Signal Mesh
A local-first graph that connects scattered notes, links and files into one navigable knowledge surface.
Problem
Knowledge gets scattered across tabs, notes and chats. Search finds documents, not relationships.
Idea
What if the connections between things were the interface, not the search box?
What we built
A browser-based graph engine that embeds every captured item locally and renders a live relationship map you can walk through.
How it works
Items are embedded on-device, clustered incrementally, and drawn with a GPU-accelerated force layout that stays interactive past ten thousand nodes.
Learnings
- Relationship density matters more than embedding quality below ~500 items.
- On-device inference removes the trust barrier almost entirely.
- People navigate graphs far better when motion is damped, not springy.
Future direction
Shared graphs with selective sync, and an API so other experiments can write into the mesh.
More from the lab
Have an idea worth exploring?
Whether it's a product, a prototype or a question nobody has answered yet — we'd like to hear it.