OpenClaw

OpenClaw is a local-first AI assistant that runs entirely on your machine. It connects to your chosen AI model and reaches the outside world through chat interfaces (Discord, Telegram, Slack, web UI, and more), performing agentic tasks such as web browsing, file operations, code execution, and scheduled jobs.

Unlike cloud-based AI assistants, OpenClaw keeps your data and conversations local: nothing is sent to a third-party service beyond the API calls you explicitly configure. You choose the model, you control the keys, and your workspace files never leave your machine without your knowledge.

LOCAL AI

On first interactive start OpenClaw detects a running lemonade-server (http://127.0.0.1:13305), offers to configure it as the model provider, and lets you choose an OpenClaw recipe from the Lemonade recipe catalog. You can rerun that setup any time with 'openclaw.lemonade'.

Or run 'openclaw.inference-snap' to pick a Canonical inference snap (gemma4, gemma3, deepseek-r1, and others) as the provider instead.

SETUP

Requires a valid API key for your chosen LLM provider, OR lemonade-server running locally for fully offline inference. Run the onboarding wizard on first launch:

openclaw onboard

The background gateway service starts automatically the first time you run any openclaw command and is managed as a standard systemd user unit:

systemctl --user status openclaw-gateway systemctl --user stop openclaw-gateway

SECURITY NOTE

This tool is capable of executing code, accessing files, and interacting with external services on your behalf. Due to the nature of autonomous AI agents, we recommend installing and running this snap inside an LXD container as best practice. A container provides an extra layer of confinement, limiting the agent's reach to the container environment and reducing exposure to your host system and personal data.

lxc launch ubuntu:24.04 my-agent lxc exec my-agent -- snap install --classic openclaw