@neo4j-labs/agent-memory (source; unreleased changes since 0.5.0)
    Preparing search index...

    @neo4j-labs/agent-memory (source; unreleased changes since 0.5.0)

    @neo4j-labs/agent-memory

    Neo4j Labs Status: Beta Community Supported

    Package installation. Install with npm install @neo4j-labs/[email protected] (package page). Source lives alongside the Python SDK at neo4j-labs/agent-memory (relocated from neo4j-labs/agent-memory-tck).

    TypeScript client for the Neo4j Agent Memory Service — short-term, long-term, and reasoning memory for AI agents, backed by Neo4j.

    ⚠️ Neo4j Labs Project

    This project is part of Neo4j Labs and is actively maintained, but not officially supported. There are no SLAs, backward-compatibility guarantees, or scheduled deprecation commitments. APIs may change without notice. For questions and support, please use the Neo4j Community Forum.

    The Python and TypeScript packages in this repository are versioned and released independently; the status badge above reflects this package's own maturity, not the Python SDK's.

    • Three memory subclients in one client: short-term (conversations, messages, three-tier context), long-term (entities, search, returned graph views), and reasoning (steps, traces, provenance, tool calls).
    • Zero-config construction — reads MEMORY_API_KEY from the environment and defaults to the hosted service.
    • Works in Node 22+, Bun, Deno, Cloudflare Workers, and Vercel Edge.
    • Five framework integrations: Vercel AI SDK middleware, MCP tools, LangChain JS, Mastra, and AWS Strands Agents.
    • Built-in request logging, request-id correlation, and edge-friendly fetch-only transports.
    • TCK Bronze conformance verified by the agent-memory-tck cross-language test suite.
    npm install @neo4j-labs/[email protected]
    

    Requires Node.js 22+. This README describes the current source; changes made since the 0.5.0 release are listed under Unreleased in the CHANGELOG. The source examples build the SDK before installing their local file: dependency.

    Get an API key from memory.neo4jlabs.com, export it as MEMORY_API_KEY, then:

    import { MemoryClient } from "@neo4j-labs/agent-memory";

    const client = new MemoryClient();

    const conv = await client.shortTerm.createConversation({ userId: "alice" });
    await client.shortTerm.addMessage(conv.id, "user", "Hello!");

    const entity = await client.longTerm.addEntity("Alice Johnson", "person", {
    description: "Software engineer working on graph memory.",
    });

    const ctx = await client.shortTerm.getContext(conv.id);
    console.log(ctx.recentMessages, ctx.observations, ctx.reflections);

    Edge runtimes differ in how they expose environment variables: Cloudflare Workers pass bindings to the handler, while others (such as Vercel Edge) may provide process.env. Pass the key explicitly:

    export default {
    async fetch(req: Request, env: { MEMORY_API_KEY: string }) {
    const client = new MemoryClient({ apiKey: env.MEMORY_API_KEY });
    // ...
    },
    };

    All five ship as subpath exports. See each integration's example and how-to guide for a runnable walkthrough.

    Integration Import Example
    Vercel AI SDK @neo4j-labs/agent-memory/middleware/vercel-ai examples/vercel-ai
    MCP tools @neo4j-labs/agent-memory/mcp, …/mcp/register examples/mcp
    LangChain JS @neo4j-labs/agent-memory/integrations/langchain examples/langchain
    Mastra @neo4j-labs/agent-memory/integrations/mastra examples/mastra
    AWS Strands @neo4j-labs/agent-memory/integrations/strands examples/strands

    Full API reference (TypeDoc) is bundled with the docs site — see the TypeScript API reference.

    The client accepts a small options bag:

    new MemoryClient({
    endpoint: "https://memory.neo4jlabs.com/v1", // default
    apiKey: "nams_...", // falls back to MEMORY_API_KEY env
    timeout: 30_000, // ms; default 30s
    headers: { "X-My-Trace": "..." }, // additional request headers
    logger: (event) => console.log(event), // request/response/error events
    });

    connect() is optional — the first request acts as the implicit auth check. Call it explicitly if you prefer fail-fast at startup.

    We welcome contributions. See the repo-root CONTRIBUTING.md for the development setup, test commands, and PR conventions — there is a dedicated "TypeScript contributions" section covering Node setup, vitest, linting, and the TCK bridge.

    This package lives alongside the Python SDK (neo4j-agent-memory) in neo4j-labs/agent-memory. The two SDKs implement the same memory model and share the same backend (NAMS). The cross-language behavioral contract is defined and certified by the agent-memory-tck spec repo, which consumes this package from npm.

    Apache-2.0 — see LICENSE.