Configure the Model

AutoLXB needs an LLM / VLM endpoint that supports text and image input. Visual execution and semantic route adaptation require image understanding.

Entry

Open: Config -> Device-side LLM Config

Fill in:

• Request Type: the request protocol shape, such as OpenAI Chat Completions, Gemini generateContent, or Anthropic Messages.
• API Base URL
• API Key
• Model

The config page shows the final resolved request URL based on the selected Request Type. API Base URL is still the base address, for example:

• OpenAI Chat Completions: https://api.openai.com/v1
• Gemini generateContent: https://generativelanguage.googleapis.com/v1beta
• Anthropic Messages: https://api.anthropic.com/v1

Test the model

After filling in the fields, tap the test button.

The test sends a small image challenge and asks the model to read the digits in the image. A successful test means:

• The app can reach the model service.
• The API key and model name are usable.
• The model can process image input.

Choose a model with image understanding

If the model only supports text, a basic connection test may pass, but visual execution, screen observation, and semantic adaptation will not work properly.