Supported coding agents

1DesignTool drives the coding agents you already have installed — sixteen supported CLIs, detected on your machine and driven headless inside your project folder. There is no built-in model and no bundled key: the agent you pick does the building, on your plan and your quota.

The roster

Claude CodeCodexCursor Agent
GrokOpenCodeAntigravity
QwenKimiAmp
DevinClineHermes
PiOh My PiKiro
Aider

The agent picker — in onboarding, Settings → Agents, or the agent chip in the composer — lists what the app actually found: the binary path, the version, the models and effort levels that CLI exposes, and its sign-in state. A CLI that is installed but not authenticated shows as needing sign-in on its own side.

Settings → Agents — every CLI the app found, its version and default model

How the app talks to them

Each supported CLI has a bridge definition that knows its command shape, its streaming format and its model flags. The app spawns the CLI as a subprocess with your brief, the project folder and the design kit as context, and pumps its output into the transcript panel — file edits, progress and errors alike. Everything after that is the same studio, whichever agent is driving.

Choosing an agent

The strongest model you have quota for makes the best designer — design work rewards taste, and taste tracks model strength. Practical differences matter too: some CLIs expose effort levels (low → max) that trade speed for care, and the per-agent pages in this section call out what each is good at.

  • Claude Code — the most-tuned path; see Design with Claude Code
  • Codex / Cursor / Devin — strong general builders, good for interactive sites
  • Qwen / Kimi / OpenCode / Aider — solid alternatives when those are the accounts you have

Tip: Switch agents per run from the composer chip — a cheap model for rough variants, your strongest for the final polished pass. The project, history and gate don't care which engine produced the files.

If your CLI isn't in the list

The roster is fixed per release — the bridge has to know a CLI's exact command and output contract, so arbitrary shell commands can't be plugged in. If your agent of choice is missing, the changelog tracks newly supported CLIs, and the troubleshooting page covers detection issues for the ones that should appear.