Hermes Agent desktop setup with Ollama and Telegram

This tutorial sets up Hermes Agent through its macOS desktop app, then connects a local LLM through Ollama's custom endpoint. The speaker uses Gemma 4 12B and creates a variant with a 65,536-token context window. He says Hermes needs roughly 64,000 tokens for agent tools, so checking the active context is part of troubleshooting. The news-search example uses local inference but still accesses the internet.

The cloud setup links an OpenAI account and selects a GPT 5.5 model. Separate profiles keep model defaults, sessions, skills and configuration apart. The speaker also explains persistent memory files, tool permissions, working directories and Docker execution settings. He describes this open source AI agent as improving through saved preferences and reusable workflows; the examples show memory and skill updates rather than model retraining.

The practical workflow creates a Telegram bot with BotFather, adds its token and an allowed user ID, then schedules a daily briefing at 8:00 a.m. Gemma struggles with messaging in his setup, so he switches to ChatGPT. Cron jobs start fresh sessions without messaging history. After feedback updates memory but leaves the reporting skill unchanged, he adds a second cron job to update that skill before the next report. He recommends a dry run and notes that the web tool needs external API keys, while browser automation can encounter blocking.