Task Orchestratortion layer on top of - Code

A local Kanban companion for T3 Code that plans your goal, runs each task in its own worktree, and reviews the code before you do.

What it does

  1. 01

    Plan

    You write one large goal. A planner agent reads your repo and proposes 3 to 40 tasks with dependencies, acceptance criteria, complexity, and the code paths each touches. You edit or refine, then accept.
  2. 02

    Board (Kanban)

    Tasks flow through Backlog, Ready, Running, AI Review, Human Review, and Done, plus Needs Input and Failed.
  3. 03

    Run tasks

    Each task runs as a T3 Code thread in its own git worktree and branch. A scheduler starts Ready tasks when their dependencies are Done, within a parallel limit, and avoids tasks touching the same files.
  4. 04

    AI review loop

    Each finished task gets a separate AI reviewer thread. Blocking findings go back to the task agent for fix rounds, up to maxFixRounds, before you ever look at it.
  5. 05

    Needs Input

    When an agent asks a question, the card moves to Needs Input and you get a desktop toast. Answer in T3 and the card moves back on its own.
  6. 06

    Human review

    You read the diff and AI findings, then approve and merge locally with git merge --no-ff, open a PR, or request changes.

Run T3 Orchestrator locally

T3 Code must be installed and running on your system. T3 Orchestrator never runs agents itself; it talks to the running T3 Code app through the t3 CLI.

  • Node 22 or newer and pnpm 10.
  • git.
  • The T3 Code desktop app running.
  • Windows CLI launcher: ~/.t3/bin/t3.cmd

Clone the repository

git clone https://github.com/nishantsingh02/T3-Orchestrator.git
cd T3-Orchestrator

Install and run

pnpm install
pnpm dev

The server starts on port 4417 and the web app on port 5173. Open http://localhost:5173.

Without real agents

pnpm dev:try

Uses a scripted fake T3 client for UI work and demos.

Production

pnpm build
pnpm start

Serves the API and the web app on http://127.0.0.1:4417.

Use T3 Orchestrator

One board carries the whole workflow: plan, run, review, merge.

  1. 01

    Create a board

    Pick a T3 Code project or type a repo path, then choose a base branch. Leave "Learn about the project from its T3 threads" on to import existing thread context into the board brief.
  2. 02

    Write a plan

    Open Plans, or press P. The planner reads the repo and proposes tasks with dependencies. Edit the table, refine, then accept. Tasks land in Ready or Backlog.
  3. 03

    Run tasks

    Press Start on a card, or turn on Auto-run. The scheduler starts Ready tasks whose dependencies are Done, within the parallel limit, and avoids tasks that touch the same paths.
  4. 04

    Review automatically

    Each finished task gets its own AI reviewer thread. Blocking findings go back to the task for fix rounds before you see it.
  5. 05

    Answer questions

    When an agent asks something, the card moves to Needs Input and you get a desktop toast. Answer in T3; the card moves back on its own.
  6. 06

    Review and merge

    Read the diff and findings, then approve and merge into your base branch, open a PR, request changes, or mark done.
  7. 07

    Run the project

    The Run button in the toolbar starts the root checkout and gives you its link. The same button in a task drawer runs that task worktree so you can try a branch before merging.
  8. 08

    Track progress

    Overview shows progress, pace, blockers, and where the time went. Roadmap holds milestones and the suggestions inbox.
  9. 09

    Ask, audit, and add context

    Insights answers questions about the project. Security runs audits, secret scans, and dependency scans. Context holds files and the knowledge graph.