Visit website Free to browse · updated 2026-06-29
What is PMB?
PMB functions as a local memory layer that agents access through the MCP protocol. It injects relevant stored information into prompts without requiring explicit tool calls from the agent and writes new events asynchronously to avoid blocking interactions. All data remains in a single user-owned file on disk. The included dashboard visualizes memory as an interactive entity graph and a chronological timeline of events. Lessons receive scores based on actual follow-through rates, allowing users to identify and prune ineffective entries over time. By maintaining continuity across different agents and sessions, PMB reduces repetitive explanations and preserves context when switching between coding environments. The approach emphasizes transparency, with all operations visible through terminal commands or the local web interface.
Key features
Local-first memory stored in a single SQLite file on disk
MCP-native integration with agents like Claude Code, Cursor, Codex, Zed
Automatic memory recall before agent reasoning (4-16 ms)
Hybrid recall using BM25 + dense vectors + entity graph with RRF
Live entity graph visualization (Map) and git-graph timeline journal
Lesson scoring based on agent follow-rate for pruning
Fully offline with no API keys or cloud dependency
What you can use PMB for
Persistent Agent Memory Across Sessions
PMB stores decisions, lessons, and project facts in a local SQLite file that AI agents can recall automatically on every prompt, eliminating the need to re-explain context.
Context Sharing Between Multiple Editors
The same local memory file is accessible by Claude Code, Cursor, Codex, and Zed, allowing seamless switching between tools without losing project history or conventions.
Visual Exploration of Agent Activity
The built-in dashboard renders memory as an interactive entity graph and git-style timeline, letting users inspect facts, decisions, and connections directly on their machine.
How to use PMB
- 1Run pip install pmb-ai in your terminal
- 2Launch your supported agent (Claude Code, Cursor, Codex or Zed)
- 3Enable the MCP integration so the agent connects to PMB
- 4Work normally while PMB records and recalls memory automatically
- 5Open the local dashboard to explore the Map and Timeline
PMB pricing
Pricing model: Open Source. Plan details are indicative — check the site for current prices.
Open Source
Free
- Apache 2.0
- Local-first memory
- No API keys
- No cloud
- MCP-native
Editor's verdict
Pros
- +100% local and private with Apache 2.0 open source license
- +Shared memory across multiple supported agents and editors
- +Sub-35 ms recall with 94.5% recall@10 accuracy
Cons
- –Requires local installation and MCP wiring per agent
- –Dashboard is a local web app only
Our take: PMB is a solid coding & dev choice. It's valued for 100% local and private with apache 2.0 open source license and shared memory across multiple supported agents and editors. The main trade-off is requires local installation and mcp wiring per agent. A good pick if you want capable AI without a high upfront cost.
Frequently asked questions
PMB runs entirely locally with no cloud dependency and requires no API keys; all memory is stored in a single SQLite file on disk.
Summary
PMB is a solid coding & dev choice. It's valued for 100% local and private with apache 2.0 open source license and shared memory across multiple supported agents and editors. The main trade-off is requires local installation and mcp wiring per agent. A good pick if you want capable AI without a high upfront cost.
Did you find this helpful?
User reviews
Verified reviews from the community shape this tool's rating.
Loading reviews…
PMB alternatives
Similar coding & dev tools worth comparing.
