API guide

doorflow uses the OpenAI API. Set the base URL, use a df- key, and choose a model.

Quick start

export OPENAI_BASE_URL=http://127.0.0.1:52005/v1
export OPENAI_API_KEY=df-…

Set these two values in your tool's provider settings. The fields may be called Base URL, API base or api_base.

Python

from openai import OpenAI

client = OpenAI(
    base_url="http://127.0.0.1:52005/v1",
    api_key="df-…",
)

reply = client.chat.completions.create(
    model="qwen3.8-flash-next-q2_0",
    messages=[{"role": "user", "content": "hello"}],
)
print(reply.choices[0].message.content)

JavaScript

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "http://127.0.0.1:52005/v1",
  apiKey: "df-…",
});

const reply = await client.chat.completions.create({
  model: "qwen3.8-flash-next-q2_0",
  messages: [{ role: "user", content: "hello" }],
});

curl

curl http://127.0.0.1:52005/v1/chat/completions \
  -H "Authorization: Bearer df-…" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-flash-next-q2_0","messages":[{"role":"user","content":"hello"}]}'

Add "stream": true to stream. Put the key in Authorization: Bearer <key> or x-api-key.

Check GET /v1/models first. If it returns models, your URL and key work.

Limits

Keys can limit requests, tokens, spend, request and token rate, and concurrent requests. Token limits cover input, cache write, cache read and output, and can be daily, monthly or lifetime.

When a limit is reached the gateway returns 429 with the limit and the amount used:

{
  "error": {
    "type": "insufficient_quota",
    "code": "insufficient_quota",
    "message": "You exceeded your current quota: daily request quota reached (500/500).",
    "limit": "requests_per_day",
    "limit_value": 500,
    "used": 500
  }
}

Rate limits also include requested, reset_s and a Retry-After header. Successful responses include x-ratelimit-remaining-* headers.

Check a key with the usage checker, or send GET /me with the key.

Priority

A key can carry a request priority. normal (the default) is served at once; high is served as soon as the current generation finishes; low only while the gateway has no other activity; and ultra preempts the requests in flight.

Errors

All errors use the same OpenAI shape: error.type, error.code and error.message.