One key. Every model. Built for agents.
Route to every major provider through a single OpenAI-compatible endpoint — with automatic fallbacks, unified billing and observability, and a Responses API that works everywhere. Bring your own keys when you want them; let us manage the credentials for agentic tools when you don't.
from openai import OpenAI
client = OpenAI(
base_url="https://gateway.maniac.ai/v1",
api_key="mnc_...",
)
resp = client.chat.completions.create(
model="openai/gpt-4o-mini", // any provider slug
messages=[{"role": "user", "content": "Why is the sky blue?"}],
)Routing, billing, and observability in one place
Use a single API key and dashboard to access models, track spend, and keep workloads resilient.
One key, every provider
Access OpenAI, Anthropic, xAI, Google, OpenRouter and more through one OpenAI-compatible endpoint — with unified billing and observability across your whole stack.
Built-in fallbacks
Automatic failover across providers, so a model outage or rate limit doesn't take your app down. Define a chain; the gateway handles the rest.
Unified observability
Every request logged with spend, latency, and traces in one dashboard — no per-provider consoles to stitch together.
The Responses API, on every model
A Responses-compatible endpoint for stateful, agentic loops — streaming events and background mode included — that works across every provider, not just one. Build once, then switch models with a string.
{
"model": "openai/gpt-4o", // or any provider slug
"input": "Plan next week and book the meetings.",
"tools": [{ "type": "web_search" }, { "type": "mcp" }],
"stream": true, "background": true
}Tools, with the keys handled
Managed web search, code execution, memory, and MCP servers — credentials provisioned and rotated for you. Your agents call tools through the gateway without you managing a single tool-provider key.
credentials provisioned & rotated for you — no tool keys to manage
Your keys, or ours
Run on platform keys by default, or bring your own provider keys per org or project when you need your own rate limits, data terms, or pricing. Same endpoint, same observability, no code changes.
OpenAIAnthropicxAIGoogleOpenRouterOne key for every model
Point your existing OpenAI client at the gateway and start routing in minutes.