rclabs · distillateexperiment EXP-007
runningOpenClaw Admin Dashboard
- Question
- When agents run on their own all day, how do you know what they are doing?
- In short
- Mission control for my personal AI agents
- Started
- 2026
- Apparatus
- PythonHTMLRailway
- Specimens
- Source ↗
- Write-up
- Read the full story of how it works →
Curious about
Autonomous agents with cron jobs and memory files are a bit like having employees you never meet. I wanted a window into them.
Built
A stdlib-only Python server that asks the openclaw CLI for agent status and cron jobs and reads the agents’ memory files. It’s protected by basic auth, per-IP rate limiting and security headers, and deployed on Railway. On top sits a mobile-first dashboard with panels for agents, cron, sessions, logs, memory and actions.
The catch, and the reason this is still running: the dashboard is still rendering sample data and hasn’t been wired to the server yet.
Learned
- Observability comes first for agents, exactly as it does for servers. The DevOps playbook transfers almost unchanged.
- Monitoring has to be verified, not generated. A screen showing sample data is confidently wrong.
- An admin panel is an attack surface. Secure defaults are not optional.
next specimen → EXP-008 · Apple Health → Railway