OpenAI-compatible API

The LLM API follows the OpenAI wire format. Point any OpenAI SDK or compatible tool at the BreachLine base URL and authenticate with a BreachLine key.

Setting	Value
Base URL	https://api.breachline.io/api/v1/llm/v1
API key	Your bl_live_ key
Model	nebula-4.5 (see Models)

OpenAI SDK configurationCopy

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ["BREACHLINE_API_KEY"], base_url="https://api.breachline.io/api/v1/llm/v1")

Endpoints

POST /api/v1/llm/v1/chat/completions

Chat completions with tools, structured output and streaming. Guide

Scope: llm:access or llm:chat

GET /api/v1/llm/v1/models

Available models with prices. Guide

Scope: llm:access

POST /api/v1/llm/v1/responses

The Responses API, with stored threads. Guide

Scope: llm:access or llm:chat

POST /api/v1/llm/v1/embeddings

768-dimensional embeddings. Guide

Scope: llm:access or llm:embeddings

POST /api/v1/llm/v1/rerank

Cohere-compatible reranking. Guide

Scope: llm:access

POST /api/v1/llm/v1/batches

Asynchronous bulk chat completions. Guide

Scope: llm:batch, llm:chat or llm:access

Differences from OpenAI

• Models. Only Nebula models exist. Unknown and vendor ids resolve to nebula-4.5.
• Errors. Errors use the BreachLine envelope (error.status, error.code, error.detail), not OpenAI’s. SDKs still raise the right exception class from the HTTP status. See Errors.
• Idempotency. Idempotency-Key refuses a replay with 409 instead of returning the original response. See Idempotency.
• Responses API. Hosted tools (web search, file search, code interpreter) are not supported and are dropped; function tools work.
• Embeddings. Float vectors with 768 dimensions only, with tight per-request size limits.
• Endpoints. Only the endpoints in the API reference are available. The legacy /v1/completions endpoint does not exist.

Tip

Send a stable user value (a session or end-user id) on every turn of a conversation. Requests that share a value let prompt caching reuse earlier context.