Desigma is an open-source Figma alternative written in Rust

A product team that uses Figma pays per seat, hands the file to a design surface someone else owns, and goes through a REST API with its own quota each time an agent reads a frame. deepfounder-ai published Desigma, an open-source interface design tool written in Rust and shipped as a native desktop app for Windows, Linux and macOS. The README says it is dual-licensed MIT or Apache-2.0 and that the project is in early alpha, so the author states it is "not yet a replacement for professional daily work."

What is in the editor today

The feature list the README calls complete covers the ground a working designer uses in a Figma session: frames with auto layout, components with boolean, text, variant and instance-swap properties, prototyping with triggers, overlays, transitions, easing curves, springs and Smart Animate, and a vector tool with pen, shape builder, outline stroke, flatten, simplify, join and split. Text supports per-run styles and the usual spacing and case controls. Fills carry solid, linear, radial, angular and diamond gradients with blend modes. Variables and modes cover colors, numbers, strings and booleans with collections and mode switching. Boards cover whiteboarding with stickies, connectors, shapes, stamps and sections. Decks cover 16:9 presentation frames with speaker notes and PDF export.

A dev mode on each frame generates code in "CSS, Tailwind, SwiftUI, Jetpack Compose or React," which is how a design hands off to the code tree.

How a Figma file gets in

The README names two paths. The first is Figma's REST JSON from GET /v1/files/:key, which is the documented route. The second is the .fig file itself, which Desigma unpacks from its zip wrapper and decodes against the kiwi schema, pulling node changes, embedded images, shared styles, instance overrides, layout grids and prototype interactions. The author is honest about the limit: "The .fig format is undocumented. Our decoder follows community reverse engineering and is tested against synthetic files, so expect gaps on real-world files." Export covers .desigma, Figma REST JSON, PNG, JPG and SVG, and the editor imports PNG, JPEG and WebP for images.

5,000 layers under one frame

The one hard number the README gives is a rendering one. The renderer runs on the CPU, on tiny-skia, with a "spatial index and parallel 256 px tile cache," and the author reports that "panning and dragging on a 5,000-layer page stay well under a frame." The project also carries more than 500 tests across its Rust workspace, and the GitHub API reports 211 commits on the main branch for a repository that is four days old. Stars sit at 182 and forks at 13.

What is still missing

The README lists gaps rather than hiding them. "Collaboration, plugins, PDF export" are called out as not shipped, together with real-file .fig features the decoder has not seen, code-signed release builds, Deck transitions and template gallery, a presenter display, and text selectability inside the PDF export, which currently rasters at high resolution. First-time launches on Windows and macOS need platform-specific security overrides, since the binaries are not signed.

A working design team on Figma has one thing to look at: whether a local, offline tool that already handles frames, components and auto layout, and that generates React or Tailwind or SwiftUI from a frame, is worth running next to the paid file for the parts of the workflow an agent can reach. The project is a few days old, early alpha, and the author says so in the README. Run it on a copy of a real file before trusting it.

Source

• deepfounder-ai/desigma on GitHub, MIT OR Apache-2.0