Docs SkillsModelsDev logAbout Roadmap Telegram Security

Docs

OverviewInstallProvidersDashboardLibre WebUITelegramSchedulesSkillsSecurity
GitHub GitLab mirror
OPEN SOURCE Local-first APACHE-2.0

Libre Claw

A terminal-native autonomous coding agent harness for developers who want real runs, visible tools, persistent memory, Telegram control, and local-first safety.

Start building View source OpenRouter app
OPENROUTERMOONSHOT AIOLLAMA CLOUDCODEX OAUTHANTHROPICOPENAILOCAL OLLAMA
libre-claw — Auditable DAEMON LINKED
Libre Claw terminal TUI showing a live agent session
RUN persistent MODEL Model routing STATE persistent TOOLS Permissioned

Four surfaces

Use the same agent from terminal, browser, or phone.

The daemon is the center of gravity. The TUI, Libre WebUI, dashboard, Telegram bridge, and schedules can all observe or resume durable runs.

  1. 01

    TUI

    interactive

    Interactive coding, file explorer, approvals, run timeline, and artifacts.

    libre-claw tui
  2. 02

    DAEMON

    persistent

    Background execution, dashboard, schedules, run polling, and Telegram supervision.

    libre-claw start -d
  3. 03

    TELEGRAM

    remote

    Remote chat, inline approvals, model switching, scheduled reports, and cancellation.

    libre-claw telegram up
  4. 04

    WEB UI

    visual

    Libre WebUI adds a browser control surface: run timelines, approvals, automations, and usage.

    libre-claw start
Local dashboardOpen the control plane
Libre Claw dashboard with runs, schedules, usage, and approvals
Telegram AI coding agentBot workflow
Telegram conversation with Libre Claw

Why it exists

The harness matters as much as the model.

Libre Claw is built around the practical parts that make agents useful in production: state, approvals, resumability, logs, cost visibility, screenshots, and memory.

About Libre Claw
  1. 01
    MESSAGETUI · WEB · TELEGRAM · CRON
  2. 02
    REASONPRIMARY + FALLBACK PROVIDERS
  3. 03
    ACTALLOW · ASK · DENY
  4. 04
    REMEMBEREVENTS · ARTIFACTS · MEMORY

Local-first

Built to leave evidence.

Libre Claw is built around the practical parts that make agents useful in production: state, approvals, resumability, logs, cost visibility, screenshots, and memory.

01

Local-first

Runs, memory, config, and artifacts live under your local home directory.

~/.libre-claw
02

Permissioned

Writes, shell, browser actions, git commits, and MCP calls ask first.

ALLOW / ASK / DENY
03

Auditable

Append-only JSONL logs make every tool call and decision reviewable.

EVENTS.JSONL
04

Portable

Skills, soul files, and provider routing move across projects and models.

SKILL.MD / SOUL.MD

Need a browser? Libre WebUI adds a full control surface over the same daemon — see the integration.

Five-minute start

Clone it, install it, run the TUI.

One repository. One daemon. Every surface stays attached to the same run.

Install guide GitLab mirror
bash zsh - install libre-claw 
git clone https://github.com/kroonen-ai/libre-claw.git
cd libre-claw
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -e ".[dev]"
libre-claw tui

FAQ

Short answers for builders evaluating the harness.

01Is Libre Claw local-first?

Yes. Runs, session archives, memory, configuration, browser profiles, and artifacts are stored locally under the user's home directory by default.

02Which model providers work?

Libre Claw supports Moonshot AI, OpenRouter, Ollama and Ollama Cloud, Anthropic, OpenAI, Codex OAuth through the Codex CLI, and local-compatible endpoints.

03Can Libre Claw run from Telegram?

Yes. The Telegram bridge connects to the same daemon and durable run store, supports slash commands, and can approve tool calls inline.