PMB
A free tool that gives AI coding assistants like Claude Code or Cursor a real memory — so you stop re-explaining the same project decisions to them every session.
🔗 Visit PMB
Description
Anyone who's used an AI coding assistant knows the frustration: you explain a project's quirks and past decisions once, close the session, and the next day the assistant has completely forgotten and you're re-explaining everything from scratch. PMB fixes that by giving these assistants a persistent memory — like a notebook they can actually read and write to — that lives entirely on your own computer, with nothing sent to any company's servers.
PMB (Personal Memory Brain) is a local-first memory layer for AI coding agents (Claude Code, Cursor, Codex, Zed) that stores decisions, lessons, and project facts in a single SQLite file on disk, exposed to agents over MCP (Model Context Protocol). It uses hybrid retrieval — BM25 full-text search plus dense vector embeddings plus an entity graph — reporting 94.6% recall@10 at roughly 35ms latency. It includes an interactive graph visualization of stored entities, a git-style timeline for reviewing what was learned when, and automatic memory decay so unused facts get archived rather than cluttering recall forever. It's Apache 2.0, has zero telemetry, requires no API keys, and can optionally use a local Ollama model for offline summarization.
💬 Our review
The short version: if you use Claude Code, Cursor, or a similar AI coding agent and are tired of re-explaining your codebase's history every session, PMB is a genuinely free, local, no-catch way to give it persistent memory.
What makes it worth trying over just pasting context into a system prompt is the retrieval quality — combining keyword search, embeddings, and an entity graph is a more serious approach than a flat notes file, and the reported 94.6% recall number backs that up. The honest tradeoff is that this is still a young, single-file SQLite tool rather than a managed service, so there's no cloud sync or team-sharing story out of the box, and you're trusting a smaller open-source project with something that becomes load-bearing for your workflow over time. Compared to Mem0, which is a hosted, API-first memory layer aimed more at production AI applications, PMB's differentiator is being fully local and specifically tuned for the coding-agent use case — worth adopting if privacy and zero cost matter more to you than a polished managed backend.
📊 Global score
🤖 AI-enriched data
100% gratuit, licence Apache 2.0, aucune télémétrie, aucune clé API requise.
Pros
100% local, aucune donnée envoyée à un serveur, aucune clé API
Recall hybride (BM25 + vecteurs + graphe d'entités) avec 94.6% recall@10
Décroissance automatique des souvenirs inutilisés
Compatible avec les principaux agents de code (Claude Code, Cursor, Codex, Zed) via MCP
Cons
Pas de synchronisation cloud ni de partage d'équipe natif
Projet open-source plus jeune, moins de garanties qu'un service managé