Skip to content
constansinoPublic
forked from pingdotgg/t3code

About

Resources

Stars

1 star

Watchers

0 watching

Forks

 
 

Latest commit

 

History

891 Commits

Folders and files

Repository files navigation

T3 Code

T3 Code is a Codex-first web GUI for coding agents.

This fork focuses on one thing: letting a Windows machine use a Codex-style UI to browse, mirror, and operate Codex sessions from multiple hosts, including the local machine, a LAN MacBook, and remote Linux servers.

It stays compatible with upstream T3 Code concepts, but adds a stronger remote-Codex workflow on top.

What this fork adds

Multi-host Codex sources

  • Local Codex source
  • Remote SSH Codex sources
  • Remote WebSocket Codex sources
  • Per-source connection settings from the Settings page
  • Built-in default sources for the current local machine, LAN Mac, and cloud server
  • Manual add/remove/edit workflow for more hosts

Host-style session browser

  • One top-level source/project per host
  • Source-specific folder labels derived from the real host/home path
  • Collapsible source folders in the sidebar
  • Source threads bound to the correct host automatically
  • A layout that behaves more like a Codex host/session browser instead of a single local-only list

Codex history mirroring

  • Auto-bootstrap of Codex history for local and remote SSH sources
  • Manual history refresh from Settings
  • Bulk import of recent sessions
  • Manual import by session id
  • Per-session mirrored thread ids so history stays stable across refreshes
  • Legacy duplicate mirror cleanup
  • Stale mirror cleanup when the source no longer exposes a session

Better parity with codex resume

  • Reads real local session files from ~/.codex/sessions
  • Uses created-time ordering for mirrored source threads so the sidebar matches codex resume more closely
  • Filters out transient codex exec sessions from the local history view
  • Preserves real session titles instead of falling back to short placeholder ids whenever possible

On-demand content sync

  • Mirrored history rows appear quickly from metadata import
  • Opening a mirrored source thread triggers a full content sync for that session
  • Messages, reasoning items, and activity summaries are imported into the T3 thread model
  • Source folder labels are refreshed from the synced session metadata

Windows + Mac remote workflow

  • Included start-remote-codex.ps1 / start-remote-codex.cmd helpers
  • Included stop-remote-codex.ps1 / stop-remote-codex.cmd helpers
  • SSH tunnel workflow for exposing a Mac-hosted codex app-server as a local Windows WebSocket endpoint
  • A practical setup for Windows UI -> Mac Codex backend

Upstream T3 capabilities still available

This fork keeps the regular T3 Code app model and still supports the core upstream experience, including:

  • Web UI for Codex-backed threads and projects
  • Local or remote Codex app-server runtime selection
  • Diff panel / chat layout integration
  • Project and thread orchestration model
  • Custom model / service-tier app settings
  • Terminal, git, and worktree-oriented T3 flows already present upstream

Quick start

1) Local-only usage

Requirements:

  • Bun
  • Node.js
  • Codex CLI installed and already authenticated

Run:

bun install
bun run build --filter=t3
cd apps/server
node dist/index.mjs --host 127.0.0.1 --port 3773 --no-browser

Open:

2) Windows UI + Mac Codex backend

If you already prepared the mac-codex SSH alias and the remote start/stop scripts on the Mac, use the bundled helpers:

.\start-remote-codex.ps1

or:

start-remote-codex.cmd

That script:

  • starts the remote Mac Codex app-server
  • opens the SSH tunnel to Windows
  • starts the local T3 web server
  • opens http://127.0.0.1:3773/

Stop everything with:

.\stop-remote-codex.ps1

Managing hosts and history

Use Settings -> Codex Sources to:

  • add a local / remote SSH / remote WS source
  • configure sshHost, appServerUrl, and remoteCodexHome
  • list Codex history for that source
  • import selected history rows
  • import recent history in bulk
  • bind mirrored projects and threads back to the correct host

Current limitations

  • Full history crawling is currently implemented for local and remote SSH sources.
  • Remote WebSocket sources can be used as a live backend, but history crawling is not implemented there yet.
  • This is still a fork of a very early-stage project, so you should expect rough edges.

Remote setup notes

For the detailed remote setup guide, see REMOTE.md.

Upstream

About

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages