Roadmap: the direction, and the things we have ruled out

Flockbay is a desktop app for Windows and Linux, with an earlier Mac build. Godot is bundled inside, so the game you build is a real Godot project and it runs locally on your own machine.

A project starts from your description of the game and one question, 2D or 3D, with the answer read from that description, and nothing created until you confirm it. Both are the same app, the same builder and the same chat.

The work splits two ways. You build the world by hand with manual controls, and the controls are the ones that world actually has: terrain, water, weather and placement in 3D; in 2D, ground painted a cell at a time with the tiles, the edges and the collision falling out of the coverage, and sprites put down with the mouse. Flockbay writes the code, as GDScript, inside your project. You play the result and say whether it is right.

Building needs an internet connection because the chat, asset generation and world data all come over the network. Playing the game does not.

Published dates on unbuilt features are a way of making promises with no cost attached, and they age badly. We would rather say what direction the work is going in and let the releases speak for what has actually landed.

More of the world becomes something you can grab. Every part of a game that has a shape, a position or a feel should have a control attached to it rather than requiring a sentence. The measure of progress is how much of your world you can build with your hands.

Quality problems get solved by building capability, not by reaching for a bigger AI model. If something the tool produces is not good enough, the answer is a better control or a better capability underneath, because that is what holds up as projects get larger.

And it stays general. No genre templates, no catalogue of supported game types, no list of allowed things. A feature that only works for the example in the demo is not finished.

Cloud building and GPU streaming. This was considered and cancelled. Games run on your machine, and that is a decision rather than a stage we are passing through.

A browser version of the builder. The reason to install is the frame budget, and a browser version would forfeit exactly the thing that makes the tool worth installing.

An offline mode. The AI and the world data are network services and there is no honest way around that, so there will be no degraded offline build pretending otherwise.

Team and enterprise plans. This is a desktop app one person installs, so there is nothing to buy seats for. Flockbay is free on your own AI: your Claude or ChatGPT plan, your own agent, or an API key.

Specific complaints from playing your own game are the most useful thing anyone sends. "The terrain tool cannot make a cliff steep enough" is worth more than a feature request, and so is "the ground is tiles and I wanted a curved hillside", because both name a limit that showed up in real work.

Being non-technical does not disqualify you from that. You are the one playing the game and deciding whether it is right, which makes your judgement the input that matters most.

Windows gets each release first. Linux gets the same app as an x86_64 AppImage, and the Mac build, for Apple Silicon, is an earlier release that does not yet run Claude in the app or host the full editor. A result measured on one machine is not a verdict about another.

That order is deliberate. One platform carries each new version first, so a fix ships as soon as it works there, and Linux and the Mac follow from the same code rather than from a separate port.