Gateway

Overview

The Maniac Gateway is a multi-provider inference API. It speaks the OpenAI wire format, so unmodified openai / openai-node clients work against it — point the client's base URL at the gateway and authenticate with a Maniac API key. Behind that one surface it normalizes and routes requests to multiple upstream providers (OpenAI, Anthropic, OpenRouter, and more), with per-organization rate limiting, a billing pre-flight gate, and one telemetry event per request.

https://gateway.maniac.ai

What you get

• GET /v1/models — the model catalog, with Maniac extensions on top of the OpenAI base fields: pricing, context window, modalities, and supported parameters/APIs. See Models.
• POST /v1/chat/completions — Chat Completions, streaming and non-streaming, across providers. See Chat Completions.
• POST /v1/responses — the OpenAI Responses API: streaming event taxonomy, stateful storage (store, previous_response_id chaining), and background async runs with polling, reconnect, and cancel. See Responses.

The full machine-readable contract is published at /openapi.json, and the interactive API Reference is generated directly from it.

OpenAI compatibility

The gateway is a drop-in OpenAI endpoint. Maniac-specific fields (tags, metadata, trace, and a provider field on responses) are added alongside the OpenAI fields, so SDKs ignore what they don't recognize.

For OpenAI parameters with no internal analogue the gateway uses a map-or-reject policy: n > 1 is rejected; logprobs / top_logprobs, input_audio content parts, and json_schema strictness are dropped (the request still runs); unknown top-level params pass through. The specifics are called out on each endpoint's guide.

Start with the Quickstart.