- CollectionMemory & State
- Classificationagent-native
- LicenseMIT
- MCPAvailable
- Agent SkillUnavailable
- InterfacesMCP · CLI · STDIO
- Verified2026-10-19
What it does
projectmem is a local-first typed event log plus a pre-action judgment gate for coding agents. Memory lives in .projectmem/ (append-only events.jsonl: issues, attempts, fixes, decisions, notes). Git hooks classify commits (revert → failed attempt, fix: → fix). pjm precheck warns before a commit if the staged files match a failed approach.
Judgment, not just recall: official positioning is “memory + judgment” — stale memories are flagged, never silently deleted; pjm score reports a Prevention Score (hours/tokens/dollars). Cross-project gotchas live in ~/.projectmem/global/. No cloud, no account, no telemetry (optional --online update check only).
Distinct from Claude-Mem (session firehose) and agentmemory (shared memory server): projectmem’s contract is a typed development log and a pre-commit / pre-action warning, not chat-memory extraction.
Agent-facing interfaces
- Interaction pattern: SDK / CLI + MCP
- Catalog protocols: MCP, CLI, STDIO
- MCP: ✅ Available — native stdio server, 17 tools (homepage); README comparison table also says “15 focused tools”
- Agent Skill: ⚠️ No official npx skills add package published yet.
- Onboarding type:
mcp
How agents onboard
Interaction pattern: SDK / CLI + MCP
pip install -U projectmem
pjm doctor
pjm doctor --fix
pjm init # per repo — writes .projectmem/ and git hooks
See the full onboarding section in the dossier.
Protocol surface
| Interface | Detail |
|---|---|
| CLI | pjm init, doctor, precheck, wrap, score, dashboard |
| MCP | python -m projectmem.mcp_server (stdio) |
| Store | .projectmem/events.jsonl + distilled markdown |
| Wrap | pjm wrap claude / Cursor rules / clipboard |
Verification
Last verified 2026-10-19.
Sources:
- https://api.github.com/repos/riponcm/projectmem