OpenClaw: open-source autonomous AI agent framework
OpenClaw has become the open-source reference for autonomous AI agents in just a few months, with over 250,000 GitHub stars. The framework runs locally on your machine or server under an MIT license, with zero vendor lock-in. Its ClawHub ecosystem gathers 13,000+ reusable skills and offers native integrations with Telegram, Discord, Slack and WhatsApp. The project, launched by Peter Steinberger (hired by OpenAI in February 2026), embodies the self-hosted AI agent trend. For developers seeking a serious alternative to proprietary agents, OpenClaw delivers a rare balance: full autonomy, extensibility via skills, and privacy since everything runs on your own infrastructure.
For who
Developers, technical teams, AI tinkerers, and data-sovereignty-focused companies who want a powerful AI agent without cloud dependencies.
Input
Instructions en langage naturel, skills ClawHub selectionnes, cles API LLM, configuration d'integration (Telegram, Slack, etc.).
steps (6)
Prerequisites and install
infoInstall Docker, Git and Node.js 20+. Clone the OpenClaw repo and run the install script to set up the local environment.
Configure API keys
infoSet LLM API keys (Anthropic, OpenAI, or local model via Ollama) in the .env file.
Install skills from ClawHub
infoBrowse ClawHub and install skills matching your use cases (web search, file operations, code execution, etc.).
Connect an interface
infoConfigure Telegram, Slack, Discord or WhatsApp integrations to chat with your agent from your existing tools.
First run and iteration
infoLaunch the agent, observe its behavior on a simple task, then iterate by adding skills and custom system instructions.
Deploy to production
infoHost OpenClaw on a VPS or 24/7 local machine. Enable logs, persistence and backups.
Output
Reponses de l'agent autonome : actions executees, resultats structures, fichiers generes et messages envoyes via les interfaces connectees.
Example
Input
Goal: "Watch new GitHub issues on acme/backend daily, summarize those tied to 500 errors, and post the summary to Slack channel #dev-ops."
Output
OpenClaw installs the github-watcher and slack-poster skills from ClawHub, schedules a cron job, filters issues by label and keywords, and posts a structured summary to Slack every morning with direct issue links.
Customization
| Parameter | Description | Default |
|---|---|---|
| Modele LLM | Choix du fournisseur LLM (Anthropic Claude, OpenAI GPT, ou modele local via Ollama). | Claude Sonnet |
| Skills installes | Liste des skills ClawHub actives sur l'agent, selon les cas d'usage vises. | web-search, file-ops, code-exec |
| Interface | Canal de communication principal avec l'agent (CLI, Telegram, Slack, Discord, WhatsApp). | Telegram |