MD
Back

automation

Hermes Agent — Self-Hosted AI Operations Assistant

A personal AI agent running on my own VPS: it lives in Telegram and executes real tasks through the tools I configure, from inspecting the server to editing files and running commands.

Role
Solo: deployment, container setup, integrations, maintenance
Period
2026 – present
Stack
Hermes AgentDockerLinux VPSTelegram Bot APIModel routingScheduled jobs

Results

  • Runs 24/7 on a self-managed VPS with a supervised process that restarts on its own
  • Executes work instead of suggesting it: reads servers, edits files, runs commands, reports back in chat
  • Model routing through a local gateway, so the agent keeps working when I switch providers
  • Secrets and data stay on my own infrastructure, not a third-party SaaS

Problem

Running automation for other people means a long tail of small operations: checking whether a service is up, editing a config, keeping backups, writing updates. Doing each one by hand, or paying a SaaS for every single one, costs more time than the work itself.

What I built

A self-hosted AI agent (Hermes Agent, open source) running in Docker on my VPS:

  • Telegram gateway so tasks can be delegated from a phone, not just a terminal.
  • Supervised service manager so the agent restarts itself and survives reboots.
  • Tools and skills wired to real actions: shell, files, browser, scheduled jobs.
  • Model routing through an OpenAI-compatible gateway, with keys kept in local environment files.
  • Scheduled jobs so recurring work (health checks, digests, backups) runs without a terminal.

Result

One assistant that executes instead of answering: it inspects the server, edits the file, runs the command, and reports back in the same chat. Everything runs on infrastructure I control, and nothing sensitive leaves it.