wtf is deep-research-canvas?

Deep Research Canvas turns a research question into a visual board of connected cards instead of a wall of text. You type a question, and the app breaks it into sub questions, then lays out the findings as draggable cards on a canvas, with lines showing how each smaller finding feeds into the overall answer. You can pan around, zoom in, click on any card, and ask follow up questions about just that piece in a side chat panel. The interesting part is that you can switch which source powers the answers. There are three providers: gpt and claude answer from what the model already knows and draw their own charts with no outside sources, while a third provider called tako pulls in real, cited data cards from a service called Tako, complete with real numbers and an as of date showing when the data was current. Switching between these providers is meant to show the difference plainly: boards built with model memory alone look thin and get flagged with a confidence label, while boards built from tako are dense with citations. The final verdict or ranking is worked out by plain code applying the criteria you stated, not by the AI just guessing. To run it, you clone the project, install with npm, copy an example environment file, add your own API keys for OpenAI, Anthropic, and Tako depending on which providers you want to use, then start it locally with npm run dev. All of the keys stay on the server side and are never sent to the browser. It is built with Next.js and the Vercel AI SDK. The project is small and new, with a full architecture writeup kept in a separate docs file, and is released under the MIT license. It is not meant to give investment advice, even though the demo example in the README is about ranking semiconductor companies.