Sillar
July 2026 – Present
An AI design tool that ships working software: you tell the chat your idea (or drive it over MCP from Claude), it lays out the screens on a Figma-like canvas and wires them to a real database on its own. One button and it's live, with automatic deploys. Built for designers who want to deliver systems — shops, bookings, CRMs — without a programmer, and sell them to their clients through the same marketplace.
The canvas: free drawing, every screen in its three variants, with the layer tree and the inspector alongside Same pieces, different sites: shop, bookings, portfolio and blog
The problem
Every small business needs the same software in a different shape, so building one per vertical doesn't scale. The no-code builders that exist are paid for in performance and lock-in, and the person who knows design still ends up needing a programmer to wire the data.
Features
- Figma-like canvas: you draw freely, and every screen shows its three variants — desktop, tablet and phone — side by side
- Tables are data, not code: creating one is a row in collections, and every record is validated against that schema
- AI chat inside the canvas: ask for the board in plain language and it builds screens, tables and wiring; you tune it by hand afterwards
- HTML import: paste a design, it becomes JSON, it is drawn on the canvas, and the connected collections come out of it
- Server or client per screen, always on Next.js: the choice changes SEO and moves the board's price floor
- Per-screen SEO meter and sitemap, generated from the canvas itself
- An in-house ORM scoped to the workspace: CRUD written on the canvas cannot reach outside that board's data
- Board versioning: a business moves to v2 or v3 and rolls back without losing data
- Per-screen permissions, with read and write granted separately, and a log of who touched what
- Plans with limits and a price floor the system computes from the board's real consumption
- Three roles in a single app, each with its own home and metrics: builder, reseller and user
- Support with consent: the reseller requests read-only access to a client's board and the business grants or revokes it
- Revenue split between builder, reseller and platform, with holds and releases driven by payment state
- Full isolation between boards: each one with its own workspace and its own data
My responsibility
My own project, solo: the data model, the Go API, the canvas and the three roles in Next.js.
Stack
- Frontend
- Next.js 16 (App Router), React 19, Vercel AI SDK (Anthropic / Google / OpenAI / xAI), Zod, in-house Figma-like canvas
- Backend
- Go 1.25, PostgreSQL (JSONB, schema-as-data), AWS S3, per-board isolation and audit logging