Summary
Following his earlier DeepSeek V4 Pro and OpenClaw integration, Fahd Mirza turns to a different agentic framework in this tutorial: Hermes Agent by Nous Research. Where OpenClaw is terminal-first, Hermes has a built-in learning loop that creates reusable skills from experience, maintains memory across sessions, and runs natively inside Telegram — making it possible to trigger full autonomous coding workflows from a phone.
Mirza walks through the complete setup on Ubuntu: installing Hermes Agent via a one-liner, configuring DeepSeek V4 Pro through the API (noting that V4 Pro may not appear in the model list initially but becomes available shortly after launch), and creating a Telegram bot through BotFather to use as the command interface. Once paired, he verifies the integration responds correctly before moving to the real-world task.
The test application is a broken full-stack personal expense tracker — a FastAPI backend with SQLite and a frontend dashboard featuring charts, budget tracking, and multiple tabs. Bugs are deliberately planted across the codebase, and the readme bug map is removed before the test to give the model no hints. The entire prompt is sent through Telegram, and Hermes Agent powered by DeepSeek V4 Pro systematically reads every file, traces bugs across backend and frontend layers, and applies fixes. Mirza then runs the repaired application to confirm it works. The demo is a practical guide for developers who want mobile-accessible, session-persistent agentic coding without being tied to a local terminal.
📺 Source: Fahd Mirza · Published April 25, 2026
🏷️ Format: Tutorial Demo
