Which APIs and models are compatible with miku.gg?

Updated

miku.gg is designed to work with any OpenAI-like API, so you are not locked into a single model provider. According to the official documentation, the currently listed examples are Aphrodite, TextGen WebUI (ooba), and Tabby. If a backend exposes an OpenAI-compatible endpoint, it is a candidate for use with miku.gg.

What "OpenAI-like API" means here

The key compatibility requirement is not a specific model name but the API shape. miku.gg expects an interface that follows the OpenAI API convention, which typically means:

  • A chat or completion endpoint that accepts prompts and returns generated text.
  • A request/response structure compatible with OpenAI-style clients.
  • A base URL you can point miku.gg at, rather than a hardcoded provider.

Because of this, the model behind the API can vary. It could be a hosted service, a locally running model server, or a self-hosted inference engine, as long as the endpoint behaves like OpenAI's.

Currently listed compatible backends

The documentation names three examples at this time:

Backend What it is Why it fits
Aphrodite An inference engine for serving language models Exposes an OpenAI-like API
TextGen WebUI (ooba) A popular local text generation web UI Provides an OpenAI-compatible API mode
Tabby A self-hosted coding assistant / model server Offers an OpenAI-like interface

These are examples, not an exhaustive list. The broader rule is that any OpenAI-like API can be used.

How to connect a backend

The general flow is:

  1. Run or access a backend that exposes an OpenAI-like API (for example, one of the three above).
  2. Note the base URL and any API key or token the backend requires.
  3. Point miku.gg's model configuration at that base URL and credentials.
  4. Select or specify the model name the backend serves.
  5. Send a test message and confirm the bot returns generated text.

The expected result is that miku.gg routes your chat input to that backend and displays the returned output, with no change needed to miku.gg's core code.

Self-hosted and local model support

Yes, self-hosted and local models are supported in practice, because the compatibility layer is the API, not the hosting location. If you run a model locally through an OpenAI-compatible server, miku.gg can use it the same way it uses a remote endpoint. This is consistent with miku.gg being open source and runnable on your own machine.

Practical notes and common snags

  • API shape matters more than model brand. A backend that does not expose an OpenAI-like interface will not work without an adapter.
  • Base URL and model name must match the backend. Mismatches here are a frequent cause of failed requests.
  • Authentication varies. Some local backends need no key; others expect a token. Use whatever the backend requires.
  • The listed backends may change. Treat Aphrodite, TextGen WebUI, and Tabby as current examples rather than a fixed compatibility guarantee.

For the most current list and configuration details, check the miku.gg documentation directly, since supported backends can be added or updated over time.