Call an OpenAI-compatible API without hardcoding a key.

Configure one explicit HTTP Request

Use the official node behavior as the integration boundary

The official n8n HTTP Request documentation supports arbitrary REST URLs, generic credentials, headers, JSON bodies, cURL import, response settings, batching, proxy configuration, and timeouts. The credential reference lists Bearer Auth as a generic credential type.

Fetch a current model ID before opening n8n

test -n "$PRIMORDIAL_API_KEY" || {
  echo "PRIMORDIAL_API_KEY is not set" >&2
  exit 1
}

curl --fail-with-body --silent --show-error \
  https://www.primoraihub.com/v1/models \
  -H "Authorization: Bearer $PRIMORDIAL_API_KEY" \
  | jq -r '.data[]?.id'

Copy one exact returned ID into the first smoke test. Do not use a guessed name, an old blog example, or the literal MODEL_ID_FROM_V1_MODELS placeholder.

Store only the token in Bearer Auth

1. In the HTTP Request node, select Generic Credential Type.
2. Select Bearer Auth and create a dedicated credential.
3. Paste only the Primordial AI token. Do not add the word Bearer.
4. Name the credential by environment and purpose, then restrict its allowed domain to www.primoraihub.com when that control is available.

n8n's Bearer Auth implementation adds the Authorization: Bearer prefix. Adding it inside the token field can produce a duplicated prefix and an HTTP 401.

Start with a static, typed JSON smoke test

Enable Send Body, choose JSON, select Using JSON, and replace the model placeholder:

{
  "model": "MODEL_ID_FROM_V1_MODELS",
  "messages": [
    {
      "role": "user",
      "content": "Reply with OK"
    }
  ],
  "stream": false
}

Execute the node once. A successful response verifies only this token, model, endpoint, non-streaming body, and n8n execution at that time. It can consume account quota.

You can also select Import cURL in n8n. Import a cURL command without a real Authorization header, then attach the Bearer credential afterward. n8n documents that imported parameter values are strings, so re-enter the body in JSON mode to preserve false as a boolean.

Map workflow data only after the static test passes

If the previous node emits a prompt field, change the JSON body to an n8n expression that returns an object:

{{
  {
    "model": "MODEL_ID_FROM_V1_MODELS",
    "messages": [
      {
        "role": "user",
        "content": $json.prompt
      }
    ],
    "stream": false
  }
}}

Validate missing, empty, oversized, and non-string input before the request. Keep the original workflow item or a correlation field if downstream nodes must match responses to inputs.

Retry transient failures, not configuration errors

The n8n HTTP Request common-issues guide documents batching and Retry on Fail. Set a finite maximum and wait interval; do not use automatic retries to hide authentication, route, or schema defects.

Keep credentials out of nodes and exports

• Use a dedicated key for the workflow and environment.
• Never paste a real key into a header field, JSON body, sticky note, execution name, or exported workflow.
• Restrict who can edit or run the workflow and who can use the credential.
• Prune or limit execution data if prompts or responses contain sensitive business information.
• Rotate the key after suspected exposure and update only the n8n credential.

n8n's workflow-sharing documentation says editors can run workflows that use credentials even when the credential is not separately shared. Treat workflow access as permission to consume the provider account.

Frequently asked questions