lerim 0.1.72

Continual learning layer for coding agents and software projects.

Persistent memory for coding agents.

lerim.dev · docs · pypi

Summary

Lerim is a memory layer for coding agents. It watches agent sessions, extracts durable memories with PydanticAI, and saves them as plain markdown in .lerim/memory/.

Why teams use it:

• Keep project decisions and reasoning across sessions.
• Share context between different coding agents.
• Ask questions against past work with lerim ask.
• Keep data local and file-based.

Supported session adapters today: Claude Code, Codex CLI, Cursor, and OpenCode.

How to use

Prerequisites: Python 3.10+ and Docker (recommended).

Install and bootstrap:

pip install lerim
lerim init
lerim project add .
lerim up

If you want a local Docker build instead of pulling from GHCR:

lerim up --build

Use --build from a source checkout (local Dockerfile available). For normal PyPI installs, use lerim up.

Daily flow:

lerim sync
lerim maintain
lerim ask "Why did we choose this approach?"

These commands call the running Lerim service (lerim up or lerim serve).

Quick validation flow before release:

lerim down
lerim up --build
lerim sync
lerim maintain
lerim ask "What are the latest important memories?"

Runtime model

Lerim runs three PydanticAI-based agent flows:

• sync: indexes sessions and extracts memories.
• maintain: merges duplicates, archives low-value items, refreshes memory quality.
• ask: answers questions with memory context and citations.

Memories are markdown files under project scope (<repo>/.lerim/memory/) with fallback in ~/.lerim/memory/.

Configuration

lerim init can set this up for you. API keys are read from environment variables (stored in ~/.lerim/.env by default).

# ~/.lerim/.env
MINIMAX_API_KEY=your-key
# add provider keys you use:
# OPENROUTER_API_KEY, OPENAI_API_KEY, MINIMAX_API_KEY, ZAI_API_KEY

Default provider example (MiniMax):

[roles.agent]
provider = "minimax"
model = "MiniMax-M2.7"
fallback_models = ["zai:glm-4.7"]

Commands

Most-used commands:

lerim status
lerim status --live
lerim logs --follow
lerim queue
lerim queue --failed
lerim unscoped --limit 20
lerim memory list --limit 20

Setup and management:

lerim connect auto
lerim project list
lerim project remove <name>
lerim skill install

Alternative to Docker:

lerim serve

Web UI

Web UI is not bundled in this repo yet. Use lerim dashboard for current status and CLI alternatives.

Docs

• Full docs: docs.lerim.dev
• CLI reference: src/lerim/skills/cli-reference.md
• Package source map: src/lerim/README.md

Development

uv venv && source .venv/bin/activate
uv pip install -e '.[test]'
tests/run_tests.sh unit
tests/run_tests.sh quality

Contributing

Contributions are welcome.

• Read the Contributing Guide
• Browse open issues
• Agent adapter PRs are especially appreciated -- see src/lerim/adapters/ for examples