linusroxbergh/svall

> [!NOTE]
> Alpha: early and changing fast. Expect rough edges.
A macOS map of your Claude Code, Codex and OpenCode agents. Every terminal is a
robot on an island, so you can see at a glance which agents are working,
waiting on you or done.
Islands group your robots, say one per project, on a map, and a board
sorts them into columns by stage.
Each card shows its agent's status, context use, model and links such as its
PR and ticket.
Each terminal is a tmux window. Quitting a fleet's window stops its
terminals; opening a robot again resumes its session with
claude --resume, codex resume or opencode -s.
Give an island or robot instructions, links, files or folders, and
agents leave notes for the next agent. Each session starts with a short brief
of all of it, plus its repository, PR and ticket, and gets
a diff when any of it changes. The brief also tells the agent when to hand
work to a new robot.
Resources lists the instructions, skills, subagents, commands, MCP servers,
plugins, hooks and settings that Claude Code and Codex give your agents, and
opens each in an editor.
Every terminal has a browser, a file editor and the working tree's changes
beside it.
The scribe names each robot and keeps its note up to date.
Mission control's buttons start a helper agent that updates or reports on
the fleet.
A macOS banner tells you when an agent is blocked or done. A blocked one has
Approve and Deny on it.
Your phone gets the fleet and any terminal over Tailscale, with push
notifications.
The svall CLI scripts the fleet: make robots, send prompts, wait for
them.
## Quickstart
You need an Apple Silicon Mac on macOS 15 or newer, and the Claude Code CLI,
the Codex CLI or OpenCode. The Claude and Codex desktop apps don't install the
CLIs.
1. Install the Claude Code CLI, which
works with a Claude subscription or an Anthropic API key, or the
Codex CLI 0.156 or newer, which
works with an OpenAI API key alone, or OpenCode
2.0.22 or newer, which runs on OpenCode Zen's free models without an account
or on any provider you sign in to. See Using API keys.
2. Download Svall from svall.dev and drag Svall.app to
Applications, or run:
curl -fsSL https://svall.dev/install.sh | sh
The script installs to /Applications, or ~/Applications when that isn't
writable, or $SVALLINSTALLDIR. It checks the download's checksum,
developer signature and Apple notarization, then opens the app.
3. Open Svall. The setup screen lists the agents it found and asks for your
projects folder. Review changes lists the files it will write. If it
finds none of the three CLIs, it shows how to install one and checks again
when you return. Its hooks do nothing outside Svall;
What setup changes has the details. If
~/.local/bin isn't on your PATH, it shows the line to add to your shell
profile so the svall command works.
4. A new fleet opens with Island 1 holding one robot, a shell in your
projects folder, and a short tour of the map. The shell becomes an agent
when you type claude, codex or opencode. + New island on the sandbar
at the bottom makes another island, and Cmd+T another robot. With more
than one CLI installed, the app asks which one the scribe and mission
control run. Change it later in Settings or with svall agent codex.