OpenClaw is the most-starred open-source project on GitHub right now — a self-hosted AI gateway that connects your local models and Claude API to every messaging app you use. This is the complete setup guide.
248,000 GitHub stars · Dual-node setup · ~$17–47/month to run
OpenClaw is a gateway — it sits between your messaging apps and your AI models. You send a message to your Telegram bot. OpenClaw decides whether to route it to your local Ollama model (zero cost) or Claude API (complex tasks), runs it, and replies — all within seconds.
You (Telegram): "Search for the latest papers on LLM reasoning"
→ OpenClaw detects: tool call needed
→ Routes to: Claude Sonnet (web_search skill)
→ Skill runs: DuckDuckGo search + page browse
OpenClaw: "Here are the top 5 papers from the past month..."
Total time: 3.2s · Cost: $0.008
Commercial AI assistant services charge subscription fees, log your conversations, and lock you into their model choices. OpenClaw flips the model: you pay for compute directly, your conversations stay on your hardware, and you can plug in any model — Ollama locally, Claude API for power tasks.
| OpenClaw (self-hosted) | Commercial services | |
|---|---|---|
| Cost | $17–47/month (usage-based) | $20–100/month (subscription) |
| Privacy | 100% on your hardware | Conversations logged by provider |
| Models | Ollama + any API | Provider's choice |
| Channels | WhatsApp, Telegram, Slack, Discord | Depends on plan |
| Customization | Full — skills, hooks, personas | Limited |
| Always-on | Yes (VPS) | Yes |
OpenClaw sits between your models and your messaging apps. One deployment covers WhatsApp, Telegram, Slack, and Discord simultaneously — with per-channel personas and access control.
Route simple queries to local Ollama (zero cost) and complex agentic tasks to Claude API. Most setups run 60–80% of queries locally, cutting API costs by the same margin.
Extend your AI with custom tools: web search, Google Calendar, file operations, and any API you can write a JavaScript function for. Hooks trigger automations on incoming events.
A $7/month Hetzner VPS keeps your assistant reachable 24/7. Tailscale mesh networking means no open ports — your deployment is invisible to the public internet.
Your conversations never pass through a third-party service. Your models run on your hardware. Your API keys stay in your .env file. Full control, full privacy.
Docker Compose deployment, automated backups, monitoring, update procedures, fail2ban, UFW, and a monthly security checklist — built into the setup from the start.
From first VPS to production-hardened, security-audited setup. Lesson 1 is free.
Architecture overview, cost model, and how it complements Claude Code.
Hetzner VPS + WSL2 + Ollama + Docker + Tailscale — the full 60-minute walkthrough.
Configure routing rules, complexity thresholds, and fallback chains.
Connect WhatsApp, Telegram, Slack, and Discord with access control and per-channel personas.
Build custom tools and event-driven automations in TypeScript.
Monitoring, backups, cost tracking, and incident response.
Tailscale ACLs, token rotation, Docker isolation, and the monthly security checklist.
Start with the free lesson — no signup required. Upgrade to Pro when you're ready for the full dual-node setup and production ops.
Part of the Fluent course