Four products. One workshop.
A project board an AI agent can actually drive, a solid modeling kernel that runs in a browser tab, a filter that keeps AI marketing slop out of your feeds, and a game studio whose players decide what gets built. Four different problems, and every one of them was planned, built, verified and shipped on FeatureBoard itself. That is the product endorsement that counts.
FeatureBoard — the board your agent drives.
An MCP server that gives an agent a real board: a queue that refills itself, work packets that define done, worktrees and verified commits, background checks, drift audits, and a hand-off channel between machines. Stored as markdown you own, on your disk.
NativeCAD — a solid modeling kernel in the browser.
A B-rep solid modeling kernel written from scratch in JavaScript — watertight solids, STEP AP242 in and out, addressable faces and edges, DFM checks, glTF and 3D-PDF export — with zero runtime dependencies, no royalties, and a live modeler in the browser.
SlopRadar — your feed, minus the slop.
A Chrome extension that spots AI-generated marketing slop and engagement bait on LinkedIn, X, Reddit and Threads and quietly hides it — using your browser's own on-device model. Nothing leaves the browser: no server, no analytics, no third parties.
BoardGames — games built by their players.
A game studio with no publisher in the middle. Subscribers get votes and spend them on the games and features they want made, ask for the genres nobody is making, and play what they fund. Every ship, station and machine is modelled in NativeCAD, and every feature you can vote on is a real ticket on a FeatureBoard board.
Everything here was built on FeatureBoard itself, in public — the examples page has the rest of what came out of the same workshop, how we build explains the method, and about says who.