An AI agent is a language model in a loop that can take actions — run commands, edit files, browse, send messages. (The basics: agents and tool calling explained.) Here are popular open-source options and how to get started. Agent tools change very fast: always follow the project’s official README for the current install command.
In this article
The landscape
| Tool | What it is | Good for |
|---|---|---|
| OpenClaw | A self-hosted personal AI assistant that connects to chat apps (WhatsApp, Telegram and others), keeps memory and can run tasks on your machine | A personal assistant you control |
| OpenHands (formerly OpenDevin) | A coding agent that works in a sandbox: reads the repo, runs commands, writes code | Software tasks, fixing issues |
| Aider | AI pair programmer in your terminal that edits files in a git repo | Developers who like the command line |
| Open Interpreter | Lets a model run code on your computer to complete tasks | Data tasks, file conversions |
| n8n | Visual workflow automation with AI nodes | Business automations with an AI step |
| CrewAI / LangGraph | Python frameworks for building multi-agent apps | Developers building their own agents |
What you need first
- A model: an API key (OpenAI, Anthropic, Google, OpenRouter…) or a local model through Ollama.
- Node.js (for OpenClaw), Python 3.10+ (Aider, Open Interpreter, CrewAI) or Docker (OpenHands, n8n).
- Ideally a separate machine, VM or user account for agents that run commands.
Install sketches
OpenClaw
npm install -g openclaw@latest
openclaw onboard # guided setup: model provider, chat channels, permissionsAider
python -m pip install aider-install && aider-install
cd your-project && aider --model <your-model>Open Interpreter
pip install open-interpreter
interpreter # asks before running each command by defaultn8n (Docker)
docker run -it --rm -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8nThen open http://localhost:5678. OpenHands also ships as a Docker image — follow its README, as the image names and flags change between versions.
Common issues
| Problem | Usually |
|---|---|
| “Command not found” after install | The npm or pip bin folder is not on PATH; reopen the terminal or add it. |
| Agent loops or gets stuck | Weak/small model or vague goal. Use a stronger model and give one clear task with a finish condition. |
| High API bills | Agents make many calls. Set spending limits on the provider dashboard; use cheaper models for simple steps. |
| Local model too slow | Not enough RAM/GPU; use a smaller or quantized model, or a cloud API. |
| Docker errors on Windows | Enable WSL 2 and virtualisation in BIOS; start Docker Desktop first. |
Safety rules (read these)
⚠️ An agent that can run commands or read your messages can also be tricked. Web pages, emails or documents can contain hidden instructions (“prompt injection”). Researchers have repeatedly found exposed, misconfigured agent installations online.
- Run agents in a sandbox, VM or separate user; never as administrator.
- Give the minimum permissions and API keys; use keys with spending limits.
- Keep “ask before running commands / sending messages” switched on.
- Don’t expose the agent’s web interface to the internet without authentication.
- Install skills/plugins only from sources you trust, and keep the software updated.
✨ Ask AI about this article
Stuck on a step? Ask a question and the AI answers using this article.
Free · AI can be wrong