APIPool Quickstart

Base URL

https://app.apipool.dev

The base URL contains no /v1 and no protocol path. Choose the protocol by model and group. A key can access enabled models in its group; this does not mean every model supports every endpoint.

• Text — use /v1/responses for models that support Responses, or /v1/chat/completions for models that support Chat Completions. Follow the model guide.
• Images — use the Images request format and asynchronous task polling; see Image models and protocols. Do not substitute image model IDs into the text examples below.
• Vendor-native formats — use only explicitly enabled endpoints, such as Anthropic /v1/messages. Native Gemini image-generation endpoints are not currently enabled.

Some SDKs and tools append /v1 themselves. Follow their documentation or the examples below when configuring them.

Choose a model and protocol

This protocol table does not guarantee continuous model availability. Check the catalog status and your key's group before following the matching guide.

Authentication

Send your key in the Authorization header:

Authorization: Bearer $APIPOOL_API_KEY

Create a key in the console and save its full value. You can copy it again when the list shows a Copy button. If the key is masked and cannot be copied, create a replacement and update your application. Disabled or deleted keys can no longer be copied.

Get started

1. Sign up and open the console.
2. Add credit on the balance page. Usage is billed at the published rate for the model and group you pick; your balance never expires.
3. Create an API key and copy the full value from the creation result or the key list.
4. Pick a model from the model list and call it with the examples below.

curl

APIPOOL_ENDPOINT="https://app.apipool.dev"

curl "$APIPOOL_ENDPOINT/v1/responses" \
  -H "Authorization: Bearer $APIPOOL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-5.6-luna","input":"Hello from APIPool"}'

OpenAI SDK (JavaScript)

import OpenAI from 'openai';

const APIPOOL_ENDPOINT = 'https://app.apipool.dev';

const client = new OpenAI({
  baseURL: `${APIPOOL_ENDPOINT}/v1`,
  apiKey: process.env.APIPOOL_API_KEY,
});

const response = await client.responses.create({
  model: 'gpt-5.6-luna',
  input: 'Hello from APIPool',
});

console.log(response.output_text);

OpenAI SDK (Python)

import os
from openai import OpenAI

APIPOOL_ENDPOINT = "https://app.apipool.dev"

client = OpenAI(
    base_url=f"{APIPOOL_ENDPOINT}/v1",
    api_key=os.environ["APIPOOL_API_KEY"],
)

response = client.responses.create(
    model="gpt-5.6-luna",
    input="Hello from APIPool",
)

print(response.output_text)

Plain HTTP (fetch)

const APIPOOL_ENDPOINT = 'https://app.apipool.dev';

const response = await fetch(`${APIPOOL_ENDPOINT}/v1/responses`, {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.APIPOOL_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'gpt-5.6-luna',
    input: 'Hello from APIPool',
  }),
});

console.log(await response.json());

Chat Completions

Official Group Qwen, GLM, and available Kimi text models use this endpoint. Use an Official Group key and select an available text model from the model list. This example does not apply to image models.

APIPOOL_ENDPOINT="https://app.apipool.dev"

curl "$APIPOOL_ENDPOINT/v1/chat/completions" \
  -H "Authorization: Bearer $APIPOOL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.8-flash","messages":[{"role":"user","content":"Reply only: APIPOOL-OK"}],"stream":false}'

Read a non-streaming answer from choices[0].message.content. For streaming, usage, and family-specific limits, see Qwen, GLM, and Kimi.

Claude and Anthropic Messages

Claude models in the Discount Group and Discount Group 2 currently require Claude Code. They use POST /v1/messages, but ordinary curl requests or Anthropic SDK clients do not gain access merely by using that endpoint.

Follow the Claude Code guide to configure the endpoint and key. Choose a complete model ID available in your key's group. Do not reuse an old model ID or substitute Claude into a Responses example for another model.

Image generation and editing

The listed image families use JSON /v1/images/generations. Models that support file uploads also accept /v1/images/edits (multipart/form-data). Wan edits use image_urls on the generation endpoint; Z-Image supports text-to-image only. See Image models and protocols for exact model coverage.

These image requests are asynchronous: submission returns 202 Accepted, then poll the task ID with the same key. The request format resembles Images, but the response is a task, not the synchronous OpenAI Images result. Do not read the submission as data[].url. Image endpoints do not support stream: true. For GPT Image 2 billing, see the Official Group guide or Discount Group guide.

curl "$APIPOOL_ENDPOINT/v1/images/generations" \
  -H "Authorization: Bearer $APIPOOL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-image-2","prompt":"A product photo on a white background","size":"1:1","resolution":"1k","n":1}'

Submission and polling are documented in full under Asynchronous image tasks.

Streaming

The Responses, Chat Completions, and Messages endpoints accept "stream": true and return an SSE stream. The image endpoints do not support streaming.

Error reference

Follow the steps below for these errors. Replacing your API key or adding credit is usually unnecessary:

Keep the X-APIPool-Request-Id response header or error.request_id when reporting an error. Once an image submission returns a task ID, query that same task. A disconnected poll does not mean submission failed; do not immediately resubmit.

Billing

Top up in dollars on the balance page. Text model usage is charged at the token rates shown in the model list; per-call models such as images are charged at the tier rate multiplied by the number of units actually returned. Requests, tokens, per-model spend, and per-request charges appear on the usage and billing pages.