Your self-hosted AI agent - already running on VPS
OpenClaw is your private, autonomous AI agent that lives on your VPS and runs 24/7. It uses any LLM (Venice.ai, local Ollama, etc.) + real skills/tools to actually do things - read/write emails, control your browser, manage calendar/files, run shell commands, and more - while keeping all your data 100% private. Persistent memory, proactive cron jobs, multi-agent swarms, and native support for Telegram, Web UI, and TUI make it feel like a real digital assistant you own.
For comprehensive guides, troubleshooting, and advanced configuration, visit the official OpenClaw documentation:
Need the complete documentation index? Check out the llms.txt file to discover all available documentation pages.
openclaw tuiFor maximum security (no public port 18789), tunnel the dashboard through SSH so it feels like it's running locally on your Mac.
ssh -L 18789:localhost:18789 yourusername@your_VPS_IPUse this one-liner for background mode:
ssh -N -f -L 18789:localhost:18789 yourusername@your_VPS_IP
Add to your ~/.zshrc or ~/.bashrc:
alias openclaw-web='ssh -N -f -L 18789:localhost:18789 yourusername@your_VPS_IP && echo "✅ Tunnel created! Open http://localhost:18789"'
Run these from your SSH session. Start every login with openclaw doctor.
| Command | What It Does |
|---|---|
openclaw doctor | Quick health/security check (run this first!) |
openclaw onboard | Re-run setup wizard (models, channels, skills) |
openclaw gateway status | Backend server status |
openclaw logs --follow | Live logs |
openclaw skills list | See installed skills |
openclaw gateway restart → Restart after config/skills changesopenclaw gateway status --deep → Full health + channels/skillsopenclaw health → One-liner overviewopenclaw channels listopenclaw channels statusopenclaw channels add telegram → Fix Telegram if neededopenclaw models list → All available models + current active oneopenclaw models status → Usage, quotas, cost, fallback info (run this after any change!)openclaw models set <provider/model-or-alias> → Sets the permanent global default model for all new conversations and agentsopenclaw models set-image <provider/model> → Sets the vision/image model separatelyopenclaw models test → Quick speed + token usage benchmarkMake daily switching effortless with shortcuts and automatic fallbacks:
openclaw models aliases add fast venice/claude-4.6-sonnet
openclaw models aliases add opus venice/claude-4.6-opus
openclaw models fallbacks add ollama/gemma4:26b
Check them with openclaw models aliases list and openclaw models fallbacks list.
openclaw skills install <name> or openclaw skills search <keyword>openclaw skills updateopenclaw doctor --fix → Auto-fix common issuesopenclaw status --all --deep → Complete system viewopenclaw security auditopenclaw dashboard → Open Web UI from terminalalias oc='openclaw'
alias ocd='openclaw doctor --fix'
alias ocg='openclaw gateway restart'
alias ocl='openclaw logs --follow'
alias ocw='openclaw-web'
Everything lives in ~/.openclaw/workspace/ (or your configured path).
SOUL.md → Edit to change your agent's personalityMEMORY.md / AGENTS.md → Persistent memory & multi-agent swarmsskills/ folder → All your installed skillsopenclaw doctor regularlyWeb UI → Skills tab (or clawhub.ai). Start with:
Or just tell it in chat: "Install the browser skill".
"Every morning at 8am summarize my unread emails and send me a Telegram digest"
Or: "Remind me 30 min before every calendar meeting".
Edit AGENTS.md to create specialized sub-agents (coding, finance, research, etc.).
"Create a coding sub-agent that can push to GitHub"
| Problem | Fix |
|---|---|
| Web UI won't load | openclaw doctor --fix then openclaw gateway restart |
| Telegram silent | openclaw channels status → openclaw channels add telegram |
| Rate limit / model errors | openclaw models set to a cheaper model or add fallback |
| Changed model but still using old one | Run openclaw gateway restart after models set |
| Skills broken after update | openclaw skills update + gateway restart |
| Tunnel keeps dying | Use the openclaw-web alias with -N -f |
Still stuck? Run openclaw doctor and share the output in the OpenClaw community.
Start in TUI (openclaw tui) for fast testing, then live in Telegram + Web UI for daily use. You're already past the hardest part - now go make it do your life admin.
Play with skills and /think high - that's when OpenClaw stops feeling like a chatbot and starts feeling like a real teammate.
Drop your favorite OpenClaw skill, automation, or multi-agent setup in the comments below - I'll feature the best ones in a follow-up post!
Browse All Blog Posts →
💬 Comments