Skip to content
← writing

Mume Gateway: one endpoint, 500+ models

· 1 min read · llm · api · infrastructure · mume

This post used to be a full API reference. It is not any more — that lives in the Mume docs, where it belongs and where it stays current. Keeping a second copy here meant two things that could disagree with each other, and the one on my blog was always going to be the stale one.

What is worth saying here is the short version.

The idea

Mume Gateway is one OpenAI-compatible endpoint in front of 500+ models across OpenAI, Anthropic, Google, Mistral and others.

“OpenAI-compatible” is doing real work in that sentence. It means you do not learn a new SDK, rewrite your request handling, or maintain a provider abstraction of your own. You change a base URL:

import openai

client = openai.OpenAI(
    api_key="your-mume-key",
    base_url="https://mume.ai/api/v1",
)

resp = client.chat.completions.create(
    model="anthropic/claude-sonnet-5",
    messages=[{"role": "user", "content": "Hello"}],
)

Change model to something from another provider and the rest of your code is untouched. Streaming, tool calling, images and MCP all work the way you already expect them to.

The part that is actually the product

Routing between providers is the commoditised half. Several projects do it well.

What sits behind this endpoint is a credit ledger, per-user quotas, API keys with their own limits, token-denominated rate limiting, usage analytics and billing. That is the half that decides whether an AI product has a business model rather than just a demo, and it is the half I spend my time on.

I wrote about why at length in Your LLM gateway routes. Mine bills., and about what happens when you skip it in What it costs you when one user writes an infinite loop.

Full reference, model list, error codes and API key setup: mume.ai/docs.