CaviraOSS/LongMemory

Local persistent memory store for LLM applications including claude desktop, github copilot, codex, antigravity, etc.

LongMemory – Durable, temporal, governed memory for AI agents

What it is

  • A local‑first, SQLite‑backed memory engine designed for large‑language‑model (LLM) applications and autonomous agents.
  • Stores immutable “nodes” and executable edges in a Hydrograph graph, preserving provenance, timestamps, and multilingual content.
  • Provides several recall modes (strict, historical, associative, world‑grounded) that enforce temporal truth, contradiction handling, and access control.

Key capabilities

  • Temporal truth – separate record and valid times, deterministic decay, reinforcement, and consolidation.
  • Governance – project, tenant, user, role, and agent scopes with ACLs.
  • Explainable recall – evidence‑bounded results with optional single‑model call via answer_from_evidence.
  • Multilingual support – script detection, transliteration, cross‑language embeddings.
  • Integrations – native connectors for repos, files, web feeds, cloud docs; embedding back‑ends (OpenAI‑compatible, Gemini, Bedrock, Ollama, etc.); agent frameworks (CrewAI, AutoGen, LangGraph/LangChain, OpenAI Agents SDK), IDE plugins (VS Code), automation tools (n8n, Dify, Flowise), and MCP toolset.
  • Operational surfaces – TypeScript library, CLI, HTTP API, MCP (streamable), web dashboard, VS Code extension.

How to get started

  1. Library – npm install longmemory (or pip install longmemory-sdk for the HTTP client). Create a memory with createMemory(); ingest text and recall with memory.ingest() / memory.recall().
  2. CLI – npm install -g longmemory; run longmemory init, longmemory ingest …, longmemory recall ….
  3. Self‑hosted service – run from source (pnpm install && pnpm build && pnpm start) or via Docker (docker run -p 7331:7331 -v longmemory-data:/data ghcr.io/caviraoss/longmemory). A dashboard is available on port 3000 via the ui compose profile.

Architecture at a glance

  • Ingest pipeline → immutable nodes stored in SQLite → Hydrograph (graph of nodes, edges, worlds, entities).
  • Graph feeds three recall engines (strict, associative, world‑grounded) which produce a bounded context for the LLM.
  • The graph, API, MCP tools, and UI are all built on a single TypeScript codebase.

Typical use cases

  • Persistent personal or team memory for chat assistants (Chat Memory, Skills, LLM‑Wiki).
  • Project‑level knowledge bases that retain decisions, architecture notes, and code impact facts.
  • Agent tooling where actions need evidence‑backed recall (e.g., autonomous DevOps bots).

Deployment options

  • Docker / Docker‑Compose (API + optional UI).
  • Cloud platforms (Heroku, Railway, Render, DigitalOcean App Platform, Vercel – UI only).
  • Native Windows script for local background service.

Benchmarks & quality

  • Comes with a benchmark harness (pnpm bench*) that reports retrieval quality, temporal behavior, latency, and evidence metrics.

License

  • Apache License 2.0 (core); the bundled n8n node uses MIT as required.

Where to learn more

  • Docs: docs/answering.md, docs/python-sdk.md, docs/mcp.md, ARCHITECTURE.md.
  • Dashboard demo GIF in the repo root.
  • Contribution guidelines and governance docs are provided for community involvement.

Related

  • Project
    buildingjoshbetter/TrueMemory

    A local-first long-term memory system for AI agents that automatically captures and updates user preferences and facts across sessions to eliminate AI amnesia.

  • Project
    lxfight-s-Astrbot-Plugins/astrbot_plugin_livingmemory

    A plugin for the AstrBot framework that provides AI agents with long-term memory by storing and retrieving user preferences, relationships, and historical context using vector search and knowledge graphs.

  • Project
    caspianmoon/memoripy

    An evidence-first, local memory runtime for AI agents that prevents noise ingestion and provides verifiable, bitemporal records with retrieval receipts.

  • Project
    sopaco/cortex-mem

    A Rust-based long-term memory framework for AI agents that uses a three-tier hierarchical architecture and a virtual filesystem to provide persistent, context-aware interactions.

  • Project
    matrixorigin/memoria

    A persistent memory layer for AI agents featuring Git-like version control, allowing for snapshots, branching, and hybrid semantic search to maintain auditable long-term knowledge.