oleksiijko

oleksiijko / pmb

Public

Local-first persistent memory for AI coding agents (Claude Code, Cursor, Codex) via MCP. 94.5% LoCoMo recall@10, 70ms p50, multilingual, zero API keys.

pypi.orgprojectpmb-ai ai-agents ai-memory bm25 claude-code codex
15
1
89% credibility
Found May 30, 2026 at 16 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

PMB (Personal Memory Brain) is a local-first memory system for AI coding assistants that helps them remember your projects, preferences, and past decisions across all your coding sessions. You connect it to your AI tool with one command, then tell it facts about yourself and your work. When you return days or weeks later, the AI recalls everything without you repeating anything. Everything stays on your own computer — no cloud services, no accounts, no tracking. It supports multiple AI coding tools and works in 50+ languages.

How It Works

1
💬 You hear about a smarter AI coding assistant

A developer friend mentions they've been using an AI coding tool that actually remembers their projects, preferences, and past decisions across sessions.

2
🔌 You connect it to your AI coding tool

With one simple command, you link the memory system to your AI coding assistant so they start working together.

3
💾 You tell it what matters about you

You say things like 'remember I prefer Postgres' or 'my cat is allergic to chicken' and the system quietly stores these facts for future sessions.

4
🔍 You ask questions weeks later

Days or weeks later, you ask your AI 'what did I decide about the database?' and it instantly recalls your earlier conversations without you repeating anything.

5
🌍 Everything stays on your own computer

Your personal facts, project decisions, and preferences never leave your disk — no cloud services, no accounts, no tracking.

6
🔄 You share memory across multiple AI tools

If you use different AI coding assistants for different projects, they all connect to the same memory workspace so nothing gets lost.

✅ Your AI assistant becomes genuinely smart

Instead of starting every session from scratch, your AI knows who you are, what you've built, and what you care about — like having a colleague with a perfect memory.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 16 to 15 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is pmb?

PMB (Personal Memory Brain) gives your AI coding assistant a memory that persists across sessions. You install it, run `pmb connect codex` (or claude or cursor), and your agent remembers your preferences, decisions, and project context forever. It runs entirely on your machine using SQLite and LanceDB under the hood, never phoning home. The default multilingual embedder handles 50+ languages out of the box, and you can query your memory from the terminal with `pmb recall` or inspect it via a local web dashboard.

Why is it gaining traction?

The benchmark numbers are eye-catching: 94.5% recall on the LoCoMo standard and 70ms p50 latency. But the real hook is the zero-API-key, local-first model. Competitors like mem0, Letta, and Zep charge per-call and require cloud access. PMB stores everything on your disk, encrypts it if you want, and syncs via git. The README is unusually honest about what carries those benchmark numbers (spoiler: BM25 dominates; most of the 13 storage layers show zero delta on ablation). That transparency builds trust.

Who should use this?

Developers who switch between Claude Code, Cursor, or Codex and are tired of repeating context every session. Solo developers working across multiple projects who want their agent to "just know" their tech stack and preferences. Anyone with privacy concerns about sending conversation history to third-party services.

Verdict

Solid engineering with a skeptical, measurement-first culture baked into the project. At 15 stars it's early-stage and the CLI cold-start latency is still rough (the MCP server stays warm, but individual CLI calls pay ~14s). The roadmap is ambitious and the docs are thorough. Worth trying if you want local memory without vendor lock-in, but treat v0.2 as the stable target.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.