AI Agents · Personal / internal build

Self-Hosted Autonomous AI Agent (OpenClaw)

A self-hosted, always-on autonomous AI agent (OpenClaw) running on a VPS with a custom web command center — a durable, recoverable agent rather than an ephemeral chat session.

OpenClawVPSSupabaseVercelGitHubTailscaleTypeScript

Self-hosted agent — gateway command-center dashboard

The problem

Off-the-shelf AI chats are ephemeral: no durable memory, no scheduled autonomy, no monitoring, no recovery. Running a real always-on agent means solving the operational problems, not just the prompt.

Approach

A VPS-hosted agent runtime behind a WebSocket gateway (over a private tailnet), a custom React command center on Vercel reading live status from Supabase, agent behavior defined in editable markdown core files, with the whole state version-controlled.

What I built

  • Always-on agent with scheduled (cron) autonomy and a live health/status dashboard
  • Multi-channel messaging (WhatsApp, Telegram, Discord, Signal, iMessage) over a private tailnet
  • Editable markdown persona/identity/tools/memory core files, edited remotely over SSH
  • Model-ladder configuration to route between models and control token cost
  • Disaster recovery: automated full VPS-to-GitHub sync plus scheduled database backups

Result

Built a production-grade autonomous agent that survives restarts, runs on a schedule, is monitored from a custom console, and is fully recoverable — the operational engineering an always-on agent actually needs.

Want something like this built?

Book a 30-minute discovery call and I'll scope your project.

AI Agent Services