Skip to content

LocalLore documentation

LocalLore gives Claude Code a private memory of earlier coding sessions. A persistent local service incrementally indexes Claude's JSONL session history and exposes keyword, semantic, and contextual retrieval through the Model Context Protocol (MCP).

Everything needed for retrieval stays on your machine: the source sessions, SQLite index, embedding model, and MCP endpoint. LocalLore does not send conversation text to a hosted embedding or search service.

  • Start with a working search

    Install LocalLore, verify the daemon, and retrieve your first memory.

    First-search tutorial

  • Complete a task

    Find operational recipes for installation, search, and troubleshooting.

    How-to guides

  • Understand the design

    Learn MCP, RAG, embeddings, indexing, hybrid search, and local storage.

    Architecture

  • Look up exact details

    Check settings, interfaces, and generated Python API documentation.

    Reference

How a memory becomes a result

flowchart LR
    A[Claude Code JSONL] --> B[Incremental indexer]
    B --> C[(SQLite + FTS5)]
    B --> D[Local embedding model]
    D --> C
    E[Claude Code] -->|MCP query| F[LocalLore daemon]
    F -->|keyword + vector retrieval| C
    C -->|ranked excerpts| F
    F -->|tool result| E

The documentation follows the Diátaxis model: tutorials teach through a guided experience, how-to guides solve specific tasks, explanations build understanding, and reference pages provide precise technical facts.

Project status

This repository is a finished, pinned snapshot and is not actively maintained. Review its locked dependencies and security assumptions before adopting it.