I Let an AI Agent Live on My VPS for Three Weeks

Short summary

A developer ran an AI agent continuously on their VPS for three weeks, automating infrastructure ops via Telegram commands backed by bash scripts and markdown context files. The agent monitors 12 containers, restarts failures, and runs health checks—enabling remote incident response. Key lesson: agents need proper documentation and tool mapping to avoid wasting context on reconnaissance.

• •AI agent runs in Docker on VPS, receives commands via Telegram with bash tool access
• •Uses markdown files (SOUL.md, USER.md, CLAUDE.md) to provide project context and prevent context-wasting reconnaissance
• •Automates routine ops: log tailing, container restarts, health checks, database queries, config updates

Generated with AI, which can make mistakes.