pi-memsearch

Long-term memory for pi, in the store your other coding agents already write to.

Every pi session starts from zero. pi ships no memory by design (“primitives, not features”), so this package adds it. Each session writes what it learned to a memory file, and weeks later pi finds it again by meaning, in whatever words you use then. The store is the same per-project one Claude Code, Codex, OpenClaw and OpenCode write through memsearch. Memory is plain markdown under .memsearch/; the searchable collection is derived from it and rebuildable at any time.

Features

• Recall in the phrasing you use weeks later: 32/35 strong hits against pi-memory’s 26/35, over 35 queries on an identical 223-file corpus (method and per-query results)
• Cross-agent: pi recalls what Claude Code learned yesterday in the same repo, and the reverse
• Writes itself: the agent_settled hook distills every content-bearing exchange into today’s memory file, in the background, on the session provider’s cheapest model
• Three recall layers: scored chunks, then the full section, then the turns around it in the origin transcript
• Plain markdown: commit .memsearch/ to share memory with collaborators, or gitignore it to keep it personal
• One external dependency: uv. memsearch runs through uvx, so there is no Python packaging to manage

Quick start

pi install npm:pi-memsearch        # all projects
pi install npm:pi-memsearch -l     # this project only

First run downloads the onnx embedding model once, about 560 MB. It is announced as a notice so the pause is not mistaken for a hang. No API key is involved.

Then leave it alone. The memory writes itself, and answers weeks later:

# .memsearch/memory/2026-08-13.md  ← written by the session, unprompted
### 22:41
- the user and the agent moved the hot cache to Redis with 5 minute TTLs

# a new session, three weeks on
you ▸ /recall how did we fix the flaky redis test?
pi  ▸ memory_search → 5 chunks; top: 2026-08-13 "moved the hot cache to Redis with 5 minute TTLs" (0.81)
      memory_expand → the full "### 22:41" section, with its session anchor
      → answered at layer 2; the origin transcript was never opened

Where the rest lives

• Install: prerequisites, the four install forms, first run, uninstall
• Configuration: every PI_MEMSEARCH_* variable, and the line between pi’s config and memsearch’s own
• Tools: the seven tools, and the three-rung recall ladder
• Memory store: the daily markdown files, and which store a session writes to
• Runtime: hook-by-hook behavior, every tunable, every degradation path
• Troubleshooting: start with memory_status
• Limits: what is deliberately unsupported, and how it compares to pi-memory
• Development: the mise tasks, and the no-build-step jiti loading