T3 Code setup with Fable 5, Grok 4.5 and Muse Spark

Learn a multi-model coding workflow in T3 Code, with separate worktrees for implementation and an authentication caveat for server hosting.

Player not loading? Watch on YouTube

The speaker presents T3 Code as an open source coding assistant interface that drives provider CLIs. The tutorial covers its desktop app and a self-hosted web server option, then connects subscription logins and a Meta Model API key. Hosting the interface on your own server is the deployment option discussed; the model setup uses provider accounts and APIs.

The workflow assigns Fable 5 to planning and review, Grok 4.5 or GPT-5.6 to backend implementation, and Muse Spark 1.1 to visual design through OpenDesign. These assignments reflect the speaker's tests and preferences. The example is a freelancer invoicing app with authentication, a database and PDF generation.

Design comes first. The speaker supplies reference screenshots to OpenDesign, exports the design context, and gives it to Fable in T3 Code's plan mode. Fable produces separate frontend and backend tasks with an API contract. Grok then implements the backend in a worktree, where an action installs dependencies automatically.

The speaker reports that Muse Spark overwrote existing files during parallel work and recommends a separate worktree with frontend-only responsibilities. They also say T3 Code uses more memory than Verdant and that its web server lacks built-in authentication, so hosted access needs password protection. For difficult backend logic, the speaker prefers GPT-5.6; Grok is their default for routine implementation.