Use Infersia with Hermes Agent

Nous Research’s terminal and desktop agent

Hermes Agent talks to any OpenAI-compatible endpoint, so this takes about two minutes and no code. The result is Hermes running on DeepSeek V4 Flash at its full 1,048,576-token context — which matters for an agent, because a long session is exactly what runs out of room first.

~2 minutesNo codeWorks on the free tier

Before you start

Hermes Agent installed. If hermes --version prints a version in your terminal, you’re set.
An Infersia API key. Create one at infersia.com/dashboard/keys. It starts with isk-v1-. Copy it somewhere now — the dashboard shows it once.

No credit card needed to try: qwen/qwen3-8b:free runs at no cost, up to 50,000 tokens a day.

Know your way around already?

These are the only values that change. The walkthrough below is the same thing, slowly.

Base URL: https://api.infersia.com/v1
Provider type: custom
Model: deepseek/deepseek-v4-flash-0731
API key: isk-v1-… create one

Setting it up

Open the provider wizard

In a terminal — not inside a Hermes chat session — run:

hermes model

This is the full setup wizard. It is the command that adds new providers and enters API keys.

Choose the custom endpoint option

From the provider list, select the entry named:

Custom endpoint (self-hosted / VLLM / etc.)

Infersia runs vLLM behind an OpenAI-compatible API, so this is the correct choice even though nothing here is self-hosted by you.

Enter the three values

The wizard asks for each in turn:

Base URL:  https://api.infersia.com/v1
API key:   isk-v1-your-key-here
Model:     deepseek/deepseek-v4-flash-0731

Any model id from our catalogue works here. Start with qwen/qwen3-8b:free if you’d rather not spend anything while testing.

Check it works

Start a session and ask it something:

hermes chat

# then, in the session:
> Say hello in exactly three words.

A reply means you’re done. The request will also appear in your activity log within a few seconds, with its token counts and exact cost.

Or edit the config file directly

The wizard writes to ~/.hermes/config.yaml. If you prefer to edit it yourself, or you’re scripting a machine setup, this is the equivalent:

~/.hermes/config.yamlyaml

# ~/.hermes/config.yaml
model:
  default: deepseek/deepseek-v4-flash-0731
  provider: custom
  base_url: https://api.infersia.com/v1
  api_key: isk-v1-your-key-here

That file is the single source of truth for model, provider and base URL. Older guides mention an LLM_MODEL environment variable — that has been removed from Hermes and is ignored.

Switching models later

Once the endpoint is configured, you can change model without leaving a session:

/model custom:qwen/qwen3-8b        # switch to a specific model
/model custom                       # auto-detect from the endpoint

The one that catches everyone: `hermes model` vs `/model`

/model, inside a session, only switches between providers you have already set up. It cannot add Infersia for the first time. If it doesn’t appear, leave the session with /quit, run hermes model from your terminal, finish the wizard, then start a new session.

Which model to pick

`deepseek/deepseek-v4-flash-0731`

The default choice for agent work. Tool calling, and the full 1M context — an agent session accumulates history fast, and this is the one that doesn’t hit a wall.

`qwen/qwen3-8b:free`

Free, 50,000 tokens a day, no card. Good for confirming the setup works before you decide anything.

`stepfun-ai/step-3.7-flash`

Accepts images, including inside tool results — useful if your workflow screenshots things and asks the agent to look at them.

If something goes wrong

“Invalid API key” or a 401: Check the key was pasted whole — they are long and a partial copy is the usual cause. Confirm it independently of Hermes:
curl https://api.infersia.com/v1/models \ -H "Authorization: Bearer isk-v1-your-key-here"
A JSON list of models means the key is good and the problem is in the config.
“Model not found”: Model ids include the publisher prefix — qwen/qwen3-8b, not qwen3-8b. The exact strings are on the catalogue page, each with a copy button.
Infersia isn’t in the model list: You are probably in a chat session using /model. See the warning above — the first-time setup has to happen from the terminal.
The context limit seems smaller than 1M: Some clients carry their own idea of a model’s window and refuse a request before it reaches us. Ours reports the real figure at /v1/models — if a client disagrees, the limit is on its side, and the fix is in its model configuration.

A shorter path, once it lands

We have submitted a provider plugin to Hermes Agent, which would make Infersia a named entry in the setup picker instead of a custom endpoint — an INFERSIA_API_KEY in ~/.hermes/.env and nothing else to configure. It is open upstream and not merged, so the custom endpoint above remains the way to do this today. We will update this page if that changes.

Still stuck?

Write to support@infersia.com with what you tried and the error text. If you maintain a tool and want your users routed here, the partner programme pays you a share of what they spend.

Hermes Agent is a product of Nous Research. Its name and logo are theirs, used here only to identify the software these instructions are for. Infersia is not affiliated with or endorsed by Nous Research.