Conductor

Run Kapable's designer on your own Claude subscription. A small daemon called kapable-conductor (Kapable Conductor) sits on your Mac. Kapable sends design turns to it instead of to Kapable's own model access. The turns then draw on the Claude plan you already pay for, at your own rate limits, on hardware you control.

curl -fsSL https://cli.kapable.ai/install.sh | sh

Why run it yourself

• Your subscription, your limits. A design turn spends your Claude plan, not Kapable AI credits.
• No platform spend cap. Kapable's own AI spend caps do not gate these turns. Kapable still records what each turn cost, so you can see it.
• Your machine. The model session runs on your Mac, under your user account.

You need a Kapable org first. The Kapable team creates that org for you and adds you as its owner. Self-serve signup is closed today, so there is no page to sign up on.

What gets installed

The CLI fetches the daemon from the same public release registry it updates itself from. It checks the SHA-256 before it installs anything.

Install

1. Install the two clients.

curl -fsSL https://claude.ai/install.sh | bash
curl -fsSL https://cli.kapable.ai/install.sh | sh

Both land in ~/.local/bin. Both print advice about adding that folder to your PATH via ~/.zshrc.

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zprofile
export PATH="$HOME/.local/bin:$PATH"

kapable --version   # 0.18.0 or later
claude --version

2. Sign in to Claude. This is your own subscription.

claude auth login --claudeai
claude auth status

The login opens a browser, and a human has to finish it. That is the whole point of this setup. Kapable never handles your Claude credential. claude auth status tells you whether the login took.

3. Sign in to Kapable.

kapable login

With no flags this runs a device-code flow. It prints a URL and a short code, you approve the code in any browser, and the command waits until you do. It never opens a browser for you, so an agent can run it on your behalf.

4. Install the conductor.

kapable conductor install

One command does the whole job. It downloads the daemon and checks the SHA-256. It registers the daemon with the platform and mints its credentials. It writes the launch agent and starts the daemon. Last, it checks that Kapable can reach your Mac.

Two variants exist for repair work. kapable conductor setup redoes the platform-side half on its own, which is what you want after a credential rotation. kapable conductor install --no-setup puts the daemon on your Mac and skips that platform-side half.

Check it worked

kapable conductor status

Two facts matter, and only these two. The daemon has a recent heartbeat. Kapable can reach it. If status reports a daemon that runs locally but is not reachable, the install did not finish — run kapable conductor setup again.

Trust heartbeat age over any word like online. An online label can survive a daemon that died weeks ago. A heartbeat from a minute ago cannot.

This page prints no sample output on purpose. Any output we typed here would be illustrative rather than real, and a made-up line is worse than none. Read your own.

Turning it on for a design session

Installing the daemon does not move your design work by itself. Each design session carries a backend setting, and Kapable sets that for you today. The designer has no self-serve toggle yet, and no org-wide default.

So: ask us to point a session at your conductor. Ask again when you want that session back on Kapable's model access. We expect to replace this step with a switch you control.

What it exposes, and what it does not

What the daemon does on your Mac:

• It opens an outbound tunnel to Kapable. Nobody has to open an inbound port on your router or firewall.
• It refuses every request that carries no valid token.
• It runs Claude Code sessions that Kapable asks for. Those sessions run as your user, with your user's reach.
• Each session works in its own scratch folder under ~/.kapable/. The daemon creates that folder.
• It will not overwrite an .mcp.json file that already exists in a project folder. It refuses instead.
• Daemon credentials live in the launch agent's environment, in a file only your user can read. They never sit on a command line, where any other local account could read them.

What stays off your Mac:

• No Kapable write credential. The designer's tools run on Kapable's side. Your Claude session calls them back over an authenticated channel, so no repository token reaches your machine.
• Your Claude credential stays local. Claude Code stores it. Kapable never reads it.
• No file sweep. The daemon runs the turns Kapable sends. It does not crawl your disk or ship files anywhere on its own.

One consequence is worth naming. Your Claude session plans for itself, so a conductor turn skips the designer's own lint and compile nag loop. Edits still compile-check when you ship them.

Removing it

kapable conductor uninstall

That stops the daemon, removes the launch agent, and drops its registration on the platform side. The same status note as above applies: we have not yet run this verb end to end on a clean Mac.

If you would rather do it by hand, or the verb is missing:

launchctl bootout gui/$(id -u)/dev.kapable.conductor
rm ~/Library/LaunchAgents/dev.kapable.conductor.plist
rm -rf ~/.kapable/conductor-scratch

Tell us before you remove the daemon. A design session still aimed at a dead conductor fails its turns. We point those sessions back at Kapable's model access, and the designer works as it did before. Removing the daemon touches neither your Claude login nor the kapable CLI.

Next Steps

Next

Getting Started · Authentication