Hermes Agent Bot

Hermes is an autonomous AI agent that lives on Telegram and manages real tasks. It’s not a chatbot — it’s an operator that reads messages, executes commands, writes code, deploys applications, and reports back. Running primarily on local infrastructure with cloud APIs available for frontier tasks when needed.

What It Does

• Code Generation — Writes, reviews, and refactors code in Python, TypeScript, Go, and Rust
• DevOps Orchestration — Deploys to Cloudflare, manages DNS, runs CI/CD pipelines
• Research & Summarization — Reads documentation, summarizes papers, searches the web via multiple providers
• File Operations — Reads, writes, patches, and manages files across the workstation
• Process Management — Runs shell commands, monitors logs, and reports on system health
• Long-Running Tasks — Handles multi-step workflows that can take minutes or hours

Architecture

The bot is built on a modular agent framework:

• Telegram Interface — Telethon-based, handles inline commands, file uploads, and threaded conversations
• Core Agent — State machine with tool use, context management, and error recovery
• Tool Registry — Extensible tools for file ops, shell, web search, git, deployment, and more
• Model Router — Switches between local models (Qwen, Granite, Phi) and cloud APIs based on task complexity
• Memory Layer — SQLite-backed conversation history and project state

Local-First Design

Hermes defaults to local Ollama models. For tasks that need heavier reasoning (like code review or complex architecture decisions), it can route to cloud APIs — but only when explicitly needed. Everything else runs on the local GPU at 40-60 tok/s.

The bot communicates with the secondary workstation for compute-heavy tasks and can delegate work across both machines in the Duo.