Vibeleaderboard
← All Intel
Intel / video

Hermes Architecture EXPLAINED: Memory, Context & Gateways

Source
youtube.com
Author
Hugging Face
Date

Context

A Hugging Face walkthrough of the Hermes agent explains that an agent like this is a simple loop. Each time a user sends a message, Hermes builds a , sends it to the , runs any tools the requests, feeds results back until the model stops asking, and returns a reply. A final step then checks whether anything from the exchange is worth saving to memory.

The context is built from a few markdown files: soul.md for the agent's personality, user.md for facts it learns about you, and memory.md for other notes about workflows and tools, plus message history and descriptions of its skills and tools. Past-conversation summaries appear only if external memory is configured, and long conversations are summarized past a threshold. The presenter says the memory step is what makes Hermes keep improving; that is his description of the design, and this portion does not test it.

Terms in this piece · Glossary
  • Context and context window — Context is the information available for a generation; the context window is the capacity available to hold it.
  • Large language model — A model trained on large amounts of data to process and generate language.
  • Model — A computational system that uses learned patterns to produce an output from information supplied to it.
  • Prompt — Information or instructions supplied to a model; in a chat, the request you type is often called your prompt.
Read the source www.youtube.com
More from Hugging Face
Recommended reads
Comments

Checking sign-in…

Loading comments…