How can I use the DeepSeek API with UFO?

[Delo44]

How can I use the DeepSeek API with UFO? I tried modifying parts of the code to add support for the DeepSeek API, but I was only able to successfully run the first of the three sample tasks. The other two tasks failed to execute successfully.

[Munshi Mahidul Haque (Kxrma47)]

UFO already has a native DeepSeek service, so modifying the provider code should not be necessary. Configure the relevant entries in config/ufo/agents.yaml with API_TYPE set to deepseek, the desired DeepSeek model, and the API key, following the DeepSeek model page in the current UFO documentation. The important limitation is that the built-in DeepSeek integration is text-only; UFO's model overview explicitly marks DeepSeek as having no visual support. That explains why a simple first task can work while sample tasks that require interpreting a screenshot fail even though authentication and text completion are correct. Use a vision-capable provider such as GPT-4o, Gemini, Claude, or Qwen-VL for HOST_AGENT/APP_AGENT when the task requires screen understanding, and keep DeepSeek for a text-only role such as planning, backup, or code generation. If the failing samples are text-only, share the agent logs and agents.yaml entries with keys removed, because the next likely issue is JSON-output compatibility: the DeepSeek adapter requests a JSON object response, and failures should be distinguished from missing vision capability.