Repolode
Submit
oleksiijko

oleksiijko/pmb

Local-first persistent memory for AI coding agents like Claude Code, Cursor and Codex, served over MCP from a SQLite file on your disk.

View on GitHubWebsite
Stars
284
Forks
32
Open issues
10
Last commit
2 months ago

What is it?

AI coding agents forget decisions, lessons and project context between sessions, so developers keep re-explaining the same constraints. PMB stores these as durable memory in a local workspace: SQLite is the source of truth, with rebuildable LanceDB vector indexes beside it. The agent reads memory through MCP tools such as prepare(message) and recall(query), which combine BM25, vector search, an entity graph and optional reranking.

On Claude Code and Codex, PMB can install lifecycle hooks that inject relevant memory before the model responds and record the agent's actions afterward. It also indexes project code, PDFs, Markdown and chat exports, and includes a local web dashboard on 127.0.0.1 for browsing entities, lessons and performance. The default embedding model covers 50+ languages, so a query in one language can find a fact stored in another.

The tool runs offline by default with no API keys, and its CLI supports snapshots, verification and export to Markdown or JSON. It is written in Python, installs with pip, and is licensed under Apache 2.0.

Key features

  • 01Stores decisions, lessons, facts and project structure in a local SQLite file.
  • 02Connects to Claude Code, Cursor, Codex, Windsurf, Zed and VS Code over MCP.
  • 03Hybrid recall combining BM25, vector search, entity graph and optional reranking.
  • 04Lifecycle hooks inject relevant memory before the model responds on supported agents.
  • 05Indexes project code, PDFs, Markdown and ChatGPT exports.
  • 06Local dashboard for browsing memory, and verified snapshots with restore.

Use cases

  • Keeping architectural decisions and project constraints available across coding sessions.
  • Giving Claude Code or Cursor persistent context about a codebase without sending data to a cloud service.
  • Searching personal notes, PDFs and past agent conversations from the terminal.

Quick start

  1. 1

    Install the package from PyPI

    pip install pmb-ai
  2. 2

    Wire PMB into your detected AI agent

    pmb setup
  3. 3

    Preload the embedding model so the first recall is fast

    pmb warmup
  4. 4

    Restart your agent, then confirm everything is wired with the doctor check

    pmb doctor

Things to know

  • The default embedder is about 450 MB and must be downloaded before first use.
  • Some features, such as hooks and the optional cross-encoder rerank, apply only to specific agents or setups.

In collections

Long-term memory for AI agentsOpen-source AI agent frameworks & appsMCP serversPrivacy & local-first toolsDatabases & SQL tools

Topics

#ai-agents#mcp#local-first#sqlite#memory#vector-search#rag#privacy#ai-memory#bm25

FAQ

What is pmb?
Local-first persistent memory for AI coding agents like Claude Code, Cursor and Codex, served over MCP from a SQLite file on your disk. AI coding agents forget decisions, lessons and project context between sessions, so developers keep re-explaining the same constraints. PMB stores these as durable memory in a local workspace: SQLite is the source of truth, with rebuildable LanceDB vector indexes beside it. The agent reads memory through MCP tools such as prepare(message) and recall(query), which combine BM25, vector search, an entity graph and optional reranking.
Is pmb free to use? What license does it have?
pmb is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is pmb still maintained?
Updated within 6 months. The last commit to pmb was on Sep 23, 2026.
What can I use pmb for?
Keeping architectural decisions and project constraints available across coding sessions. Giving Claude Code or Cursor persistent context about a codebase without sending data to a cloud service. Searching personal notes, PDFs and past agent conversations from the terminal.
How do I install pmb?
Install the package from PyPI: pip install pmb-ai Wire PMB into your detected AI agent: pmb setup Preload the embedding model so the first recall is fast: pmb warmup Restart your agent, then confirm everything is wired with the doctor check: pmb doctor
Where can I get pmb?
The source code and installation instructions are on GitHub: https://github.com/oleksiijko/pmb

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects