How Knode is installed

Status: accepted, 2026-10-19

Context

Some people want Knode only as a memory for their agents, from the terminal; others want the app. Both have to share the same core and the same rule from decision 0005: only one process opens a space.

Decision

• One binary, knode, holds the CLI, the MCP server (knode mcp) and Knode Service. Installing the CLI is enough to give every agent a memory, without the app. When Knode Service is not running, the CLI starts it in the background.
• Install the CLI: brew install knodemd/tap/knode on macOS and Linux, or curl -fsSL https://knode.md/install.sh | sh.
• The Mac app carries the same binary inside. Install it with brew install --cask knodemd/tap/knode or from knode.md. The app offers to install the knode command; if the CLI is already installed with Homebrew, it uses that one instead.
• Connect an agent: knode connect claude-code (also codex, gemini, claude-desktop), or by hand, for example claude mcp add knode -- knode mcp.
• The iPhone and iPad app comes from the App Store; the web app runs at app.knode.md.
• Libraries: knode-format and knode-md on crates.io, and the Swift package at https://github.com/knodemd/knode.
• Linux gets the CLI early, because the core is Rust; the Linux and Windows apps come much later.
• Everything stays in the knode repository. homebrew-tap only holds the formula and the cask, and knode.md/install lists every command.

Consequences

• One version for the CLI, the MCP server, the service and the apps.
• Releases build the binary for macOS (Apple silicon and Intel) and Linux, sign and notarise the Mac builds, and bump the formula and the cask.
• The install page on knode.md and this decision change together.