Build a research-notes workspace (bookmarks + annotations)
Treat VizChat as your second brain — collect web pages, parse PDFs, annotate, and let the Agent synthesize your reading.
Deep research has three pain points: collecting (you spot a useful link but can't read it now), understanding (you've piled up PDFs you'll never finish), and connecting (scattered facts don't become a viewpoint). VizChat's Workspace unifies all three on one canvas.
Workflow at a glance
- Collect: pin web pages with web-viewer widgets, which keep a reader-mode copy of the article text; upload PDFs into the file tree.
- Understand: let the Agent summarize each source into a rich-text widget.
- Connect: drag related widgets closer on the canvas, or ask the Agent "what assumption do these three share?"
Steps
- Create a Workspace, e.g.
research-few-shot-learning-2025. - Clip a page: paste a URL into chat and say "pin this page to the canvas" — the Agent fetches the article text and creates a web-viewer widget.
- Upload PDFs: upload your PDFs into the file tree; OCR parsing runs in the background (see Supported file formats).
- Ask the Agent for cards: "For each
paper-*.pdf, produce a 400-word rich-text widget with TL;DR / Method / Limitations sections." You get a uniform batch of notes easy to skim. - Add your own commentary: put it in a separate rich-text widget below the note it belongs to — that keeps your voice distinct from the source's.
- Synthesize: hold
Cmd/Ctrland click each widget to add it to the selection (or drag across an empty part of the canvas to rubber-band select a cluster of them), then ask "Across the widgets I've selected, summarize the current SOTA and the gaps." The Agent reads across them and produces a synthesis widget. - Ship the output: create a new storyboard document in the same workspace and have the Agent move the conclusions into it for a reading-group talk. For archival, export the board as a
.vizpkgpackage.
Organization tips
| Need | How |
|---|---|
| Group by topic | Cluster related widgets spatially on the canvas, and put the topic in each widget's title |
| Mark read / unread | Emoji or [x] read tags at the top of each rich-text widget |
Caveats
web-viewershows the page's extracted article text, not its original layout. Pages whose text can't be extracted — paywalled articles, JS-only apps, login-gated pages — come back empty, so check the widget right after creating it.- OCR consumes credits (see Plan limits) — pilot on a small set before batching.
Next step
Build an academic presentation (with PDF and formulas)
Turn a paper into a presentable Storyboard in VizChat — with the original PDF pinned on the canvas, LaTeX-rendered equations, and a TTS-narrated walkthrough.
Build a custom widget
Three ways to get a widget that doesn't exist yet — ask the agent to build one, author a built-in widget in the repo, or develop and publish one yourself with the CLI.