AI analysis grounded in the code graph — computed facts, not vibes · 2026-10-05T03:03:24Z
T3 Code is a web/desktop/mobile GUI wrapping multiple coding-agent CLIs — Codex, Claude Code, Cursor CLI and OpenCode — behind a single interface. Users install and authenticate the underlying CLI providers themselves, then drive them through T3 Code's chat, diff, terminal and git-actions views (ChatViewContent, DiffPanel, ThreadTerminalDrawer, GitActionsControl). It ships as an npx t3@latest runner, a desktop app (winget/Homebrew/AUR), and provider adapter layers in apps/server/src/provider/Layers (e.g. CursorAdapter, GrokAdapter, CodexSessionRuntime) that normalise each CLI's session/turn model.
The commit log shows active, near-daily feature and polish work (nightly releases roughly every 1-2 days, e.g. v0.0.29-nightly builds from 2026-07-24 through 2026-07-27) plus fast provider catch-up such as "Add Claude Opus 5 model" and Clerk/Expo upgrades — evidence of a project keeping pace with the fast-moving coding-agent ecosystem. The README's positioning as a unifying GUI for Codex/Claude/Cursor/OpenCode, combined with frictionless install paths (npx t3@latest, winget, brew, AUR), plausibly explains adoption, though the README itself states the project is "very very early" and explicitly not yet accepting outside contributions — so growth here appears driven by usefulness/timing rather than community-building claims.
What changed recently, how it's actually built (from the code graph), and whether you should care. Free account — no card, no spam.