Cherry Studio + JoinGonka Gateway — desktop AI client

Installation. Builds for every system are on the project's releases page: an installer and a portable build for Windows, a dmg image for macOS, AppImage, deb, and rpm for Linux — for both x64 and ARM. Chat history and settings are stored locally, on your machine.

V1 or V2. In August 2026 the V2 line launched with a new data structure and reworked settings; the last build of the previous line is 1.9.13. A custom provider exists in both, but the way you add it and some of the labels differ — those spots are flagged below. When moving from V1 to V2, provider and model settings carry over automatically, the original V1 data is preserved, but new data doesn't sync back, and backups from the two lines aren't compatible. Before upgrading, read the migration section in the Cherry Studio documentation.

JoinGonka key. Register on the gateway, open the "API Keys" section in your dashboard, and click "Create key". The key starts with jg- and is shown only once — save it right away. It's convenient to create a separate key for Cherry Studio: in the "Usage" section its spend will show up as its own line.

1. Click the gear icon to open the settings. Select the Model Provider section (referred to as Model Services in the English documentation).
2. Below the provider list, click Add Provider.
3. In V2, the Add Custom Provider dialog opens. Enter a Provider Name — for example, JoinGonka. In the Endpoint settings block, fill in the OpenAI line with the address https://gate.joingonka.ai/v1: a Request path line will appear below the field showing the final path — it must end with /v1/chat/completions. In the API Key field, paste jg-your-key and click Add.
4. In V1, the dialog is shorter: provider name and Provider Type — select OpenAI. The key and address are entered on the provider page itself, in the API Key and API Host fields.
5. Make sure the provider is enabled: the toggle is in the top-right corner of its page. In V2, new providers are added disabled, and models from a disabled provider do not appear in selection lists.

How Cherry Studio assembles the address. The API Host field takes the base address, and the app appends the path itself, showing the result in the Preview line below the field. The rules are the same in both lines:

Older instructions mention another rule: a slash at the end of the address supposedly disables /v1. It applied in early versions, up to and including the 1.6 line, and was removed in 1.7: in current builds, only # disables auto-substitution.

The provider is added, but it is not visible in the chat yet: Cherry Studio selection lists only include models you have added explicitly. There are two ways to do this.

List from server. On the provider page, click Sync models (in V1 — Manage, then Fetch model list). Cherry Studio will request GET /v1/models from the gateway and display the active network lineup; add the required models using the + button. Manually. The Add Model button opens a form where one field is required — Model ID: the identifier must be entered character for character.

The prices for the network models are the same, so add them all — you will have to choose between them based on behavior rather than budget:

Model types. Each model in the list has a settings button where you can mark its capabilities: Vision, Reasoning, Tool, and others. Cherry Studio sets some tags automatically, but you should verify them for gateway models. Vision should be unchecked: the network models are text-only. Tool is needed if you are connecting MCP tools — all models in the table support native tool calls. Reasoning is appropriate for GLM-5.3 Flash, as it is a network reasoning model.

Default model. In the Default Model section of the settings, you specify models for utility roles. Set DeepSeek V4 Flash as the main assistant model. For the fast model that invents dialogue titles and prepares search queries, Cherry Studio documentation recommends a light and fast model, not a reasoning one — out of the network models, MiniMax M2.7 fits this role best; do not use GLM-5.3 Flash here due to its mandatory reasoning. Any of the first two models will work for translation.

On the provider page in the key block, there is a Model Check button (in V1, it is Check next to the key field). Select a model — Cherry Studio will send it a short test request and show the result along with the response latency. If the check passes, go to the chat, select the gateway model in the model switcher, and ask any question.

To ensure that requests are indeed going through the gateway, open its dashboard, go to the "Usage" section: in the "By Keys" block, the Cherry Studio key will show requests and the time of the last access; in the "By Models" block, the selected model will appear. V2 also has its own client statistics in the Usage settings section: tokens and requests by providers, models, and keys. The cost there is for reference only — Cherry Studio estimates it based on the public prices of the models, not on the Gonka network price, so please check the gateway dashboard for exact amounts.

If something goes wrong, the reason is almost always clear from the response:

The client itself is free to use; you only pay for the inference. Through the JoinGonka Gateway, tokens cost $0.0069 per million for input and $0.021 per million for output — the price is the same for all models in the network and is pulled onto this page from a live source. Chat consumes tokens significantly more modestly than agent tools, so the approximate scale of costs as of September 2026 is as follows:

An important disclaimer about long dialogues: with each reply, the client sends the entire history back to the model, so the hundredth reply costs more than the first. The remedy is a new dialogue for a new task; in Cherry Studio, this is a single button.

Comparing models in a single chat. Cherry Studio can ask one question to multiple models at once: select them in the model switcher, and each will respond with a separate request. With a flat rate for all network models, this is the fastest way to understand which tasks are handled well by MiniMax M2.7, where you need a long response from DeepSeek V4 Flash, and where the reasoning of GLM-5.3 Flash pays off — its settings are broken down in the model review.

Second protocol with the same key. The gateway also responds in Anthropic format. In the V2 chat, there is a separate Anthropic line in the Endpoint settings block: enter https://gate.joingonka.ai there, and the Request path line should result in /v1/messages. In V1, the same result is achieved by a separate provider with Provider Type Anthropic and the same address. Under the More options button in V2, there is also an OpenAI Responses line — the gateway also services it, using the same address as the main OpenAI line.

Running terminal agents. On the Code CLI page (referenced as Coding Companion in the documentation), Cherry Studio installs and runs console agents, providing them with the provider and model from its settings; the list of providers is filtered by the protocol required by the specific tool. Therefore, the Anthropic address should be filled in advance if you intend to run Claude Code from here. How to set up the same agent without intermediaries is explained in the Claude Code + Gonka guide.

Knowledge Base. In V2, it works even without an embedding model — using BM25 text search, and with an embedding model, hybrid search is enabled. Gonka network models are generative, so for hybrid mode, use an embedding model from another provider, for example, a local one via Ollama; the gateway model will then answer based on the found fragments.

Need a shared chat for your team in a browser? Cherry Studio is a personal desktop client. For a shared web interface with accounts, check out Open WebUI: it connects to the gateway using the same address and key.