Integrations

Mixen speaks the OpenAI protocol, so it connects anywhere you can set a custom base URL. The recipe never changes: point at https://api.mixen.ai/v1 and use an mxn-... key.

Cursor

Settings → Models → enable Override OpenAI Base URL:

• Base URL: https://api.mixen.ai/v1
• API Key: your Mixen key
• Add a model whose identifier comes from the catalog, for example gpt-5.6-terra or claude-opus-4.8

Hit “Verify” afterwards — Cursor calls /models and confirms the key works.

Cline and Roo Code

Pick the OpenAI Compatible provider and fill in:

• Base URL: https://api.mixen.ai/v1
• API Key: your key
• Model ID: an identifier from the catalog

For agent work, choose models with a large context window and reasoning support — they hold up better across multi-step edits.

Open WebUI

Settings → Connections → OpenAI API:

• API Base URL: https://api.mixen.ai/v1
• API Key: your key

Open WebUI pulls the model list from /models itself, so the whole catalog shows up in the dropdown.

LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://api.mixen.ai/v1",
    api_key="mxn-...",
    model="gpt-5.6-luna",
)
print(llm.invoke("What is vector search?").content)

Vercel AI SDK

import { createOpenAI } from '@ai-sdk/openai'
import { generateText } from 'ai'

const mixen = createOpenAI({
  baseURL: 'https://api.mixen.ai/v1',
  apiKey: process.env.MIXEN_API_KEY,
})

const { text } = await generateText({
  model: mixen('gpt-5.6-luna'),
  prompt: 'Write a tagline for a book delivery service',
})

n8n, Make and other low-code tools

Anywhere an “OpenAI” node exists, the custom-host variant works: put https://api.mixen.ai/v1 and your key in the credentials. If a node hard-codes the OpenAI domain, use a plain HTTP request instead — the body format is identical.

Not supported

• Tool calling (tools, function_call) — not supported: the fields are ignored and the model replies with ordinary text. Clients that build agents on tools will not work.
• Assistants API, threads and OpenAI’s built-in file storage — not available.
• Batch API and fine-tuning — also not available; model training is not offered.

Web search does work, but as a model choice rather than a tool — see Chat and streaming.