OpenAI compatibility

Odyssey speaks the OpenAI API. If you already call OpenAI through an official SDK, the base URL and the key are the only things that change, and the model id now names the model's author as well as the model.

Change one line

import OpenAI from "openai"; const client = new OpenAI({  baseURL: "https://odysseyapi.tech/v1",  apiKey: process.env.ODYSSEY_API_KEY,});

What carries over

• POST /chat/completions, including stream, stop, temperature, top_p, max_tokens and seed.
• Tool calling with tools and tool_choice, and the tool_calls response shape.
• JSON mode through response_format, on models with the json capability.
• Image parts in messages, on models with the vision capability.
• POST /responses, including streaming and function tools. See Responses and images.
• POST /images/generations on image models, plus POST /images/edits on the ones flagged for editing.
• GET /models, returning the catalog the calling key may use.
• Server-sent events framed exactly as the SDKs expect, ending with [DONE].

What differs

• Model ids are namespaced. Use openai/gpt-5 rather than gpt-5, and any model from another author the same way.
• Organization headers are ignored. Billing follows the key, not an OpenAI-Organization header.
• Not implemented: embeddings, audio, batches, files, image edits and the assistants API.

Frameworks and tools

Anything that accepts an OpenAI-compatible base URL works: LangChain, LlamaIndex, the Vercel AI SDK's OpenAI-compatible provider, and most self-hosted chat interfaces.

from langchain_openai import ChatOpenAI llm = ChatOpenAI(    model="anthropic/claude-sonnet-5",    base_url="https://odysseyapi.tech/v1",    api_key=os.environ["ODYSSEY_API_KEY"],)