Works on
Models Velociti works with
Velociti runs compatible local models on your computer and connects supported providers with keys you already control. The names below are the popular lines on the public desk. The project stays when you change intelligence.
ChatGPT
Supported provider
OpenAI's chat product. In Velociti it enters through a key you control, for a turn you send out.
ChatGPT is a hosted chat. The thread in ChatGPT lives in an OpenAI account.
In Velociti ChatGPT is a supported provider, not the desk. You bring the key. The project stays on your computer. See Velociti vs ChatGPT.
Claude
Supported provider
Anthropic's models. Connect with a key you already control. Switch to Claude without reloading the project.
Claude is a hosted model family. Use it when the turn needs that intelligence.
The files, instructions, and approved memory stay attached. Switch models without losing the project.
Grok
Supported provider
xAI's Grok. Available as a supported provider when you send a turn out with your key.
Grok is hosted intelligence. Velociti does not move your project into a Grok account.
Local drafting can stay on your machine. Send the harder turn to Grok when you choose.
Kimi
Local or provider
Moonshot's Kimi. Use it in the picker when it is the right intelligence for the turn.
Kimi is one of the popular models Velociti is built to sit beside.
The campaign file does not reload when you move from Codex to Kimi K3 in the desk.
MiniMax
Local or provider
MiniMax models in the same workspace as the rest of your picker.
MiniMax is another supported line in the picker. The project object does not change when you select it.
Gemini
Supported provider
Google Gemini as a supported provider. Your key. Your project stays local.
Gemini is Google's hosted model family. In Velociti it is a keyed provider, not the system of record.
Work stays on your computer unless you send that turn out.
DeepSeek
Local or provider
DeepSeek can run as compatible local GGUF or through a supported provider path.
DeepSeek is widely run locally. Run local models on your computer covers RAM and VRAM placement.
A provider path still uses a key you control.
Qwen
Local or provider
Qwen is a common local GGUF family and a supported provider line.
Qwen weights are a typical local install. Velociti measures hardware, then adds a compatible build to your picker.
The workspace around Qwen is the same workspace around ChatGPT: the project stays.