How oleksiijko/pmb works
PMB is local-first persistent memory for AI coding agents. The main loop is agent activity → automatic hook capture → durable workspace storage; later, the agent requests context through MCP, which retrieves and returns relevant memories. Separate capabilities include project/PDF indexing, graph and timeline visualization, session restoration, and memory maintenance. The README documents these product areas; sampled source confirms parts of MCP serving, hybrid search, dashboard handling, and session tracking, so links for unsampled implementations are limited to documented behavior.
Main entry points: engine Memory engine · main.py PMB CLI · hooks Agent hooks · server.py Local dashboard
Components 20
Agent interfaces
- Agent hooks
src/pmb/hooks - MCP server
src/pmb/mcp/server.py - PMB CLI
src/pmb/cli/main.py
Memory runtime
- Memory engine
src/pmb/core/engine - SQLite workspace
src/pmb/core/sqlite_helper.py - Workspace management
src/pmb/core/workspace.py - Hybrid search
src/pmb/core/search.py - Embedding queue
src/pmb/core/embed_queue.py
Knowledge processing
- Entity graph
src/pmb/graph/store.py - Recall reasoning
src/pmb/reasoning/router.py - Project and PDF ingest
src/pmb/ingest - Activity signals
src/pmb/signals - Session tracking
src/pmb/signals/session.py - Session restore
src/pmb/hooks/session_restore.py
Workspace operations
- Memory health
src/pmb/health - Memory maintenance
src/pmb/maintenance - Verified snapshots
src/pmb/core/snapshots.py
Memory visibility
- Local dashboard
src/pmb/dashboard/server.py - Graph visualization
src/pmb/dashboard/viz.py
Other components
- Coding agent
Connections 20
- Coding agent → Agent hooks: triggers capture
- Agent hooks → Memory engine: records activity
- Coding agent → MCP server: requests memory
- MCP server → Memory engine: dispatches tools
- Memory engine → SQLite workspace: reads and writes
- Memory engine → Hybrid search: retrieves memories
- Hybrid search → Embedding queue: queues embeddings
- Memory engine → Recall reasoning: routes recall
- Memory engine → Entity graph: updates entities
- PMB CLI → Project and PDF ingest: indexes sources
- Project and PDF ingest → Memory engine: stores extracted content
- Agent hooks → Activity signals: collects context
- Memory engine → Session tracking: tracks sessions
- Session restore → Memory engine: builds session brief
- PMB CLI → Memory health: checks memory
- PMB CLI → Memory maintenance: runs maintenance
- PMB CLI → Verified snapshots: manages backups
- PMB CLI → Local dashboard: opens dashboard
- Local dashboard → Memory engine: reads workspace
- Local dashboard → Graph visualization: renders graph