Docs

AI Provider is a private AI you reach with a key. You send a message, you get a reply, and we never store either one. It speaks the same language as OpenAI's API, so most AI apps and code libraries already know how to talk to it.

Use it in an app

Many apps let you bring your own AI key: chat apps, writing tools, story and character apps, automation tools and coding assistants. Open the app's settings and look for something like "OpenAI-compatible", "custom endpoint" or "base URL". Then fill in three boxes:

Address (base URL)

https://api.aiprovider.org/v1Copy

API key

Your key, which starts with ap_. Make one if you need it.

Model

aiprovider-1Copy

Where it goes in each kind of app

Good to know: the AI thinks briefly before it answers. Most apps hide that thinking and show only the answer, which is what you want. If an app asks for a list of models, it finds aiprovider-1 on its own.

Quick start for code

If your code already uses the OpenAI format, change the address, the key and the model name. That's it.

1. Get a key. It shows on screen right away and starts with ap_.
2. Save it as an environment variable (a setting your program reads), for example AIPROVIDER_API_KEY.
3. Send a request to https://api.aiprovider.org/v1/chat/completions with the model aiprovider-1.

# pip install openai
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.aiprovider.org/v1",
    api_key=os.environ["AIPROVIDER_API_KEY"],
)

reply = client.chat.completions.create(
    model="aiprovider-1",
    messages=[{"role": "user", "content": "Write a two-line welcome for our help desk."}],
)
print(reply.choices[0].message.content)

You get back a normal chat completion:

{
  "id": "chatcmpl-8kq3...",
  "object": "chat.completion",
  "model": "aiprovider-1",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "reasoning_content": "They want something short and warm...",
      "content": "Hi, you've reached the help desk.\nTell us what's going on and we'll sort it out."
    },
    "finish_reason": "stop"
  }],
  "usage": { "prompt_tokens": 22, "completion_tokens": 61, "total_tokens": 83 }
}

Reasoning and replies

The model thinks for a moment before it answers. Replies include a short reasoning step, billed as output tokens. It comes back in its own field, so it never mixes with the answer:

• message.content is the answer. Show this to your users.
• message.reasoning_content is the reasoning step. Show it, keep it on your side, or ignore it.

If a reply runs out of room (it reaches max_tokens) while the model is still thinking, everything so far is in reasoning_content, content is empty, and finish_reason is length. Give it more room and try again. We suggest at least 1,000 tokens for short answers.

Streaming

Set "stream": true to get the reply piece by piece as it's written, using server-sent events (a simple standard for streaming text over HTTP). The reasoning arrives first in delta.reasoning_content, then the answer in delta.content.

stream = client.chat.completions.create(
    model="aiprovider-1",
    messages=[{"role": "user", "content": "Explain DNS in three sentences."}],
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta
    # The short reasoning step arrives first (delta.reasoning_content), then the answer.
    if delta.content:
        print(delta.content, end="", flush=True)

Add "stream_options": {"include_usage": true} to get a final chunk with token counts. The stream ends with data: [DONE].

Tool calling

Tool calling lets the AI ask your program to do something, like look up the weather or search your files, and then use the answer. Send tools the usual way and read tool_calls from the reply. Streaming works too.

Read the tool calling guide

Parameters

We accept these fields on POST /v1/chat/completions. Others are ignored.

Errors

Errors use the usual shape, so the official libraries raise the right exception. The message says what went wrong and what to do.

{
  "error": {
    "message": "You're out of credit. Add credit at https://aiprovider.org/dashboard/billing and try again.",
    "type": "insufficient_quota",
    "param": null,
    "code": "insufficient_quota"
  }
}

For 429 and 503, retry with a growing wait: 1 second, then 2, then 4, or the number in Retry-After. The official libraries do this for you.

Limits

Limits depend on your plan. "At once" means requests still running at the same moment.

• Context: up to 1,048,576 tokens per request. We estimate about 4 characters per token.
• Request size: up to about 4.5 MB of JSON, roughly 1 million tokens of plain text. That is enough to fill the whole context window.
• Reply: up to 131,072 tokens, reasoning included.
• Time: a request can run for up to about 13 minutes. Very long replies may stop there; stream them and keep what arrives.
• Busy times: your request may wait in line for up to 2 minutes before it starts. Streaming requests stay connected while they wait.

Your balance

GET /v1/account tells you how much credit is left, what you've spent this month, your plan and your limits. Use the same key. Amounts are in US dollars.

curl https://api.aiprovider.org/v1/account \
  -H "Authorization: Bearer $AIPROVIDER_API_KEY"

The same numbers, with charts, are in your dashboard.

Not supported yet

• Images, audio and files: messages must be text.
• Embeddings and fine-tuning: only chat completions for now.
• More than one reply per request (n above 1).

Need one of these? Tell us at the support address in the footer. It helps us decide what to build next.