Skip to content
vavio

Can I use Hermes Agent without an OpenAI or Anthropic account?

Yes. Hermes Agent is open-source software from Nous Research, and it needs at least one model provider, but that provider doesn’t have to be OpenAI or Anthropic. Nous Research’s provider guide lists Nous Portal, OpenRouter, DeepSeek, Z.ai (GLM), Kimi, Qwen, MiniMax, Google Gemini and many others, and says Hermes works with any OpenAI-compatible API endpoint, including one you run yourself with Ollama, vLLM or llama.cpp. Avoiding those two companies is not the same as keeping your data private, though: unless the model runs on hardware you control, your prompts go to whichever company serves it.

Four things “without OpenAI or Anthropic” can mean

People asking this question usually want one of four different things. Each gives you something different and leaves a different dependency in place.

What you want What Hermes offers What you still depend on
Agent software you can inspect and keep Hermes Agent itself is open source. Its memory, skills and settings are yours, whatever model you choose. A model provider, because Hermes does no thinking of its own.
An open-weight model Hermes can use models whose weights are published, such as GLM, DeepSeek, Kimi or Qwen. Whoever serves that model, unless you run it yourself. An open-weight model on a hosted API still runs on someone else’s computer.
A different provider at a lower cost Pay-per-use keys from providers such as OpenRouter, DeepSeek or Z.ai, or a subscription sign-in such as Nous Portal. That provider’s prices, limits and data terms.
Prompts that never leave your hardware A self-hosted endpoint on your own GPU, using Ollama, vLLM, SGLang, llama.cpp or LM Studio. Hardware big enough to run a model that can use tools well.

Decide which of these you actually need before you choose a route. Someone after a cheaper model and someone who wants nothing to leave the house need different setups.

Routes that need no OpenAI or Anthropic account

Hermes’s provider guide groups model access into three kinds of route. All three include options from other companies.

  • Sign in with a subscription. Besides ChatGPT and Claude plans, the guide lists sign-ins for Nous Portal, xAI SuperGrok, Qwen and MiniMax. Hermes’s work counts against that plan’s limits.
  • Use a pay-per-use API key. The guide lists dozens of key providers, including OpenRouter, DeepSeek, Z.ai, Kimi/Moonshot, MiniMax, Google Gemini, Fireworks, Hugging Face, NVIDIA and AWS Bedrock. OpenRouter is a single account that reaches many models from many companies.
  • Point Hermes at an endpoint. Run hermes model and choose a custom endpoint, or set model.provider: custom and model.base_url in Hermes’s config file. Any server that speaks the OpenAI-compatible API will do, including one running on your own machine.

If either check fails, see why Hermes rejects a model or ignores its tools for the startup reproduction and server checks.

Two requirements apply whatever you choose. Hermes needs a model with at least 64,000 tokens of context for agent use with tools, and it rejects smaller windows at startup. The model and the server running it must also support tool calling; for example, the guide says llama.cpp needs its --jinja flag, or it ignores Hermes’s tools entirely.

Where your prompts go

Hermes sends the model your messages and the material it is working with, such as tool results and the contents of files it reads. The company serving the model receives all of that, so read its data terms rather than judging by where the model was made.

  • Through OpenRouter. OpenRouter’s data collection guide says it doesn’t store your prompts or responses unless you opt in to logging, though it keeps metadata such as token counts. It forwards each request to an upstream provider, and that provider’s own terms apply. Its “ZDR only” privacy setting excludes endpoints that retain prompts for abuse monitoring.
  • Directly with a model maker. Terms vary a lot. DeepSeek’s privacy policy, last updated 10 February 2026, says it stores personal data in the People’s Republic of China and lists training its models among its uses, with a right to opt out.
  • On your own endpoint. Prompts stay on hardware you control, provided Hermes and the model both run there.

Using Hermes on Vavio without OpenAI or Anthropic

On Vavio, you connect a model on Vavio’s secure setup page, then choose which model Hermes uses. The route Vavio has tested end to end is an OpenRouter API key. In Vavio’s own production test on 10 November 2026, an owner entered an OpenRouter key on a Vavio computer, Hermes was set to Z.ai’s GLM 5.3, and it replied to a message. No OpenAI or Anthropic account was involved. Other providers on Hermes’s list haven’t been tested on Vavio yet.

GLM 5.3 is an open-weight model: Z.ai publishes its weights on Hugging Face under its own licence. At 753 billion parameters, running it yourself takes data-centre GPUs, which is why most people reach it through a provider. On 11 November 2026, OpenRouter’s GLM 5.3 page listed several providers serving it, including providers in the EU. Check current prices there on the day you choose.

You don’t need ChatGPT or Claude to use Vavio either. You can buy hosting, connect the model and open Hermes on the Vavio website. Connecting Claude Code or Codex to manage your Hermes is optional.

When hosted Hermes is the wrong choice

Vavio’s Hermes hosting plan is a computer with 8 GB memory and no GPU. It doesn’t include a local model, so a Vavio Hermes always sends its prompts to the provider you connect. If every prompt must stay on hardware you own, run Hermes on that hardware beside your model. Should Hermes Agent run on my own computer or in the cloud? covers that choice.

Hosting suits you if what you want is model freedom without keeping a server alive: Hermes stays online while your laptop is off, on a provider you picked. The model provider bills its usage separately from the $14.99 hosting price. How much does it cost to run Hermes Agent each month? shows how to add the two together.

Next step

Sources

Written by Vavio with AI assistance and checked against the sources listed at the end of the article.

More from the Vavio blog: Hermes Agent guides