Questions founders actually ask

OpenClaw is an open-source, local-first AI agent platform. Unlike cloud AI chatbots, it runs entirely on your own hardware — a laptop, Mac Mini, or VPS. It connects to your messaging apps (Telegram, WhatsApp, Discord) so you can interact with it naturally, and it executes real tasks: running shell commands, reading/writing files, browsing the web, and calling APIs. Think of it as a personal AI teammate that actually does work, not just responds.

Yes. OpenClaw itself is free and open-source (MIT License). The only cost is your AI model API usage — for example, Anthropic Claude or OpenAI GPT-4o charges per token. If you use Ollama with a local model, even this is free. Infrastructure costs (VPS, server) are optional and depend on your setup.

Cloud AI services are stateless and isolated — they can't access your files, run commands, or act proactively. OpenClaw runs on your hardware, which means: (1) your data never leaves your machine, (2) it has full access to your local environment, (3) it can run scheduled tasks 24/7 without a session, and (4) you control exactly what model it uses. It's the difference between a chatbot and an actual agent.

OpenClaw is designed for founders, solo builders, indie hackers, technical operators, and AI-first teams. You need to be comfortable running a terminal command and following setup steps — but you don't need to write code to use it effectively. If you can install Node.js and type a few commands, you're good to start.

For most people, 5–10 minutes from zero to a working Telegram bot. The onboarding wizard walks you through everything interactively. The most common time sink is creating a Telegram bot via BotFather (2 minutes) or gathering your API key. Discord takes a bit longer due to the Developer Portal setup, but still under 15 minutes total.

OpenClaw uses modern JavaScript APIs that aren't available in older Node.js versions. Older versions cause silent failures — no helpful error messages, just a Gateway that doesn't start. Always run node --version first. We recommend v24.

Check in this order: (1) Run openclaw gateway status — if the Gateway isn't running, start it. (2) Run openclaw doctor for a full diagnostic. (3) For Discord: verify Message Content Intent is enabled in the Developer Portal. (4) Check that your API key is valid and has quota remaining. (5) Open the Dashboard at http://localhost:18789/ui to see error logs.

The pairing code is a security feature that prevents unauthorized users from interacting with your agent. When you connect a Telegram channel, OpenClaw sends a one-time code to your bot's chat. You enter this code during setup to confirm that you control both the CLI and the Telegram account. It's not something you create — it arrives in your Telegram chat.

For most founders: Anthropic Claude Sonnet gives the best balance of speed, reasoning, and cost. For coding-heavy tasks, Claude is particularly strong. For cost-sensitive setups, DeepSeek is excellent value. If data privacy is critical, use Ollama with a local model — completely free, no data leaves your machine. You can switch models at any time without losing context.

Yes. Install Ollama (ollama.com), pull a model (e.g., ollama pull llama3.3), and select "Ollama" as your provider during OpenClaw setup. This is completely free and 100% private — zero data leaves your machine. Performance depends on your hardware. For most research and writing tasks, 7–13B parameter models work well.

Yes. Edit your AGENTS.md file to change the model configuration, then restart the Gateway. You can also set different models per agent if you have multiple agents running.

No. OpenClaw has no central server. All agent configuration, conversation history, files, and API keys stay on your machine. The only data sent externally is what you send to your chosen AI model provider (Anthropic, OpenAI, etc.) — governed by their privacy policies. If you use Ollama, nothing leaves your machine at all.

We strongly recommend creating dedicated bot/service accounts instead. OpenClaw can execute shell commands and access files — you want that running under an isolated identity, not your primary personal account. For Telegram, create a new bot via @BotFather. For WhatsApp, get a dedicated SIM or number. This protects you from accidental data exposure and permission conflicts.

Built-in skills are safe. Community skills should be reviewed before use — check what tools they request access to and what instructions they include. Skills requesting shell access should be treated with extra caution: they can run arbitrary commands on your machine. Only install skills from trusted sources or review them yourself before activating.

Just message it like you'd message a person. "Research the top 5 alternatives to Linear." "Summarize my unread Slack messages." "Draft a cold email for this prospect." The agent picks the right skill, executes the task, and responds. No special commands or syntax required for basic usage.

Yes — this is the heartbeat feature. Define scheduled tasks in your SOUL.md (e.g., "every morning at 7AM, send me a briefing"). The agent runs these autonomously and messages you with results. This is how you build a real 24/7 assistant rather than just a chatbot you have to poke.

Yes. OpenClaw supports multi-agent routing. You can have a "work" agent connected to Slack, a "personal" agent on Telegram, and a "coding" agent on Discord — all running simultaneously on the same machine, each with their own workspace, skills, and model configuration.

When installed with --install-daemon, OpenClaw registers as a system service and starts automatically on reboot. However, it only runs 24/7 if your machine is on 24/7. For true always-on, deploy on a VPS (DigitalOcean, Hetzner) or a dedicated home server. Don't rely on a laptop you take to meetings.