Use Autark in n8n.

n8n's OpenAI Chat Model node accepts a custom base URL, so Autark drops in as the model behind any AI Agent, Basic LLM Chain, or chat workflow. One setting has to be changed or every request fails, and it is not the one people expect.

First: turn off “Use Responses API”

The OpenAI Chat Model node ships with Use Responses API switched on. That sends requests to /v1/responses, which Autark does not serve. The node reports:

404 page not found
The resource you are requesting could not be found

// This is the Responses API toggle, not a bad key or a bad URL.

Open the node's Options and turn Use Responses API off. The node then calls /v1/chat/completions and works. This catches everyone pointing n8n at an OpenAI-compatible provider, not just Autark.

Set up the credential

The credential only holds the key. Leave Organization ID blank: it is an OpenAI account concept and Autark ignores it.

1. Credentials → New → search "OpenAI"
2. API Key:        autark_sk_YOUR_KEY
3. Organization ID: leave blank
4. Save

// The Base URL is NOT here. It lives in the node's Options.

Point the node at Autark

The Base URL field is in the node's Options, not in the credential. This is the step people hunt for, because n8n's credential page does not mention it.

1. Add an "OpenAI Chat Model" node (under AI → Language Models)
2. Credential:      select the one you just created
3. Options → Add Option → Base URL
4. Base URL:        https://api.autark.ai/v1
5. Use Responses API:  OFF   <-- required, see below
6. Model:           autark-agent

// The base URL must include /v1. n8n appends /chat/completions itself.

The model list fills itself

n8n loads the dropdown from GET /v1/models using the Base URL you set, so all six Autark models appear in the list. If the dropdown is empty, switch the Model field to By ID and type the name instead: the request works either way.

Verify it works

Test the key outside n8n first. If this succeeds and the node still fails, the problem is the node's configuration, not your key.

curl https://api.autark.ai/v1/chat/completions \
  -H "Authorization: Bearer autark_sk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "autark-agent",
    "messages": [{ "role": "user", "content": "Reply exactly: OK" }]
  }'

// Returns a chat completion if the key and model are valid.

When the node is not enough

The OpenAI Chat Model node covers chat, agents and chains. For anything it does not expose, an HTTP Request node reaches the API directly. Use Header Auth with the name Authorization and the value Bearer autark_sk_YOUR_KEY.

POST https://api.autark.ai/v1/chat/completions

Headers:
  Authorization: Bearer autark_sk_YOUR_KEY
  Content-Type:  application/json

Body (JSON):
{
  "model": "autark-agent",
  "messages": [
    { "role": "user", "content": "Summarize this ticket in one line." }
  ]
}

// Works for any endpoint in the API reference, including streaming.

Troubleshooting

404 page not found

Almost always the Use Responses API toggle. Turn it off. If it is already off, check that the Base URL ends in /v1 and does not repeat it.

The model dropdown is empty

Switch the Model field to By ID and enter the name directly, for example autark-agent. The dropdown is a convenience, not a requirement.

401 or 403

The key is wrong, expired, or not being passed. Run the cURL check above. If that works, the credential is not attached to the node.

The AI Agent node returns nothing

Check the model supports tool calling. All Autark models do, but the agent needs a model connected to its Chat Model input, not just configured in the workflow.

All integrations API reference