Skip to main content

RikkaHub with OpenCheese

/ API docs

← API quickstartView as Markdown

RikkaHub is an Android chat app that works with OpenCheese through either of its two provider types: OpenAI, for every model, or Claude, the Anthropic format for Claude models.

Add the provider

Open the sidebar, tap Settings, then Providers under Models & Services, and tap +.

For the OpenAI type:

Field Value
Name OpenCheese (any name)
API Key your OpenCheese API key
API Base Url https://api.opencheese.dev/v1
API Path /chat/completions (the default)
Response API off; or on, with the assistant's Stream Output on (see below)

For the Claude type:

Field Value
Name OpenCheese Claude (any name)
API Key your OpenCheese API key
API Base Url https://api.opencheese.dev/v1

Both types keep /v1 in the base URL: RikkaHub appends /chat/completions, /responses, /messages or /models itself. Do not add a trailing slash.

Save, open the Models tab and pick models from the list RikkaHub fetches, then use the plug icon's Test Connection to check them.

To see the same list from a computer, and check the key at the same time:

bash
curl -s https://api.opencheese.dev/v1/models \
  -H "Authorization: Bearer $OPENCHEESE_API_KEY" | jq -r '.data[].id'

Set the Fast Model

RikkaHub names conversations and suggests replies with its Fast Model. On a fresh install it points at no real model, and RikkaHub shows "Fast model not found. Please select one in settings." before sending anything. Open Settings → Default Model and Prompt, and under Fast Model choose an OpenCheese text model; a cheap or free one is enough. Turn suggestions off if their cost per turn matters to you.

Reasoning levels

The reasoning level picked in a chat is sent with the request. Choose one the model lists in supported_reasoning_levels (see Models), or Auto for the model's default. A level the model doesn't offer is refused with a message naming the levels it does.

Image models

RikkaHub lists image models in the chat picker too, and in a chat they answer with a picture:

  • Add the image model with type Chat and input modalities Text + Image. RikkaHub does not know these model IDs and treats them as text-only otherwise, turning every picture you send into text, so edits become new pictures.
  • On the Response API, keep the assistant's Stream Output setting on: RikkaHub shows a non-streamed image answer as an error.
  • Chat Completions mode works with streaming on or off.
  • For RikkaHub's image page, add the same model again with type Image and an image count of 1. Up to 5 reference images are accepted per edit.

A background call such as a title or a suggestion aimed at an image model is refused with a message naming the Fast Model setting, so it never buys a picture by mistake.

Usage

Each RikkaHub conversation, with its titles, suggestions and summaries, appears in the dashboard's Transactions as one session. Pictures from the image page are a session each.

The cheese has holes. Your infrastructure should not.