PMB | Local-first memory for AI

PMB maintains a single local database that records agent activity automatically through hooks. These hooks classify incoming messages and retrieve matching entries from the memory store in milliseconds, allowing the agent to reference prior context without explicit calls. The system also logs new outcomes after each turn to build an evolving record of project details. A visual dashboard presents the stored information as an interactive entity graph and chronological journal. Users can explore connections between facts, decisions, and lessons while the underlying hybrid retrieval combines keyword, vector, and graph methods to rank results. Lessons receive ongoing evaluation based on actual usage patterns to keep the memory relevant and concise. Integration requires a single setup command that links supported agents to the shared memory workspace. Once connected, multiple tools access the same records, enabling seamless transitions between environments while preserving full offline operation and user control over all data.