The build team
LoopCodeLab uses agents as interchangeable workers. Available choices include Claude, Codex, Kimi, Grok, Qwen, Gemini, Vibe, and GLM.
Who does what
Section titled “Who does what”| Role | Responsibility | When you choose it |
|---|---|---|
| Planner | Turns the idea into stories and acceptance criteria | During plan review |
| Worker | Builds one or more stories | Before the build or while rerouting work |
| Master | Reviews, accepts, and integrates the stories | Before the build |
You can choose a team for your build. If one story gets stuck, you can swap that story to another agent while the build is running.
Choose build models
Section titled “Choose build models”The build-model picker only shows models that are currently connected to your account. Connect an API key or complete a supported CLI sign-in under Settings → Agents before choosing it for a build. See the Settings tour for connection options.
Every team needs one lead model to review and integrate the workers’ results. If a previously selected model is no longer connected, LoopCodeLab removes that stale selection so the build cannot start with an unavailable model. If no build models are connected, the empty picker shows Connect models, which takes you to Settings to add one.
Solo mode
Section titled “Solo mode”Solo mode also works. One agent can build every story on a lower-cost model, then review the work on a stronger model. It is a simple choice for small builds. A team is more useful when stories are independent and can run in parallel.
Shared rules for the work
Section titled “Shared rules for the work”LoopCodeLab adds skills to each story when they fit the task. These are practical instructions for work such as PWAs, Flutter, media, or Firebase. They help keep the same quality rules in place across different agents.