Descriptions:
This step-by-step tutorial covers the complete setup of Hermes Agent, an open-source AI agent framework, built entirely on free and local tools for maximum privacy. The video walks through cloning the GitHub repository, configuring Ollama as a local model server, downloading Google’s Gemma 4 model, and wiring everything together with a Telegram bot — all without relying on paid cloud APIs.
For the model layer, the creator shares direct testing results: the 7GB Gemma 4 E2B variant failed to follow web search instructions reliably, while the 9.6GB E4B model handled those tasks correctly. The Ollama API endpoint (localhost:11434/v1) is used to expose locally downloaded models to Hermes Agent, and the video demonstrates how to list available models and switch between them at runtime.
Configuration details include setting context compression to 0.5 (critical for keeping smaller models coherent over long conversations), choosing a session reset strategy — either a daily 4 a.m. reset or an inactivity timeout — and enabling full tool visibility to monitor every micro-step the agent executes. The tutorial also covers creating a Telegram bot via BotFather, pasting the bot token into Hermes Agent, and the creator’s recommendation to run the agent on a dedicated device rather than a primary machine to limit file system exposure.
📺 Source: Bart Slodyczka · Published April 08, 2026
🏷️ Format: Tutorial Demo
