Descriptions:
Alex Finn makes the case for Hermes Agent — built by New Computer — as a lightweight, self-improving alternative to OpenClaw (Open Interpreter) for running a persistent AI agent on your local machine. The video covers installation from the terminal, model selection, and a direct comparison of Hermes versus OpenClaw across performance, update reliability, and cost: Hermes runs effectively on GPT 5.5 at $20/month, whereas Claude Opus requires the $200/month tier for comparable agent workloads.
Finn demonstrates several practical use cases: building a 3D rocket ship simulator in Three.js from a Telegram message while away from the computer, using the slash-steer command to inject mid-task corrections, and triggering multi-agent setups with five parallel Hermes instances. The self-improving skill system — where the agent builds and refines its own capabilities over time based on usage — is highlighted as the key differentiator over static agent frameworks.
The video also covers Hermes’s new portal plan ($20/month without requiring a separate ChatGPT or Claude subscription), making it accessible for users who want a capable desktop agent without managing multiple API accounts. For developers and power users evaluating persistent local agents, this is a useful hands-on overview of where Hermes currently stands relative to the competition, with honest notes on both strengths and the remaining rough edges of OpenClaw-style tooling.
📺 Source: Alex Finn · Published April 28, 2026
🏷️ Format: Tutorial Demo
