Use Dolphin LLM in n8n workflows

Create the credential

1. In n8n open Credentials, click Add credential and choose OpenAI.
2. Paste your key into API Key. Leave Organization ID empty.
3. Set Base URL to https://dolphinllm.com/v1.
4. Save. n8n tests the credential by asking for the model list; a success message means the key and URL work.

Create the key in your account. If you have none yet, sign up; $0.50 of trial credit is added when you first open the account.

Use it in an AI Agent or LLM chain

1. Add an AI Agent or Basic LLM Chain node.
2. Under Chat Model, add OpenAI Chat Model and pick the credential you created.
3. Set Model to By ID and enter uncensored (or pick it from the list if n8n shows it).
4. In Options, set Maximum Number of Tokens (for example 800), Sampling Temperature and a Timeout of 120 seconds or more for long answers.

The API supports tool calling, so the AI Agent node can use n8n tools with this model. JSON mode (response_format: json_object) is available when you need structured output for the next node.

Or call it with an HTTP Request node

For full control over the request body, use a plain HTTP Request node:

• Method POST, URL https://dolphinllm.com/v1/chat/completions
• Authentication: Generic Credential Type → Header Auth, name Authorization, value Bearer YOUR_KEY
• Body content type JSON:

{
  "model": "uncensored",
  "messages": [
    {"role": "system", "content": "Summarize the text in three bullet points."},
    {"role": "user", "content": "{{ $json.text }}"}
  ],
  "max_tokens": 400,
  "temperature": 0.3
}

The answer is in choices[0].message.content, and the token counts in usage.

Limits, retries and batches

• A trial key allows 2 parallel requests and 60 a minute; after the first top-up, 8 parallel and 300 a minute. Workflows that process many items at once can hit HTTP 429 (concurrency or rate_limited).
• Use Loop Over Items with a batch size of 2 to 8, or the node's batching options, to stay under the limit.
• Turn on Retry On Fail in the node settings with a wait of a few seconds. 429 and 503 (upstream_busy) answers are safe to retry and are never charged.
• 402 no_credit means the balance is empty; retrying will not help. Add an error branch that notifies you.

Cost

A workflow that summarizes 1,000 articles a day at 2,500 tokens in and 300 out spends 1,000 × ($0.000375 + $0.00015) = $0.525 a day, about $16 a month. The pricing page has the calculator and the top-up bonuses.