offset = 0, limit = 10) {
+        const flatItems = [];
+        const flatVectors = [];
+        const scoreless = [];
+        for (const item of candidates) {
+            const vectors = this.getVectors(item);
+            if (vectors.length) {
+                for (const vector of vectors) {
+                    flatItems.push(item);
+                    flatVectors.push(vector);
+                }
+            }
+            else {
+                scoreless.push(item);
+            }
+        }
+        const scores = this.cosineSimilarity(queryVector, flatVectors);
+        const sortedResults = scores
+            .map((score, i) => [score, flatItems[i]])
+            .sort((a, b) => b[0] - a[0]);
+        const seen = new Set();
+        const kept = [];
+        for (const [score, item] of sortedResults) {
+            const key = `${item.namespace.join(":")}:${item.key}`;
+            if (seen.has(key))
+                continue;
+            const ix = seen.size;
+            if (ix >= offset + limit)
+                break;
+            if (ix < offset) {
+                seen.add(key);
+                continue;
+            }
+            seen.add(key);
+            kept.push([score, item]);
+        }
+        if (scoreless.length && kept.length < limit) {
+            for (const item of scoreless.slice(0, limit - kept.length)) {
+                const key = `${item.namespace.join(":")}:${item.key}`;
+                if (!seen.has(key)) {
+                    seen.add(key);
+                    kept.push([undefined, item]);
+                }
+            }
+        }
+        return kept.map(([score, item]) => ({
+            ...item,
+            score,
+        }));
+    }
+    paginateResults(results, offset, limit) {
+        return results.slice(offset, offset + limit);
+    }
+    extractTexts(ops) {
+        if (!ops.length || !this._indexConfig) {
+            return {};
+        }
+        const toEmbed = {};
+        for (const op of ops) {
+            if (op.value !== null && op.index !== false) {
+                const paths = op.index === null || op.index === undefined
+                    ? this._indexConfig.__tokenizedFields ?? []
+                    : op.index.map((ix) => [ix, (0, utils_js_1.tokenizePath)(ix)]);
+                for (const [path, field] of paths) {
+                    const texts = (0, utils_js_1.getTextAtPath)(op.value, field);
+                    if (texts.length) {
+                        if (texts.length > 1) {
+                            texts.forEach((text, i) => {
+                                if (!toEmbed[text])
+                                    toEmbed[text] = [];
+                                toEmbed[text].push([op.namespace, op.key, `${path}.${i}`]);
+                            });
+                        }
+                        else {
+                            if (!toEmbed[texts[0]])
+                                toEmbed[texts[0]] = [];
+                            toEmbed[texts[0]].push([op.namespace, op.key, path]);
+                        }
+                    }
+                }
+            }
+        }
+        return toEmbed;
+    }
+    insertVectors(texts, embeddings) {
+        for (const [text, metadata] of Object.entries(texts)) {
+            const embedding = embeddings.shift();
+            if (!embedding) {
+                throw new Error(`No embedding found for text: ${text}`);
+            }
+            for (const [namespace, key, field] of metadata) {
+                const namespaceKey = namespace.join(":");
+                if (!this.vectors.has(namespaceKey)) {
+                    this.vectors.set(namespaceKey, new Map());
+                }
+                const namespaceMap = this.vectors.get(namespaceKey);
+                if (!namespaceMap.has(key)) {
+                    namespaceMap.set(key, new Map());
+                }
+                const itemMap = namespaceMap.get(key);
+                itemMap.set(field, embedding);
+            }
+        }
+    }
+    getVectors(item) {
+        const namespaceKey = item.namespace.join(":");
+        const itemKey = item.key;
+        if (!this.vectors.has(namespaceKey)) {
+            return [];
+        }
+        const namespaceMap = this.vectors.get(namespaceKey);
+        if (!namespaceMap.has(itemKey)) {
+            return [];
+        }
+        const itemMap = namespaceMap.get(itemKey);
+        const vectors = Array.from(itemMap.values());
+        if (!vectors.length) {
+            return [];
+        }
+        return vectors;
+    }
+    cosineSimilarity(X, Y) {
+        if (!Y.length)
+            return [];
+        // Calculate dot products for all vectors at once
+        const dotProducts = Y.map((vector) => vector.reduce((acc, val, i) => acc + val * X[i], 0));
+        // Calculate magnitudes
+        const magnitude1 = Math.sqrt(X.reduce((acc, val) => acc + val * val, 0));
+        const magnitudes2 = Y.map((vector) => Math.sqrt(vector.reduce((acc, val) => acc + val * val, 0)));
+        // Calculate similarities
+        return dotProducts.map((dot, i) => {
+            const magnitude2 = magnitudes2[i];
+            return magnitude1 && magnitude2 ? dot / (magnitude1 * magnitude2) : 0;
+        });
+    }
+    get indexConfig() {
+        return this._indexConfig;
+    }
 }
 exports.InMemoryStore = InMemoryStore;
 /** @deprecated Alias for InMemoryStore */

package/dist/store/memory.d.ts

@@ -1,12 +1,57 @@
-import { BaseStore, type OperationResults, type Operation } from "./base.js";
+import { BaseStore, type OperationResults, type Operation, type IndexConfig } from "./base.js";
+/**
+ * In-memory key-value store with optional vector search.
+ *
+ * A lightweight store implementation using JavaScript Maps. Supports basic
+ * key-value operations and vector search when configured with embeddings.
+ *
+ * @example
+ * ```typescript
+ * // Basic key-value storage
+ * const store = new InMemoryStore();
+ * await store.put(["users", "123"], "prefs", { theme: "dark" });
+ * const item = await store.get(["users", "123"], "prefs");
+ *
+ * // Vector search with embeddings
+ * import { OpenAIEmbeddings } from "@langchain/openai";
+ * const store = new InMemoryStore({
+ *   index: {
+ *     dims: 1536,
+ *     embeddings: new OpenAIEmbeddings({ modelName: "text-embedding-3-small" }),
+ *   }
+ * });
+ *
+ * // Store documents
+ * await store.put(["docs"], "doc1", { text: "Python tutorial" });
+ * await store.put(["docs"], "doc2", { text: "TypeScript guide" });
+ *
+ * // Search by similarity
+ * const results = await store.search(["docs"], { query: "python programming" });
+ * ```
+ *
+ * @warning This store keeps all data in memory. Data is lost when the process exits.
+ * For persistence, use a database-backed store.
+ */
 export declare class InMemoryStore extends BaseStore {
     private data;
-    batch<Op extends Operation[]>(operations: Op): Promise<OperationResults<Op>>;
+    private vectors;
+    private _indexConfig?;
+    constructor(options?: {
+        index?: IndexConfig;
+    });
+    batch<Op extends readonly Operation[]>(operations: Op): Promise<OperationResults<Op>>;
     private getOperation;
-    private searchOperation;
     private putOperation;
     private listNamespacesOperation;
     private doesMatch;
+    private filterItems;
+    private scoreResults;
+    private paginateResults;
+    private extractTexts;
+    private insertVectors;
+    private getVectors;
+    private cosineSimilarity;
+    get indexConfig(): IndexConfig | undefined;
 }
 /** @deprecated Alias for InMemoryStore */
 export declare class MemoryStore extends InMemoryStore {