Documentation

Everything to install and run your own AI agent.

1 · Install on Windows / Mac

Windows: download the installer and run it. If Windows shows "Unknown publisher" (normal for a free, unsigned app), click More info → Run anyway. SelfAgent installs per-user — no admin needed — and adds a Start-Menu + desktop shortcut.

⬇ Download SelfAgent-Setup.exe

macOS: download the .dmg, open it, and drag SelfAgent to Applications. First launch: right-click → Open (to bypass Gatekeeper for unsigned apps).

⬇ Download SelfAgent-Mac.dmg

The app opens its own window and runs everything locally — no terminal needed.

2 · Self-host on a Linux VPS

Run SelfAgent 24/7 on any Ubuntu/Debian VPS. One command installs Docker (if missing), builds, and starts it with auto-restart:

$ curl -fsSL https://selfagent.dev/install.sh | bash

Then open http://YOUR_VPS_IP:8080 in your browser. Useful commands:

$ docker compose logs -f          # live logs
$ docker compose restart          # restart
$ docker compose down              # stop

To expose it on a domain with HTTPS, put it behind a reverse proxy (Caddy/Nginx). Set a password in Settings → Login before exposing publicly.

3 · Add a model key

Open API keys in the app and paste at least one model key. Groq and Gemini have generous free tiers, so SelfAgent can run completely free. Supported: Groq, Gemini, DeepSeek, OpenAI, Anthropic, OpenRouter, xAI, Azure, AWS Bedrock, and any OpenAI-compatible/custom endpoint — plus local Ollama for fully offline use.

Click 🧪 Test all models and SelfAgent probes which models actually reply (and which can run tools), then uses only the working ones. Auto picks the best one per task.

4 · Using the agent

Just type (or speak) what you want. Examples:

Turn on Voice mode (headset icon) for hands-free conversation. Watch the agent's browser/desktop live in the Live Screen tab. Connect your own Chrome for logged-in, block-free browsing.

5 · Powers, Spec mode & Steering

Powers are on-demand capability bundles (expert guidance + tools) that activate only when relevant — install built-ins or import any from a GitHub URL. Spec mode: for big builds the agent writes a Requirements/Design/Tasks spec, you approve it, then it executes task-by-task. Steering rules are preferences the agent always follows; Hooks automate things before/after every reply. Autopilot plans and executes large multi-step goals end-to-end.

6 · Connectors (MCP)

Open the Connectors tab to add MCP servers in one click — Gmail, Calendar, Slack, Notion, Google Drive, GitHub, databases and more. The agent can also read your inbox over IMAP and send email/Telegram directly.

7 · Updating

The desktop app shows a "New version available" banner when an update ships — click to update & restart. On a VPS, re-run the install command or git pull && docker compose up -d --build.

8 · FAQ

Is it really free?

Yes. SelfAgent is free and open. You only pay (optionally) for whatever model provider you choose — and Groq/Gemini free tiers + local Ollama mean you can run at zero cost.

Is my data private?

Everything runs on your machine/VPS with your own keys. SelfAgent has no central server and collects no data. See our Privacy Policy.

Which model is best?

Use Auto. For actions (browser/computer) the agent uses a function-calling–capable model automatically. Run "Test all models" so it only uses ones that work with your keys.