Persistent sessions per card
Every card owns its own agent session. Drag a finished card back to Backlog or press Continue, and the agent picks up exactly where it left off — no re-explaining.
A Futuretea product
One board for all your coding agents.
Orchestea is a desktop command center for AI coding agents. Give it a task, and an organizer agent breaks it into subtasks, runs them through Claude Code and other agents, and keeps every session, approval and log on a single Kanban board.
Currently in private beta for macOS.
Runs locally · Claude Code, Codex and Antigravity · Built with Tauri 2
Running several agents at once turns into tab chaos.
Coding agents are now good enough to work in parallel. Managing them isn't. Sessions live in separate terminals, approvals wait unnoticed, context is lost between runs, and you can't tell which agent is near its usage limit until it stops.
Orchestea gives that work a single place to live.
How it works
Create a work order.
Write the task, pick the project folder, choose an organizer and the pool of providers and models allowed to work on it. New tasks always land in Backlog.
Let the organizer split it.
The organizer agent breaks the work into subtasks. They stay inside the parent card, not scattered across the board, so the big picture stays readable.
Watch, approve, continue.
Follow live logs, approve tool requests from the board, and when a card finishes, send a new instruction — it continues in the same session, with its full context.
Features
Tool approvals in one place
When Claude Code asks for permission, the request waits on the board, not in a terminal you forgot about.
Quota awareness
Usage for Claude, Codex and Antigravity is read automatically every five minutes. If a non-required card hits a rate limit, Orchestea reroutes it once to another provider.
Parallel work without conflicts
Optional Git worktree isolation lets agents work on separate branches at the same time, with a merge flow when the work is ready.
Live model lists
Available models are fetched from each provider, so your choices are always current.
Project notes
Pin notes to a project and, when you choose, include them in the first turn of every new card. Off by default — you stay in control of what the agent sees.
Clean history
Archive finished work and clear detailed logs to reclaim space, while keeping titles, summaries, final reports and run metrics.
Built around Claude Code
Claude Code is the agent we rely on most for long, multi-step work, and Orchestea is designed around how it works: persistent sessions you can resume, a permission model surfaced as on-board approvals, and usage tracking that reads Claude's own /usage.
Orchestea also works with Codex and Antigravity, so you can choose the right agent for each task without leaving the board.
Local by design
Orchestea runs on your machine. There is no Orchestea cloud and no extra server between you and your agents — each agent talks only to its own provider, using your existing CLI setup. Task data stays in a local folder you control.
Under the hood
- Interface
- Vue 3 + Vite
- Server
- dependency-free Node.js
- Desktop
- Tauri 2, sharing the same web interface
- Testing
- an offline test suite with mock providers — no CLI or network needed
FAQ
Which agents does Orchestea support?
Claude Code, OpenAI Codex and Google Antigravity. Gemini CLI can be added optionally.
Do I need API keys?
No. Orchestea uses the agent CLIs you already have installed and signed in.
Does my code leave my machine?
Only in the way it already does when you use these agents: each agent sends what it needs to its own provider. Orchestea itself adds no external service.
Which platforms are supported?
macOS (private beta)
How much does it cost?
Free during beta
Stop juggling terminals. Start conducting.
Questions or feedback? info@futuretea.ltd