PMB.

PMB: Local-First Memory for Your AI Coding Agent

PMB gives your AI coding agent persistent memory that lives entirely on your machine. Every decision, lesson, and project fact is stored in a SQLite file you own, automatically recalled through MCP before the model thinks—no API keys, no cloud, and recall in ~35 milliseconds.

Product Highlights

• Local-First Architecture: All data stays on your disk in SQLite and LanceDB—no network calls, no telemetry, no accounts required
• MCP-Native Integration: One command connects Claude Code, Cursor, Codex, Zed, and other MCP-aware agents to shared memory
• Hybrid Recall Engine: BM25 + dense vectors + entity graph fused with Reciprocal-Rank-Fusion delivers 94.6% recall@10 in under 35 ms
• Automatic Memory Injection: Hooks surface relevant context before the model reasons—no tool calls for the agent to remember
• Honest Memory Scoring: Tracks whether lessons get followed, flags dead rules, and surfaces what actually helps
• Live Dashboard: Interactive Map and Timeline visualize 5,000+ entities and connections as a browsable knowledge graph

Use Cases

• Cross-Session Context Preservation: Stop re-explaining your project—every session starts knowing your conventions, decisions, and past bugs
• Multi-Agent Workflows: Switch between Claude Code, Cursor, Codex, and Zed without losing context—the same memory follows you
• Offline Development: Work completely offline with zero API costs; optional local LLM support via Ollama
• Team Knowledge Retention: Shared workspaces let multiple agents (and teammates) build on accumulated project intelligence
• Memory Hygiene at Scale: Automatic decay, deduplication, and archival keep recall sharp after thousands of sessions

Target Audience

PMB is built for developers and teams using AI coding agents who are frustrated by context loss between sessions and want persistent, trustworthy memory without surrendering data to cloud services.