OpenAI API Compatibility — Boole AI Integration Guide

Boole AI implements the full OpenAI API spec. Swap your base URL and API key and every existing SDK, tool, and integration works instantly.

Boole AI implements the OpenAI API specification end-to-end. Any code, tool, or integration that already works with OpenAI works with Boole — change the base URL and API key, and you’re done. No new SDKs, no wrapper libraries, no refactoring required.

What’s Compatible

Boole AI supports the following endpoints and features from the OpenAI specification:

POST /v1/chat/completions

Chat completions with streaming, tool calls, and structured outputs.

POST /v1/completions

Legacy text completions for single-turn prompts.

GET /v1/models

List all models available on your account.

POST /v1/audio/transcriptions

Audio transcription powered by Whisper Large v3.

Beyond endpoints, the following API-level features are fully supported:

• Streaming via server-sent events (SSE) — set stream: true in any chat or completion request
• Tool / function calling — pass a tools array and handle tool_calls in the response
• JSON mode and structured outputs — set response_format: { type: "json_object" } or supply a JSON schema
• System prompts, temperature, top_p, max_tokens, and stop sequences — all standard sampling parameters work as documented

Configuration

Switch between the Boole Cloud API and a locally running binary by changing two values.

Cloud API

base_url = "https://api.boole.dev/v1"
api_key  = os.environ["BOOLE_API_KEY"]

Local Binary

base_url = "http://localhost:8000/v1"
api_key  = "any-non-empty-string"

The local binary does not validate API keys. Pass any non-empty string to satisfy the SDK’s requirement for an api_key argument.

Drop-In Example

The migration from OpenAI to Boole is a two-line change. Everything else — messages format, model parameters, response parsing — stays identical.

Before (OpenAI)

from openai import OpenAI

client = OpenAI(api_key="sk-...")

Streaming

Enable streaming by passing stream=True. The SDK surfaces each token as a delta chunk as Boole sends it — at up to 312 tokens/sec on Llama 3.3 70B, the response starts appearing almost immediately.

streaming.py

for chunk in client.chat.completions.create(
    model="llama-3.3-70b-instruct",
    messages=[{"role": "user", "content": "Tell me a story."}],
    stream=True,
):
    print(chunk.choices[0].delta.content or "", end="")

Tool Calls

Pass a tools array exactly as you would with the OpenAI SDK. Boole returns a tool_calls field in the response when the model decides to invoke a function.

tool_calls.py

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get current weather for a location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "City name"}
                },
                "required": ["location"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="llama-3.3-70b-instruct",
    messages=[{"role": "user", "content": "What's the weather in Paris?"}],
    tools=tools,
)

Model names in Boole differ from OpenAI’s. When migrating, update your model string — for example, replace "gpt-4" with "llama-3.3-70b-instruct". Run GET /v1/models or check the Models page for the full list of available model IDs.