AI and agents
Claude Code
Flockbay runs Claude Code, Anthropic’s coding agent, on your own paid Claude account and shows its work in Flockbay’s chat. You pay Anthropic for the subscription you already have, and Flockbay nothing. On Windows, macOS and Linux; on a Mac it uses the Claude Code you have installed or installs it with Homebrew, and on Linux it uses the Claude Code you have installed.
Pick Claude Code once
A new chat runs on Flockbay’s hosted AI, which is free to start. Press Use your own AI in the composer’s usage gauge menu, then Sign in to Claude Code. Or open the model menu, the pill beside Send, and pick the Claude Code tab. Connectors has the same Claude Code row.
The choice belongs to the chat, like any model. Flockbay starts Claude Code for each message and resumes the same session, so the conversation carries on across messages and across restarts of the app.
The first time: sign in
- On Windows, Claude Code comes with Flockbay: Anthropic’s own signed program, unchanged, the version that release of Flockbay was built with. Nothing is installed, and a copy you already have is left as it is. On a Mac, Flockbay uses the Claude Code you have installed, or installs Anthropic’s signed package with Homebrew.
- Press Sign in to Claude Code. Your browser opens on Anthropic’s page, you sign in there, and the card says Claude Code is ready. If the browser did not open, the card has Open the sign-in page and Copy link.
The sign-in is Claude Code’s own, the same one your terminal uses. If you already use Claude Code in a terminal, you are signed in here too.
Which Claude accounts work
A paid Claude account: Pro, Max, Team or Enterprise. The free Claude plan does not include Claude Code, so it cannot build here: the card says "Your Claude account can’t build here", with Upgrade Claude and Sign in again, and you upgrade on Anthropic’s site.
Two more refusals are Anthropic’s: "Claude Code isn’t offered in your country", and "Claude Code isn’t available on this Claude account", when its plan ended or its organization turned Claude Code off. The card says which.
What Flockbay reads, and what it never touches
Flockbay asks Claude Code three things about your account: its auth status (whether you are signed in, as whom, and on which plan), the models it offers you, and how much of your plan’s usage limits is left. The card shows that account so you can see which one is in use, and the usage bar shows the limits.
A chat on your Claude never reads or writes Claude Code’s sign-in or history, and reads its settings file only on a Mac or Linux, to see which update channel you follow. It does not modify or wrap the Claude Code program, and it sets no API key and no proxy for it, so your usage goes to Anthropic on your own plan. Install skills and hooks, in Connect your agent, is the one thing that writes to Claude Code’s files: it adds Flockbay’s skills and three hooks to your Claude Code user settings, and those hooks then run in every Claude Code session on the computer.
Flockbay’s tools and instructions are handed to Claude Code for each session, on its command line. Your own Claude settings and MCP servers load beside them and are not edited.
What Flockbay counts is that Claude was picked, signed in and used, and which kind of Claude plan it ran on. Once per install, before you pick Claude, it also checks whether Claude Code on this computer is already signed in, and on which plan. Those are events, filed under your Flockbay account, with no prompt, no reply and no file name in them.
One kind of text goes to Flockbay’s server from a chat on Claude Code: the request behind anything you have Flockbay make for you, a picture, a sound, a voice-over, music or a 3D model, because Flockbay’s server makes it. Privacy lists everything that leaves your computer.
What runs without asking, and what asks
A new chat starts in Edit automatically. There, Flockbay’s own tools run without an approval card, including the ones that run scripts in your game, the Godot editor and Blender, and so do file edits inside the project you have open. Other shell commands ask first, as Claude Code’s own accept-edits mode decides, on a card in the chat with Allow and Deny.
The Permissions menu in the composer changes that per chat, and Ctrl⌘+Shift+M steps through the modes. On Claude Code it has four: Plan mode, where every message starts with a plan and nothing changes until you press Build it; Ask before edits, which asks before any of Flockbay’s tools that change something, and before every edit and command; Edit automatically, the default above; and Full access, which runs commands and reads and edits files anywhere on this computer without asking. Your own Claude Code settings can allow more in any mode.
A permission prompt is one card with Allow, Always allow and Deny, and a question from Claude Code arrives as a question card, the same one Flockbay’s own questions use.
Model and Effort
The model menu’s Claude Code tab lists the models Claude Code offers your account, under their real names, with Default first. Effort offers the levels Claude Code gives for the model you picked, up to Extra high and Max; a model with no levels shows no control. Both are remembered per chat.
Updating Claude Code
On Windows, the Claude Code that comes with Flockbay updates with Flockbay: each release carries the Claude Code it was built with.
A copy you installed yourself keeps updating the way it always did, and both copies read the same sign-in and settings. When a message fails the way an older Claude Code does, the card offers Update Claude Code, with the line to run in your own terminal.
Stopping
Stop, where Send is while a turn runs, ends the turn and the Claude Code process behind it.
To stop using Claude Code in a chat, pick another model in the model menu. Flockbay has no sign-out for Claude Code, because the sign-in is Claude Code’s own: run /logout in Claude Code itself.
A turn on your own AI may run for up to twelve hours, and a question card waits as long as you do.
Where it breaks
It runs on Windows, macOS and Linux.
On a PC whose organization blocks programs it does not know, Windows can refuse to start Claude Code. The card says "Windows won’t start Claude Code on this computer", and nothing in Flockbay can change that.
AppImage, x86_64. Ubuntu 22.04, Debian 12, Fedora 39 or newer.
Intel or AMD Chromebook, with Linux turned on.
