Yichuan Wang
6e80426e25
The agent model was not reliably using uploaded images for search because
the tool description said "OR" (text or image) and the system prompt was
too soft ("you can add a text query"). Testing showed that image-only
search gives the best visual match (score 0.50 vs 0.41 for combined),
while vague text like "who is this" actively dilutes the visual signal.
Updated the system prompt with a nuanced strategy: image-only first for
identification questions, descriptive keywords for specific queries, and
never pass vague questions as text. Also added a "Connect your own agent"
section to the API docs page showing text, image, and joint search curls.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Next.js template
This is a Next.js template with shadcn/ui.
Adding components
To add components to your app, run the following command:
npx shadcn@latest add button
This will place the ui components in the components directory.
Using components
To use the components in your app, import them as follows:
import { Button } from "@/components/ui/button";