At this point, its the users

ContextCanvas(https://github.com/milind-soni/contextcanvas) I'm working on a visual interface for managing AI context. Instead of wrestling with text, you draw selections on an embedding map and chat with the selected data.

The insight: points close together in embedding space are semantically similar. So lasso a cluster, ask questions about it, refine your selection, repeat.

Built with Svelte 5 and DuckDB WASM for fast spatial queries. Everything runs client-side—your data never leaves the browser. It is based on Apple Embeddings Atlas

The vision is turning these visual selections into API endpoints that agents can query programmatically.

Looking for suggestions and feedback!