.gui vs Figma files: portable text vs a vendor’s app

A Figma file is only fully readable inside Figma — the app, the API, or an export. It’s an excellent design environment, but the file itself is not a portable artifact you can open in a text editor, diff in git, or hand to an agent without going through Figma’s servers. A .gui file is plain text. Unzip it and design.guix is readable XML — open it in any editor, git diff it line by line, generate it from a script with no design tool involved at all.

.gui doesn’t require Figma

The Figma plugin exports any Figma layer tree to a .gui file — auto-layout, fills, gradients, effects, and tokens carried over as a 1:1 mapping of Figma’s own layer model, not approximated. That makes Figma one entry point among several, not a requirement. An agent can write a .gui file directly from a prompt with no design tool touched at any point — see Prompting your agent.

The loop, not a one-way export

Because the same file is legible to both a design tool and an agent, it can move in either direction: an agent writes home.gui, a designer opens it in Figma to nudge a color, the file goes back to the agent to wire up. Every stop in that loop reads and writes the same file — nothing gets re-derived from a screenshot at each step. That loop is the actual point; Figma export is one leg of it, not the whole story.

No live connection required

Figma’s own AI integration path runs through MCP (Model Context Protocol) — a live tool-calling connection where an agent queries Figma’s API, often across several round-trips, to reconstruct a design piece by piece. Exporting to .gui once removes that dependency: the result is a static file an agent reads directly, with no persistent connection, no repeated API calls, and no dependency on Figma’s servers being reachable at generation time.