Does my code or data ever leave my machine?

No. Everything lives in a local SQLite file with vectors in LanceDB. No network calls on read path, no account, no telemetry.

How is this different from RAG or a vector database?

Recall is hybrid (BM25 + dense vectors + entity graph) and automatic – memory is injected before the model thinks. No pipeline to build.

Will it slow my agent down?

No. Recall takes ~35ms, writes <1ms (async). The turn is never blocked.

Which agents and operating systems are supported?

Any MCP-aware agent: Claude Code, Cursor, Codex, Zed, Windsurf. PMB runs on macOS, Linux, and Windows.

What if a memory is wrong or unhelpful?

PMB scores whether each lesson is actually followed and flags dead ones so you can prune them.

Yes. Apache-2.0 open source, free forever. No paid tier, no seats.

AI Developer Tools

PMB

Q: Which agents and operating systems are supported?

Any MCP-aware agent: Claude Code, Cursor, Codex, Zed, Windsurf. PMB runs on macOS, Linux, and Windows.

Q: What if a memory is wrong or unhelpful?

PMB scores whether each lesson is actually followed and flags dead ones so you can prune them.

Q: Is it really free?

Yes. Apache-2.0 open source, free forever. No paid tier, no seats.

Local-first memory for AI coding agents. Automatically stores and recalls project context via MCP, offline, with ~35ms recall.

PMB

Visit website

What is PMB?

PMB is an open-source, local-first memory system for AI coding agents. It automatically stores and recalls project context (decisions, lessons, facts) using hybrid search (BM25 + dense vectors + entity graph) and hooks into agents like Claude Code, Cursor, and Codex via MCP. All data resides on your disk in a single SQLite file; no cloud, no telemetry.

PMB vs Similar AI Tools

	PMB	mcploitable	Cynative	Magpie
Pricing Model	Free	Free	Free	Free
Free Credits
Key Features	Local-first: all data on your disk in SQLite + LanceDB Hybrid recall (BM25 + dense vectors + entity graph) with RRF fusion Auto-recall on every agent prompt (4–16 ms read)	Seven breakable boxes covering OWASP Agentic Top-10 vulnerabilities Three guided simulations for cascading failures, human-agent trust, and rogue agents Network-isolated Docker containers for safe execution	Code-to-runtime reasoning across cloud, Git, and Kubernetes Action-gate enforces read-only policy on every API call Sandboxed JavaScript execution for concurrent research	Append-only, SHA-256-addressed event history through Jaybase AES-256-GCM encryption for stored node payloads Unified RBAC for ledger, notes, snapshots, and audit reads
Pros	100% local and offline Open source (Apache 2.0)	Covers full OWASP Agentic Top-10 in a realistic manner Docker isolation prevents accidental damage	Read-only by construction prevents accidental writes Evidence-backed verification cross-checks every finding	Opinionated and secure accounting CLI with immutable audit trail Designed for AI agent integration with JSON output
Cons	Requires Python environment (pip install) Limited to MCP-compatible agents	Requires Docker and technical setup Not for production use; only for lab environments	Requires an LLM API key, incurring token costs Limited to read-only operations, cannot remediate	Pre-1.0, limited feature set No native QuickBooks import (agents must normalize data)
Best For	Developers using AI coding agents who want persistent project memory Teams sharing context across multiple agents	Security researchers focusing on AI agent vulnerabilities Developers building MCP-based applications	Security engineers DevOps teams	Small teams needing secure, auditable accounting with AI agent support Developers integrating automated bookkeeping workflows

How to use PMB?

1Install: `pip install pmb-ai`
2Connect your agent: `pmb connect claude-code` (supports Cursor, Codex, Zed, etc.)
3Work as usual – PMB records and recalls automatically.
4Open the dashboard: `pmb dashboard`

PMB Key Features

Local-first: all data on your disk in SQLite + LanceDB
Hybrid recall (BM25 + dense vectors + entity graph) with RRF fusion
Auto-recall on every agent prompt (4–16 ms read)
Sub-millisecond async writes (<1 ms, non-blocking)
Lesson scoring: flags unused memories for pruning
Live entity graph dashboard with timeline and map
Works with multiple agents (Claude Code, Cursor, Codex, Zed, Windsurf)
Fully offline, no API keys, no telemetry
Apache 2.0 open source

PMB Use Cases

Stop re-explaining your project to coding agents every session
Switch between agents (Claude Code ↔ Cursor ↔ Codex) without losing context
Maintain honest, self-cleaning memory that highlights helpful lessons

PMB Pricing & Free Credits

PMB currently operates on a Free model.

This tool is completely free to use

Open Source (Apache 2.0)

Free

Fully free, no paid tier, no seats, no telemetry. You own the file and the code.

PMB Pros & Cons

Pros

100% local and offline
Open source (Apache 2.0)
Fast recall and writes
Works with multiple MCP-aware agents
Automatic memory hygiene (decay, archive, dedup)
No API keys or cloud dependencies

Cons

Requires Python environment (pip install)
Limited to MCP-compatible agents
Still relatively new, smaller community

What is PMB best for?

Developers using AI coding agents who want persistent project memory
Teams sharing context across multiple agents
Anyone valuing data ownership and offline operation

PMB FAQ

Top free alternatives to PMB

Openbase

A voice-controlled IDE that enables developers to initiate AI coding sessions with Codex or Claude Code, approve commands, and review diffs from their phone.