Connect n8n to OpenLLM Buddy

Use these values in your HTTP Request node (steps 1–5 below). You need a chat-completions URL (where n8n sends POST requests) and an API key (sent as a Bearer token). For this walkthrough we use placeholder names YOUR_ENDPOINT and YOUR_API_KEY — substitute your real values from the console when you go live.

The endpoint must be the full path to /v1/chat/completions, not just the host. OpenLLM Buddy speaks the same JSON format as the OpenAI Chat Completions API, so n8n does not need a custom integration — only the generic HTTP Request node.

YOUR_ENDPOINT

http://openllmbuddy-proxy.botbuddytech.workers.dev/v1/chat/completions

YOUR_API_KEY

ob_sk_000000000000000000000000000000000000000000000001

Keep your API key secret — do not commit it to git or paste it into public workflow exports. In n8n you can later move the key into credentials or environment variables; this guide uses a header field so you can see exactly what is sent on the wire.

Every call to OpenLLM Buddy from n8n goes through a single HTTP Request node. If you already have a trigger or data source (webhook, schedule, another app), add this node immediately after it so the model receives the incoming message as JSON.

1. Open your workflow in the n8n editor — create a new workflow or open an existing one.
2. Click the + button on the canvas (between nodes) or use the node panel on the right to add a step.
3. In the search box, type HTTP Request and select HTTP Request — the description should mention making an HTTP request and returning response data.
4. Leave the node selected; the next steps configure it on the Parameters tab.

You do not need the OpenAI node or any community node pack — the standard HTTP Request node is enough for OpenAI-compatible APIs.

With the HTTP Request node selected, open the Parameters tab in the right panel. These settings tell n8n where to send the request and which HTTP method to use. We use POST because chat completions accept a JSON body in the request.

• MethodPOST — required for chat-completions with a JSON payload.
• URLPaste your full completions URL, e.g. http://openllmbuddy-proxy.botbuddytech.workers.dev/v1/chat/completions (YOUR_ENDPOINT). Use http:// or https:// exactly as shown in the console — a missing scheme or wrong path will return connection errors.
• AuthenticationNone — we add the API key manually under Send Headers in the next step so you can see the Authorization header format.

Do not enable "Ignore SSL Issues" unless you are debugging a self-signed certificate in a dev environment. For production, fix the certificate or use a proper HTTPS endpoint.

OpenLLM Buddy expects your API key in the Authorization header using the Bearer scheme — the same pattern as OpenAI. In the same HTTP Request node, scroll down to Send Headersand add one header row; do not use Basic Auth or the generic "Predefined Credential Type" unless you have already stored the key in n8n credentials.

• Send HeadersToggle on so n8n includes custom headers on the request.
• Specify HeadersChoose Using Fields Below so you can type name and value explicitly.
• Header nameAuthorization — spelling and capitalization must match exactly.
• Header valueBearer ob_sk_000000000000000000000000000000000000000000000001 — the word Bearer, one space, then your YOUR_API_KEY with no extra quotes. A 401 response usually means a missing Bearer prefix or a wrong key.

If you rotate keys in the console, update this header (or your credential) before running production workflows again.

The request body tells OpenLLM Buddy which model to run and what message to answer. n8n sends it as JSON in the POST body. Scroll to Send Body on the same HTTP Request node and enable JSON mode — then paste the template below and adjust the model id if needed.

• Send BodyToggle on.
• Body Content TypeJSON — sets Content-Type: application/json automatically.
• Specify BodyUsing JSON — shows a JSON editor for the payload.

Use gemma4:26b for the model field (this page's deployment). The {{ $json.chatInput }} expression is an n8n placeholder: replace it with whatever field carries the user message from the previous node (e.g. {{ $json.body.message }} from a webhook).

Optional fields such as temperature and top_p control randomness; you can tune them later. The messages array must include at least one user role entry with non-empty content.

{
  "model": "gemma4:26b",
  "messages": [
    { "role": "user", "content": "{{ $json.chatInput }}" }
  ],
  "temperature": 1.0,
  "top_p": 0.95
}

Click Execute step on the HTTP Request node (or run the whole workflow). A successful call returns HTTP 200 with OpenAI-compatible JSON. n8n often wraps the parsed body in an array with one object — the assistant text you need is at choices[0].message.content.

Wire that path into the next node: a Set node, Slack message, email, or another HTTP Request. If you see an empty reply, open the execution data and confirm choices exists and finish_reason is stop rather than an error payload.

• content — the assistant reply (use this in downstream nodes)
• reasoning — optional; some models include chain-of-thought text
• usage — prompt, completion, and total token counts for billing and monitoring
• model — echoes the model id that served the request (should match gemma4:26b)

[
  {
    "id": "chatcmpl-145",
    "object": "chat.completion",
    "created": 1779715432,
    "model": "gemma4:26b",
    "system_fingerprint": "fp_ollama",
    "choices": [
      {
        "index": 0,
        "message": {
          "role": "assistant",
          "content": "The assistant's reply to your prompt appears in this field.",
          "reasoning": "(optional — included by some models)"
        },
        "finish_reason": "stop"
      }
    ],
    "usage": {
      "prompt_tokens": 31,
      "completion_tokens": 1417,
      "total_tokens": 1448
    }
  }
]