Nous Research

Nous Research

Nous Research is an open-source AI lab that trains models, builds agents and develops training infrastructure. It is known for the Hermes family of open-weight models, fine-tuned from base models such as Llama, and more recently for Hermes Agent, an MIT-licensed agent that remembers what it learns, turns experience into reusable skills and runs on your machine or in the cloud. Nous Research also runs Nous Portal, a subscription that gives Hermes Agent access to hundreds of models, hosted tools and cloud hosting.

Production credibility: Nous Research's releases include the Hermes 4 models from August 2025, up to a 405B version built on Llama 3.1, Hermes-4.3-Seed-36B trained on its decentralised Psyche network, NousCoder-14B and the Atropos reinforcement-learning framework. Hermes Agent, released in February 2026, had more than 260,000 GitHub stars on 8 November 2026; Nous describes it as the most widely used open-source agent of its kind, which is its own claim. In November 2026 it published Arete-27B, a fine-tune of Qwen3.8-27B with its eval, weights and training data. Hermes Agent itself is free. Nous Portal has a free pay-as-you-go tier limited to free models, Plus at $20 a month, Super at $100 and Ultra at $200, each adding paid models, hosted tools such as web search and a cloud agent.

Key Features

• Hermes open-weight models, including Hermes 4 up to 405B parameters
• Hermes Agent with persistent memory and self-generated skills
• Connects the agent to Telegram, Discord, Slack, WhatsApp, Signal and email
• Sandboxed execution on local, Docker, SSH, Singularity or Modal backends
• Nous Portal with 200+ models, hosted tools and one-click cloud hosting
• Open research releases such as the Atropos RL framework and Psyche network

Ideal Use Case

Developers and power users who want an always-on personal agent they can self-host and inspect, reachable from the chat apps they already use, and researchers who want open-weight Hermes models that they can download and fine-tune. Teams experimenting with fine-tuning also use Nous's datasets and RL tooling.

How Nous Research differentiates

DeepSeek and Mistral AI train their own base models and sell API access; Nous mostly post-trains existing open models and puts its effort into the agent layer. Against OpenClaw, Hermes Agent's case is its learning loop that writes and refines skills from experience, and an optional Portal subscription that bundles models, tools and hosting in one account.

FAQ

Q: What is Nous Research? A: Nous Research is an open-source AI lab that trains the Hermes family of open-weight models and builds Hermes Agent, an open-source AI agent with persistent memory and self-generated skills.

Q: What is Hermes Agent? A: Hermes Agent is Nous Research's MIT-licensed AI agent. It remembers what it learns across sessions, creates skills from experience, and runs from the terminal, a desktop app, messaging apps or the cloud.

Q: Is Nous Research free? A: Hermes Agent and the Hermes models are free and open. Nous Portal has a free pay-as-you-go tier limited to free models, with Plus at $20 a month, Super at $100 and Ultra at $200.

Q: What are the Hermes models? A: Hermes models are Nous Research's open-weight language models, fine-tuned from base models such as Llama and Qwen. Hermes 4 includes 14B, 70B and 405B versions with a hybrid reasoning mode.

Q: Can I self-host Hermes Agent? A: Yes. Hermes Agent installs on macOS, Linux and Windows from the terminal or as a desktop app, and works with your own model provider or with Nous Portal.

tl;dr

Nous Research is an open-source AI lab behind the Hermes open-weight models and Hermes Agent, a free, MIT-licensed self-improving agent. Its optional Nous Portal subscription starts free, with paid plans from $20 a month.