Skip to content
ai agent · devops

DevOps Agent

Telegram-based DevOps agent: Claude in an agentic loop executes SSH commands on your servers — with permission levels and dangerous-pattern filtering as a safety layer.

Role
Full-Stack
Period
2025–2026
Status
Production

DevOps Agent lets you administer servers from a Telegram chat: you describe the task in natural language, and a Claude-driven agentic loop plans and executes SSH commands over asyncssh, streams results back and iterates until done.

The core design problem is safety: an LLM with shell access must not be able to destroy a system because of a hallucination. The agent implements permission levels and a dangerous-pattern filter that blocks destructive commands before they reach the shell.

Stack: Python, aiogram 3 for the Telegram surface, asyncssh for connections, SQLite for state. The agentic loop keeps conversation context so multi-step operations ("find the leak, restart the service, verify") work as a single session.

Key Decisions
Agentic loop: Claude plans, executes and iterates over SSH autonomously
Safety layer: permission levels + dangerous-pattern filter before the shell
Natural-language server administration from Telegram
Persistent context for multi-step operations
Tech Stack
Python Claude aiogram 3 asyncssh SQLite