Hermes Agent adds Hugging Face provider with 28 curated models
Hermes Agent now treats Hugging Face as a first-class inference provider and surfaces 28 curated models in its picker, plus a custom path to the broader catalog. That broadens model choice for a persistent local agent workflow without requiring users to wire a provider manually.
TL;DR
- Nous Research's launch thread says Hermes Agent now treats Hugging Face as a first-class inference provider, so users can pick HF-hosted models directly from the agent instead of wiring a provider manually.
- Teknium's update post says the new picker exposes 28 curated models by use case, with a custom entry path for the broader Hugging Face catalog of "100+ other models they serve."
- The merged change shown in the PR screenshot organizes those models into categories like flagship, reasoning, coding, vision, and small/fast, mapping several entries to existing OpenRouter analogues.
- Hugging Face CEO Clément Delangue's demo post frames the integration as part of Hermes Agent's persistent workflow: an open-source agent with memory that can now run models such as GLM-5 "served through Hugging Face."
What changed in Hermes Agent?
Hermes Agent now has Hugging Face as an official model provider, and the entry point is simple: Teknium's announcement says users get it via hermes update. The practical change is in model selection. Instead of surfacing a raw provider list, Hermes now shows 28 curated HF models organized by task, while still leaving an "Enter custom model name" path for the much larger catalog shown in the PR screenshot.
That curation is doing product and runtime work at the same time. The categories in the PR screenshot include flagship models such as Qwen3.5-397B and DeepSeek-V3.2, reasoning models such as DeepSeek-R1, coding models including Qwen3-Coder-480B, plus vision and small/fast options. The same screenshot says providers with 15 or more curated models "skip the live /models probe entirely," which keeps the picker "clean and fast" instead of dumping 124-plus live API results into the UI.
The integration also broadens the kinds of open models Hermes users can run inside the agent's existing workflow. Delangue describes Hermes as an agent that "remembers what it learns" with "persistent dedicated machine access," and his demo post shows a live session reporting "I'm running on zai-org/GLM-5, served through Hugging Face." For engineers, the launch is less about a new agent capability than about faster access to a wider model set inside the same persistent agent loop.