Cherry Studio

Add a custom provider in Cherry Studio and reach the OpenAI, Anthropic and Gemini protocols with one key

Cherry Studio is an open-source desktop AI client for Windows, macOS and Linux. These steps were checked against the interface of Cherry Studio v2.2.2 (released 2026-11-13) and the official docs. Older versions had a "provider type" dropdown; this version asks for an address per protocol instead.

Get an API key#

Create one at qd.zsopc.com/dashboard/keys (sk-gpushare- plus 64 hex characters; you can re-reveal it on its detail page at any time).

Add the provider#

  1. Settings in the left bar → Model Provider → + Add Provider at the bottom of the list, which opens "Add Custom Provider"
  2. Fill in:
    • Provider Name: anything, e.g. xiaozhan
    • API Key: your sk-gpushare-...
    • Endpoint settings → OpenAI: https://qd.zsopc.com
    • Optional: Anthropic with https://qd.zsopc.com too; under More options there are also OpenAI Responses and Gemini, again https://qd.zsopc.com
  3. Click Add, then turn on the enable switch at the top right of the provider page (while it's off, the models don't appear in the selector)

Each address field previews the final request path below it. Cherry Studio adds the API version itself (/v1 for OpenAI, Anthropic and Responses, /v1beta for Gemini), so the bare domain is all you enter:

FieldEnterFinal request
OpenAIhttps://qd.zsopc.comhttps://qd.zsopc.com/v1/chat/completions
Anthropichttps://qd.zsopc.comhttps://qd.zsopc.com/v1/messages
OpenAI Responseshttps://qd.zsopc.comhttps://qd.zsopc.com/v1/responses
Geminihttps://qd.zsopc.comhttps://qd.zsopc.com/v1beta/models/{model}:generateContent

Do not end an address with #: it means "don't add the version", and the request loses its /v1.

Add models#

  • Click Sync models to pull the list from GET /v1/models, or the + next to it to enter a model ID by hand (such as claude-sonnet-4-6, gpt-6.1-sol, gemini-3.1-pro-low or deepseek-v4.1-flash)
  • The "Model management" dialog that sync opens ticks every new model by default. Image and video models don't work in a chat, so untick them. Full list: Models
  • Then click Model Check to test the connection; it uses the chat model you added last

Which protocol each model uses#

By default every model goes over the field marked "Default" (the OpenAI field unless you changed it). The OpenAI field alone reaches every text model — Claude and Gemini included, because the gateway translates the protocol for you, see the compatibility matrix.

To send Claude over the native Anthropic protocol, fill in the Anthropic field too, then in the model's settings set Model purpose → Chat → Chat protocol to Anthropic. That option only appears once more than one protocol is configured. Gemini works the same way.

Cherry Studio can also connect MCP servers. To generate images, video and music from a chat, connect this platform's MCP media tools.

Troubleshooting#

SymptomWhat to check
401 invalid_api_keyWhether the key was copied in full (76 characters); re-reveal it on the key detail page
404Check the request path previewed under the address field: no trailing # and no doubled /v1
The provider is missing from the model selectorThe enable switch at the top right of the provider page is off
400 model_not_foundModel ID spelling — it must match Models exactly
400 model_not_allowedThis key has a model allowlist that excludes the model
402 quota_exceededBalance exhausted. Every key shares one account balance — top up on the Top up page

Full error reference: Error codes.

Other clients#