Witsy

An open-source desktop AI assistant and MCP client that runs on any OpenAI-compatible model.

What Witsy is

Witsy is an open-source desktop AI assistant for Windows, macOS and Linux. It sits in your menu bar, works as an MCP client, and lets you prompt AI from any app, all on your own API key. Through a custom engine, you can point Witsy at TokenMix and run Claude, GPT or DeepSeek.

Before you start

• Install Witsy from its GitHub releases. It runs on Windows, macOS and Linux.
• Sign up at tokenmix.ai, open the Console, go to API Keys, and create a key that starts with sk-tm-.

Connect TokenMix

1. Open Settings, then the Models tab.
2. At the bottom of the engine list, click Create to open Create Custom Engine.
3. API Specification: choose OpenAI.
4. Name: enter TokenMix.
5. API Base URL: enter https://api.tokenmix.ai/v1 and keep the /v1.
6. API Key: paste your sk-tm- key, then click Create.
7. Open the TokenMix engine and refresh its model list, then pick a model.

Check it works

Start a chat, select a TokenMix model, and send a short message. A reply means Witsy is connected to TokenMix. You can see the spend in the TokenMix Console.

Common questions

No models after saving: open the engine and refresh the model list so Witsy pulls the models from TokenMix.

No reply: confirm the API Base URL is https://api.tokenmix.ai/v1 with the /v1, and that your key is active in the TokenMix Console.

Which models to pick: any text model works. Claude Sonnet 4.6, GPT-5.4 and DeepSeek V4 Pro are good starting points.