Skip to content

Lerim

Your coding agents forget everything after each session. Lerim learns — across all of them.

Lerim Logo

Lerim is the continual learning and context graph layer for AI coding agents — it watches sessions, extracts structured knowledge, and builds a shared intelligence graph across agents, projects, and teams. Current runtime architecture is PydanticAI-only for sync, maintain, and ask.

Lerim network animation

The problem

You spend 20 minutes explaining context to your coding agent. It writes great code. Next session? It's forgotten everything. Every decision, every pattern, every "we tried X and it didn't work" — gone.

And if you use multiple agents — Claude Code at the terminal, Cursor in the IDE, Codex for reviews — none of them know what the others learned. Your project knowledge is scattered across isolated sessions with no shared intelligence.

This is agent context amnesia, and it's the biggest productivity drain in AI-assisted development.

The solution

Lerim solves this by:

  • Watching your agent sessions across all supported coding agents
  • Extracting decisions and learnings automatically using a PydanticAI extraction agent
  • Storing everything as plain markdown files in your repo (.lerim/)
  • Refining knowledge over time — merges duplicates, archives stale entries, refreshes the memory index
  • Unifying knowledge across all your agents — shared files under .lerim/memory/
  • Answering questions about past context: lerim ask "why did we choose Postgres?"

No proprietary format. No database lock-in. Just markdown files that both humans and agents can read.

Get started

  • Quickstart

    Get from zero to first working command in under 5 minutes

    Quickstart

  • Installation

    Detailed installation instructions and prerequisites

    Installation

  • CLI Reference

    Complete command-line interface documentation

    CLI Reference

  • How It Works

    How Lerim works under the hood

    How it works

Key features

Multi-agent support

Works with any coding agent that produces session traces

Plain markdown storage

No proprietary formats — just .md files in .lerim/

Automatic extraction

PydanticAI agents extract decisions and learnings from sessions

Continuous refinement

Merges duplicates, archives stale entries, maintains index.md

Natural language queries

Ask questions about past context in plain English

Local-first

Runs entirely on your machine with Docker or standalone

Supported agents

Agent Session Format Status
Claude Code JSONL traces Supported
Codex CLI JSONL traces Supported
Cursor SQLite to JSONL Supported
OpenCode SQLite to JSONL Supported

More agents coming soon

PRs welcome! See the contributing guide to add support for your favorite agent.

How it works

Connect your agents

Link your coding agent platforms. Lerim auto-detects supported agents on your system.

lerim init
lerim connect auto

Sync sessions

Lerim reads session transcripts and runs the PydanticAI extraction flow with the [roles.agent] model. The agent uses tool functions to read the trace, take notes, prune context, search existing memories, write or edit markdown, and save a session summary:

flowchart TB
    subgraph runtime_sync["Agent flow"]
        RT[LerimRuntime · run_extraction]
    end
    subgraph lm["LM"]
        L[roles.agent]
    end
    subgraph syncTools["Sync tools (7)"]
        t1["read · grep"]
        c1["note · prune"]
        wm["write · edit"]
        v1["verify_index"]
    end
    RT --> L
    RT --> t1
    RT --> c1
    RT --> wm
    RT --> v1

Maintain knowledge

Offline refinement merges duplicates, archives low-value entries, and consolidates related learnings. The maintain flow uses the same [roles.agent] model with maintain-only tools:

flowchart TB
    subgraph runtime_maintain["Agent flow"]
        RT_m[LerimRuntime · run_maintain]
    end
    subgraph maintainTools["Maintain tools (6)"]
        t2["read · scan"]
        wm2["write · edit"]
        ar[archive]
        v2[verify_index]
    end
    RT_m --> t2
    RT_m --> wm2
    RT_m --> ar
    RT_m --> v2

Query past context

Ask Lerim about any past decision or learning. Your agents can do this too.

lerim ask "Why did we choose Postgres over MongoDB?"
lerim memory list

Dashboard

Dashboard UI is not released yet. The local daemon exposes a JSON API on http://localhost:8765 for CLI usage.

See Dashboard (Coming Soon).

Quick install

pip install lerim

Then follow the quickstart guide to get running in 5 minutes.

Next steps

  • Quickstart

    Install, configure, and run your first sync in 5 minutes

    Get started

  • Connecting agents

    Link your coding agent platforms for session ingestion

    Connect agents

  • Memory model

    Understand how memories are stored and structured

    Memory model

  • Configuration

    Customize model providers, tracing, and more

    Configuration