Snippet everything you know.
One knowledge base for you and every model you work with. Every note carries where it came from, so a claim can be traced back years later.
A note is small, and it remembers its source
A snippet is one idea in Markdown — a sentence worth keeping, a finding, a quote. It is small on purpose: small enough to tag precisely, cite in several places, and still mean something on its own a year later.
What makes it a snippet rather than a note is the citation. Each one can name the source it came from and the exact locator inside it — page 32, 1:23:45. Nothing is stranded without a provenance.
Sources are first-class, not a text field
A source is its own record, shared by every snippet that cites it. It has a type, ordered authors, a publisher, a year, and the fields its type actually has — a book has an edition, a lecture has a date and a location, a video has a running time.
Because authors are records too, a person is a way through the library: open one and you see every source they wrote and every snippet taken from it. The citation for any snippet is built from those fields, not typed by hand.
Book, paper, lecture
The bibliographic types, with the fields each one needs and nothing it doesn't.
Video, podcast, article
Time locators instead of pages, so a claim points at the minute it was made.
Meeting
Date, location and attendees. The attendees are authors, so they reach the same views as a book's.
AI chat
A conversation with an assistant is a source like any other. Paste the link and the type fills itself in.
Tags are how the library is organised
There are no folders. A snippet carries as many tags as it deserves, and the tags are what you browse: a cloud sized by how much each one holds, narrowing as you pick. Pick two tags and see only what carries both; exclude one and it drops out.
Tags also co-occur, which is the part a folder tree cannot do. Picking memory shows you that it keeps turning up beside interface design — a connection you did not file and would not have looked for.
Sources, authors and publishers work the same way. Each is a dimension you can filter by, and they combine: everything tagged cognition, from lectures, by one author.
Tags set the context for AI work
This is the one that changes how the rest is worth keeping. Your snippets are reachable over MCP, so any AI client that speaks it reads from the same knowledge base you do, over a connection you authorise and can revoke.
Tags are the selector. Instead of pasting background into every new chat, you name the tags that matter and the model pulls exactly those snippets into its context — with their sources attached, so it can cite what it used and you can check it.
- Tag what belongs to a piece of work as you capture it.
- Connect your AI client to Snippets once.
- Ask for the work. It searches those tags and loads what it finds — no re-briefing.
- What it writes back is a snippet too, in the same library, tagged the same way.
Posts: publish straight from your snippets
A post is long-form writing composed from snippets you already have. You cite them inline, the way a paper cites its references, and the post carries those citations with it — so a reader can follow any sentence back to the note it came from and the source under that.
Publish and it becomes a real page: server-rendered, with its own address, social tags and a line in the sitemap. Your public knowledge base is the subset you chose to publish, and nothing else. Everything is private until you say otherwise.
Each post also draws its own citation graph — post, snippets and sources as one picture of what the argument rests on.
What else is in there
Search that knows the grammar
Full-text across snippets, tags and sources, with tag: and -tag: filters.
Diagrams from text
Fenced blocks compile to flowcharts — a recipe as steps, a story as cause and effect, a question as an issue tree.
Attachments
Attach a file to a snippet. Text is extracted from a PDF so it is searchable with everything else.
Surfaced connections
Snippets that look related are offered for review. You accept the link or dismiss it; nothing is filed behind your back.
Export
Markdown out whenever you want it, and flashcard decks for what you are trying to remember.
Live everywhere
A snippet captured anywhere turns up in an already-open dashboard on its own.
Capture from where you already are
The web app is the full surface — capture, browse, filter, compose and publish. The terminal client is the fast one: a prompt where a line of text is a snippet, with the source and the page carried over from the last one.
You can also mail a snippet in to your own ingest address, or let an AI client write one over MCP. Native apps for iOS and macOS and a browser clipper are in development.
One account covers it, with a password or a passkey. The web app also runs without an account at all — everything stays in your browser until you sign in, and then it moves across.
Snippet everything you know.
Tasks everything you do.
Tasks is the other half: the same tags, the same clients, the same knowledge base — for the things you have to do rather than the things you know.
A Bang Labs product.