Weave turns a messy problem into a picture you can interrogate. Break it into the ideas it rests on: every node has a kind — problem, hypothesis, solution, gap, seed, synthesis — and every link has a type, so the structure of your reasoning is recorded rather than implied. Anything unsettled is drawn dashed, so an unfinished argument never looks finished. And a blocker can flip into the spark for a new idea. Runs entirely in your browser. No account, no server. Free and open source.
Hi Product Hunt,
I work on large, slow-moving projects — the kind with long timelines,
many stakeholders, and decisions that are expensive to reverse. Most of
the problems I deal with arrive unstructured: no clean brief, just a
pile of information, opinions and constraints that somebody has to turn
into a decision.
What I noticed over the years is that things rarely go wrong for lack of
information. They go wrong because nobody can see the shape of the
argument. Which assumption everything rests on. Which objection never
actually got answered. Which option quietly depends on a premise no one
said out loud.
Notes and mind maps didn't help me, because they connect ideas without
ever saying HOW they connect. So in Weave that part is not optional.
Every node is labelled for what it is, and every link carries a
relationship — this decomposes that, this answers it, this blocks it.
Two things I'm most attached to:
Anything you haven't settled is drawn with a dashed outline. The tool
refuses to let an unfinished thought look finished, which is the
opposite of what a tidy diagram usually does to you.
And a blocker isn't a dead end. Select a gap and it can flip into a
spark — it becomes a loose idea on the creative side of the canvas,
which can then combine into something new. So the thing that stopped
your solution becomes the input to your next one.
A few things that mattered to me while building it:
It runs entirely in your browser. No account, no sign-up, and no server
of mine ever sees what you're working on. If you connect an AI, you
bring your own key and it talks to your provider directly. Without a key
the AI is simply off and everything else still works. It's free and open
source under AGPL-3.0, and it will stay that way.
This is a version one, built by one person, and there are rough edges.
What I'd really like to know: does the vocabulary hold up for your kind
of problem? Six node kinds and seven link types is a strong claim about
how thinking decomposes, and I'd rather hear now that it's wrong.
Thanks for taking a look.