Potion collects the videos, articles and ideas you find, turns each one into a note you can describe, filter and link, and hands the whole thing to your AI.
Most notes are not typed, they are collected. Find a video, an article or a thread, and the Clipper, in Chrome or in any iOS share sheet, saves it as a note: the link, a playable embed or the article as clean text.
Tag it as you save, and it lands in your feed next to the notes you write yourself.
Every note has a title and markdown content that renders embeds, math, diagrams and live pages. Around it, the note carries what describes it.
Tags say which sets it belongs to. Properties give it typed values: text, numbers, dates, a choice from a list, even free-form JSON. Files attach documents you can search. Presets fill all of it in the moment a note is saved.
Because notes are objects, they can be filtered. A view keeps a filter and a sort: films rated 8 or more, articles saved this month, anything tagged rust or go that is not archived.
It runs again every time you open it, so it always shows what matches now.
Write [[another note]] and two notes are linked. Every note knows what it points to and what points back.
The graph draws those links as a map of what you know, and a lens saves the way you like to look at it: your film notes with their tags, or everything you linked to but never wrote.
Describe a chart, a table or a dashboard, and your AI writes it as an html block that lives inside a note. It can use a curated set of libraries, and read your own notes through potion.
Like a design? Make it a layout, and every film note can render through the same card.
// inside the block
const books = await potion.getNotes({
filter: { match: 'all', children: [
{ type: 'tags', op: 'has_all', tags: ['book'] } ] },
fields: ['read_on'],
});Potion speaks the Model Context Protocol, so the assistant you already use connects to it: it does what you do, in your words — saves and writes notes, tags them, builds views and lenses, reads your files, designs blocks.
Any MCP client works: Claude, Cursor, VS Code, Zed, and whatever comes next. It works as you, in one workspace, under the same rules that protect your data.
Save the sources: the paper, the talk, the thread. Your AI reads them and keeps the pages, one per idea rather than one per source, each linked to what you already know. A new source updates the page it adds to instead of piling up beside it, and an index at the top keeps the whole thing findable.
Then it thinks with you. Hand it one idea and it follows the links through everything you have gathered, and expands on it from there: a co-pilot that starts from your own notes, not from a blank page. The gaps show too; a page that is linked but never written is where to look next.
“Attention”, from the paper, linked to 3 pages.
“Why attention scales”, drafted from 6 linked notes.
“Scaling laws”: linked twice, never written.
Write, build views, explore the graph.
Your notes and views in your pocket.
Save from Chrome and any iOS share sheet.
Any AI assistant, working in your workspace.
| Free | Pro $5/mo | |
|---|---|---|
| Notes | 1,000 | Unlimited |
| Views | 5 | Unlimited |
| Workspaces | 1 | Unlimited |
| Image storage | 100 MB | 10 GB |
| Web app, iOS, Chrome extension | ✓ | ✓ |
| AI connector (MCP) | — | ✓ |