Open-weight general agent

Hermes

by Nous Research Configurable (human-in-the-loop to autonomous) Last reviewed June 2026

Hermes is the family of open-weight models and agent frameworks from Nous Research. The agent builds on the base model with tool use, function calling, and a configurable runtime that you can self-host or fine-tune.

Visit website
Coding Writing

Best for

  • Self-hosted, open-weight deployments
  • Users who want to fine-tune the underlying model
  • Privacy-sensitive or local-only environments

Not best for

  • Hosted-only workflows
  • Users who do not want to manage their own inference

Quick facts

Creator
Nous Research
Category
Open-weight general agent
Autonomy
Configurable (human-in-the-loop to autonomous)
Platforms
Self-hosted, Linux, macOS, Windows
Pricing
Free and open-weight. You pay only for the inference hardware or hosting you use.

In Depth: Hermes

What it can do

  • Open-weight base model
  • Function calling and tool use
  • Fine-tuning support
  • Local inference

Setup and requirements

  • GPU or hosted inference endpoint
  • Optional tool integrations

Safety and limitations

Agents take real actions on your behalf. Read this section before you connect anything important.

  • Open-weight — fine-tuning can change behavior
  • Run untrusted fine-tunes in a sandbox before connecting real data

Pricing and access

Free and open-weight. You pay only for the inference hardware or hosting you use.

Pricing changes often. We update this writeup as vendor pages change. Always confirm the latest on https://hermes.nousresearch.com before committing.