OpenClaw — Personal AI Agent Fleet

Challenge

Most AI agent demos are single-shot proof-of-concepts. A real personal automation layer needs agents that survive across days, hand work to each other reliably, stay within safe action boundaries, and remain auditable after the fact.

What we built

We built a fleet of named, persistent AI agents — each with its own folder, encrypted secrets, and a Telegram identity. Agents are scheduled, monitored, and audited centrally. Every action is logged, replayable, and traceable to the human approver who authorised it. A separate orchestrator project handles deployment and loop scheduling across the entire fleet.

Result

The fleet runs continuously and handles tasks across multiple professional workflows daily. Telegram is the primary interface — alerts, approval requests, and results all land in a single thread per agent. The architecture has run stable for months of daily operation.

Tech used

• Long-lived LLM agent loops
• Telegram bot API
• Encrypted per-agent secrets
• Central orchestrator + audit log
• Human-approval gates