Summary
Fahd Mirza demonstrates the new `/learn` command in Hermes Agent, the open-source AI operator from Nous Research that runs in a terminal and connects to Telegram and Discord. The `/learn` feature — released just hours before the video was published — allows Hermes to ingest any source (a URL, a local directory, or pasted notes) and automatically distill a reusable `skill.md` file capturing the workflow, style, or codebase structure it discovers. The resulting skill becomes a callable slash command that loads automatically when relevant, eliminating the need for repetitive context-setting in future sessions.
Mirza walks through three live demos on Ubuntu using the GLM 5.2 model. In the first, he points `/learn` at his own blog (fahadmirza.com) and Hermes analyzes multiple articles to infer writing style, formatting conventions — including M-dash usage in section titles that was never explicitly stated — and post structure, saving everything as a reusable blog-drafting command. In the second demo, he targets a local FastAPI codebase for a World Cup 2026 application; Hermes maps five API endpoints, the SQLite schema, and frontend wiring into an application skill, and even flags stale bug comments it noticed while reading the code. A third demo shows `/learn` applied to an in-session conversation.
The feature requires no manual prompt engineering or additional tool configuration and is compatible with any Hermes-supported model. For developers managing complex codebases or writers maintaining consistent voice across platforms, `/learn` offers a practical path to persistent, grounded AI assistance without ongoing manual context management.
📺 Source: Fahd Mirza · Published June 26, 2026
🏷️ Format: Tutorial Demo
