Hermes Agent: VPS setup, Telegram and reusable skills

This tutorial sets up Hermes Agent as a self-hosted assistant on a VPS and connects it to Telegram. The speaker describes it as an open source alternative to OpenClaw and reports fewer setup problems in personal use. That comparison reflects the speaker's experience rather than a reliability benchmark.

The demonstration uses Ubuntu 24.04 LTS on a sponsored hosting.com server with four CPU cores and 8 GB of RAM. After connecting through SSH and installing a dependency, the speaker follows the full setup wizard. Model options include subscriptions, API keys and a local model, but this walkthrough uses an OpenAI ChatGPT/Codex subscription with 5.4 Mini. It does not demonstrate local LLM inference. The gateway runs as a background service, and Telegram setup automatically detects the user's ID.

The main workflow asks the AI agent to summarize recent Hacker News model announcements, produce a PDF and save the process as a reusable skill. The speaker invokes that skill through a phone voice memo and a fresh terminal conversation, then schedules a daily version for 21:30 Vienna time.

Scheduling needs follow-up: the agent reports an overly precise time check, and a later PDF arrives after roughly three minutes and 30 seconds. The speaker changes models but concludes that the change was unnecessary because the scheduled result arrives afterward.