Openclaw is a unified AI agent control plane running on a single VM you own. Claude Code sessions, Hermes autonomous agents, scheduled morning briefs, and semantic search — all reachable from Telegram and managed through one dashboard.
Claw runs Claude Code in sandboxed containers. Hermes runs autonomous AI with skill learning and 200+ models via OpenRouter. Both are reachable from Telegram, both managed from the same dashboard at claw.biswas.me.
Automated daily routines delivered straight to Telegram. Six brief types, full 5-field cron scheduling, CRUD via the dashboard UI and REST API, and HTML-formatted results with clickable links. Each routine runs as a Go-native fetcher — no Python, no external dependencies. Toggle, edit, or trigger any routine from the operator console.
The /learn command searches learn.biswas.me and enhances results with Claude. Get summaries, relevant excerpts, and direct links — all from Telegram.
Openclaw runs Claude Code in YOLO mode — on purpose. That's the whole value prop. So the rest of the stack is hardened to compensate: no inbound surface for the agent, container-per-session, everything on your box.
Bootstrap issues a Cloudflare origin cert, hardens the box via Ansible, deploys the bot container, and wires up nginx. You bring a Telegram token and your Anthropic key.
# 1. bootstrap a fresh ubuntu 22.04 vm $ curl -fsSL openclaw.dev/install.sh | bash # 2. paste your secrets when prompted ? TELEGRAM_BOT_TOKEN ******** ? TELEGRAM_ALLOWED_IDS 8417342395 ? ANTHROPIC_API_KEY sk-ant-******** ✓ CF dns record created ✓ origin cert issued (15yr) ✓ ansible playbook green (22 tasks) ✓ openclaw online on claw.biswas.me # DM @clawdy on telegram to begin. $
Every release cuts from main, tagged and signed. The roadmap lives in the same repo as the code.