alexevoy.dev/flightdeck

You don't run one agent anymore. Run the fleet like one.

The moment you go from one AI coding agent to five, you are drowning in terminal tabs, guessing which one is done, and pasting the next task by hand. Flightdeck is the cockpit: every agent in a pane, a board that hands queued work to whichever agent is free, and completion caught for you.

It runs on your own machine and drives the agent CLIs you already pay for. I built it to replace a paid tool I use every day and found slow and clumsy. Now I am opening it up to the people hitting the same wall.

Join the pre-order list ↗ See the cockpit
Pre-release · v0.1.0 · Linux first

The Deck. Panes stream, the board dispatches. You supervise.

The Problem

One agent is a tab. Five is a mess.

A single coding agent fits in a terminal tab. A fleet does not. You lose track of which agent finished, you copy the next task in by hand, and a crashed window takes a running agent down with it. The work is parallel now, but the cockpit never caught up.

Flightdeck is that cockpit. It owns the terminals in a Rust core, so a UI reload never kills a running agent, watches for completion three ways, and moves finished work to a Review lane. It also reads the memory files your agents already use, so the thing steering the fleet knows what the fleet knows.

The Cockpit

Four parts, one window

Nothing here is a demo. This is the tool I open every morning to run my own fleet of agents. Four parts, and they work together.

Agent panes

Live terminals

Every agent in its own pane, side by side, watchable and interruptible. Real terminals, GPU-drawn, with searchable scrollback that survives a restart. The Rust core owns the sessions, so a webview crash never takes an agent with it.

The dispatch board

Auto-assign

A kanban board where queued tasks assign themselves. A task waits, an agent goes idle, the board hands it over, up to a concurrency cap you set. It catches completion three ways, by a hook signal, a process exit, or an idle terminal, then moves the card to Review. Drag a card onto a pane whenever you want to override it.

Editor and files

Git gutters

Read and tweak what an agent wrote without leaving the deck. A real editor with syntax highlighting, in-file search, and a git gutter showing every change against HEAD. Open a file from the browser, or from the list of files a task touched.

The memory bridge

Reads your files

A viewer over the memory your agents already use: CLAUDE.md, AGENTS.md, memory folders, all rendered, with wikilinks resolved and a graph of how they connect. It reads your files where they live. Flightdeck never keeps its own copy of your memory.

Screenshot at launch · the deck
Personal build · in daily use Desktop App

The deck

Agent panes streaming, with the dispatch board alongside. The one screen where the whole fleet is visible at once.

Screenshot at launch · the board
Personal build · in daily use Auto-dispatch

The board

Queued to running to review, cards moving on their own as agents pick up work and finish it. Manual drag always wins when you want it to.

How It Was Built

Built with the agents it now runs

I was paying for a tool that did this job badly: sluggish, clumsy, buggy. So I built my own, using the exact fleet of agents Flightdeck was made to run. The numbers below are straight off the repo, counting hand-written code only, not lock files.

That is the honest pitch. It is the tool I made to solve my own problem, and it works well enough that I stopped paying for the other one. If you run agents the way I do, it will probably save you the same headache.

Build window
Seven active days, from first commit on 2 July 2026 to a working cockpit on 11 July. 41 commits.
Hand-written code
About 11,800 lines of source: 6,448 of Rust across 39 files, 5,200 of TypeScript across 60. Lock files not counted.
What shipped
All four parts, plus the plumbing under them: git-worktree isolation with a merge queue, session restore, and a way to turn a written brief straight into finished changes.
Who built it
One person, in the open, with a fleet of coding agents supervised through an early version of this exact tool.
Straight Answers

Before you pre-order

The things I would want to know first, answered plainly. If yours is not here, email me and I will tell you honestly.

Is it finished?
It is v0.1.0 and I daily-drive it. All four parts work. It is pre-release, so there are rough edges, and pre-order is for people who want in early and want a say in what gets sharpened next.
What do I need to run it?
Your own agent CLIs and their subscriptions, for example Claude Code or Codex. Flightdeck runs the tools you already pay for. It does not include them and does not resell anyone's tokens.
Which platforms?
Linux first. Built and used daily on Fedora with Wayland, and that is what ships first. Mac and Windows are a build target away, but I will not promise a date until they are tested properly.
Does my code leave my machine?
Flightdeck itself is local. It owns the terminals, keeps sessions and layouts in a local database, and phones home to nothing. The agents you run inside it keep their own network behavior. Flightdeck does not add a layer in the middle.
What does the pre-order buy?
A founding license at a price locked for early users, plus the first public release when it lands. Founding users also get direct access to me and a real say in the roadmap.
Who supports it?
One person: me. You get direct access rather than a ticket queue, and founding users shape what comes next. But it is honest to say this is a solo project, not a funded company.
Will you keep working on it?
I use Flightdeck every day to run my own work, so it stays maintained as long as I do. I am not going to promise a roadmap I cannot keep. The honest version: this is the tool I depend on, so it gets fixed when it breaks and sharpened where I feel the friction.
Pre-order

Get in early

One founding tier while it is pre-release. The price locks for early users, and you help decide what gets built next.

Founding pre-order Opening soon

founding price, locked for early users

A founding license to the tool I built to run my own fleet, at a price that will not go up on you. Most tools like this rent you access or meter your usage on top of the Claude subscription you already pay for. This one you own outright.

  • Yours outright: no subscription, no credit meter
  • Founding price, locked for early users
  • The first public release the day it ships
  • Direct access to me, not a ticket queue
  • Linux first, with Mac and Windows to follow
Status: pre-release, v0.1.0 ↗ Join the pre-order list

Want to ask something first? Email me and I will give you a straight answer. If you are here for my client work instead, see app builds, automation, and operator dashboards.