English · 简体中文
Persistent, evidence-aware memory for Pi. Keep useful preferences, project context and work progress across sessions—without repeatedly asking the assistant to remember them.
- Learn and update automatically. Capture requirements, corrections and compaction summaries. Update tracked project states from tool observations, with provenance, history, gradual decay and evidence-aware ranking.
- Inject relevant context. Select memories by the current topic and recent user context. Filter weak or stale matches; no relevant match means no unrelated filler.
- Recall across sessions. Find background from another session or directory.
A read-only
memory_recalltool lets the assistant look up missing context mid-task.
Learning defaults to Pi's active model and existing authentication. Quota/rate limits or repeated failures can use an available model from another configured provider, without changing the foreground model. Fallback and budgets are configurable; no separate API key, embedding service or vector database is required.
Requires Pi 0.85+ with a working model. npm-based Pi requires Node.js 22.19+.
pi install npm:pi-memory-evolutionOr install the Git default branch, with Git and npm available locally:
pi install https://github.com/btnalit/pi-memory-evolutionChoose one source, then run these commands inside Pi:
/reload
/memory status
SQLite ok (schema 7) confirms storage initialization. Continue using Pi normally;
learning and recall run automatically.
Describe your requirements in a conversation:
Our priorities for atlas-service are automatic backups and reliable recovery.
Later, in a fresh session sharing the same memory store:
What do you remember about atlas-service's priorities?
Learning is asynchronous and selective—not every message becomes a memory.
| Command | Purpose |
|---|---|
/memory list |
Browse memories |
/memory search <topic> |
Search relevant claims |
/memory show <id> |
Inspect content and evidence |
/memory learning |
Inspect capture and actual update results |
/memory explain |
Explain the last automatic injection |
/memory correct <id> <text> |
Correct a record |
/memory forget <id> |
Suppress a record from recall |
See the usage guide for all commands, migration and troubleshooting.
For the npm installation:
pi update npm:pi-memory-evolution
pi remove npm:pi-memory-evolutionFor Git, substitute the repository URL used during installation. Reload or restart Pi afterward. Uninstalling does not delete memory data. Before a schema-changing upgrade, stop Pi processes sharing the database and back up the state directory.
Local SQLite state lives in ~/.pi/agent/agent-suite/memory-evolution/.
PI_CODING_AGENT_DIR changes that prefix; different working directories share the
same store by default.
Learning sends filtered source content to the active model, or an allowed fallback provider, and consumes that provider's quota. Cross-provider fallback is enabled by default; restrict its allowlist or disable it in recovery configuration. Memories are historical evidence, not independently verified facts. Matching and secret filtering are imperfect; verify important claims. See privacy and storage.
npm ci --ignore-scripts
npm run check
npm run test:install
npm run test:pi
npm run buildCI checks types, regressions, package contents, installation and fake-model host integration. Builds produce an installable npm archive and checksums, not a separate compiled runtime. Release PRs automate versions and changelogs; merging a verified release PR triggers npm publication. Dependency updates arrive as gated PRs.
See testing and release automation.