open beta
Your thinking has continuity,
now your tools do too
Scaffold is a local-first Mac app that studies what you give it: files, links, videos, conversations, repos, and more. It builds knowledge over time and follows you into every AI tool you use.
how it works
01
give it your world
Talk to it. Drop in PDFs, articles, videos, repos, voice memos. Connect GitHub, Linear, or Obsidian and your work flows in on its own.
02
it studies
Nothing is just saved. New information gets weighed against everything already known: duplicates merge, contradictions get resolved, corrections carry a date.
03
ask anywhere
What it learns lives on your machine, and every AI you use can draw on it. Ask in Scaffold, in Cursor, in Claude. Same understanding either way.
the graph
Every person, project, and idea it studies gets its own entry: a description, a dated history, and links to everything it touches.
Closer to a wiki than a chat log, except it writes and tends itself.
the cloud
Alongside the graph it keeps a second view, arranged by meaning instead of structure, so nearby things are related things.
Both are queried in parallel, so what it tells you is structurally sound and contextually rich at once.
what becomes possible
Here are this week's voice memos. What am I circling around?
How has my thinking on this shifted over the last six months?
Pick up where Cursor and I left off last month. What were we in the middle of, and why?
Watch my Linear board this month and tell me what's actually slipping, not just what looks busy.
I think this connects to something I saved months ago. Find it and show me the link.
what it learns, every AI knows
The model is not the mind. Scaffold keeps what it learns outside any single model or app, so Cursor knows what you told Claude, and next year's model picks up where this year's left off.
Connect any MCP client and it draws on the same knowledge. Switch models whenever you want. What it learned stays.
yours to keep
Scaffold runs on your Mac and stores everything it learns as files you own. Use cloud models for speed, or run fully local so nothing ever leaves your device. Either way, the knowledge stays on your machine.
export anything
Everything it learns lives in open files on your Mac. Export any part as a .scaffold bundle to share, archive, or move. Portable, readable, and yours. No lock-in and no platform holding your knowledge hostage.
learning, not logging
Most AI memory stops at storage. This is the difference.
Most AI memory is an append-only log. It stores what you said, retrieves what looks similar, and never asks whether the new thing contradicts the old one. The pile grows, retrieval gets muddier, and the system knows more while understanding less.
Scaffold treats this as a coherence problem. Before anything is written, an analyzer weighs it against existing knowledge and decides: store it new, merge it, update what's there with a dated correction, or reject it as a duplicate.
Knowledge is held in two forms, a graph for structure and vectors for meaning, queried in parallel and actively maintained as it grows. The result doesn't just know facts. It knows the history of its beliefs about them.
More on how it works, on the blog.
common questions
A few things people ask before they try it. Straight answers.