Skip to content

Kast

Kast gives Copilot and other agents compiler-backed Kotlin answers without asking them to guess from text search. The install model has two scopes: install the kast binary once on the machine, then add Copilot integration files to each repository where you want agents to use Kast.

The golden path

Run these commands on a developer machine, then restart IDEA or Android Studio so the IDE host loads the Homebrew-managed plugin and Copilot picks up the repository-local files.

Install Kast globally, then add it to one repository
brew tap amichne/kast
brew install kast

cd /path/to/your/repository
kast install copilot

Two scopes, one setup

brew install kast is the macOS machine-level install. It puts the global kast binary on PATH and installs or refreshes the matching kast-plugin cask, the same plugin path exposed by brew install --cask kast-plugin, so local JetBrains profiles link to a version-coupled IDE backend. kast install copilot is a repository-level install. It writes managed files under this repository's .github directory so Copilot can start kast lsp --stdio, load Kotlin instructions, and expose Kast tools.

When to rerun kast install copilot

Run it once in every repository where Copilot should use Kast. Rerun it with --force after upgrading the global binary or when the repository files look stale.

Where the IDEA plugin fits

The IDEA or Android Studio plugin is part of the macOS Homebrew developer install, not a per-repository choice. Homebrew manages the cask and links it into local JetBrains profiles so Kast can reuse the IDE project model and indexes on developer machines. Use brew reinstall --cask kast-plugin only when profile links need direct repair.

What this gives your agent

Kast is for the work that happens after a prompt asks for real Kotlin understanding: find the exact declaration, prove a usage list, inspect a call tree, or plan a safe rename.

  • Symbol identity: resolve the declaration the compiler sees, not every line that happens to match a string.
  • Bounded evidence: report whether reference and hierarchy results are complete, truncated, or limited by a configured bound.
  • Safe edits: plan rename and edit operations with file hashes before writing anything.
  • Workspace awareness: answer from the Gradle workspace instead of a pile of unrelated files.

Headless Linux servers

Use the Linux headless bundle when the machine is a CI runner, hosted agent, server snapshot, or image build with no developer IDE. That path installs its own kast binary and bundled headless runtime on the server.

Install on Ubuntu or Debian from the headless bundle
export KAST_UBUNTU_DEBIAN_VERSION="v1.2.3"
./scripts/install-ubuntu-debian.sh install
kast up --backend=headless
Why this is separate from Homebrew

Homebrew is the developer-machine path. The Ubuntu/Debian bundle is the headless-server path. Use it when the agent needs its own binary, config, runtime libraries, and backend without relying on a human shell profile or an already-open IDE. It is not the local macOS developer-machine fallback.

Where to go next

Choose the page that matches the job in front of you. The first two pages are the main path; reference material stays available after that.

  • Install

    Install the macOS developer-machine binary and plugin, then add repository-local Copilot files.

    Install

  • Use with agents

    Understand what the Copilot package gives an agent and when to use the direct CLI fallback.

    Agent setup

  • Supported use cases

    See where Kast is meant to help, where it excels, and which complexity can stay out of the first install.

    Use cases

  • Reference

    Keep the detailed API, CLI, backend, and architecture material available when you need exact behavior.

    CLI reference