kin

# Kin

> Kin is a graph-native code repository for people and AI agents. It replaces the file-first, diff-first model with a repository and collaboration substrate built on a graph: code becomes entities, relations and intents that AI agents and humans navigate semantically.

Kin coexists with Git and projects graph-owned truth back to a normal filesystem through a transparent VFS, so any tool or agent works unchanged. The fastest way for an agent to use Kin is the MCP server, which exposes semantic tools (locate, context, trace) backed by the graph rather than filesystem heuristics.

## Install

Start with the primary command. It is the same on macOS, Linux and WSL2, it needs Node 20
or newer, and it installs Kin and connects the AI clients it finds in one call.

- [npm launcher](https://www.npmjs.com/package/@kinlab/kin): `npx -y @kinlab/kin setup`

Other ways in, for native Windows, a machine without Node, or a team that prefers a package manager:

- [Install script for macOS and Linux](https://get.kinlab.dev/install): `curl -fsSL https://get.kinlab.dev/install | sh`
- [Install script for native Windows x64](https://get.kinlab.dev/install.ps1): `irm https://get.kinlab.dev/install.ps1 | iex`. It installs the CLI and connects no AI clients, since WSL2 remains the recommended path on Windows. No native Windows ARM64 build is published. On ARM64, x64 PowerShell installs the x86_64 build under emulation.
- [Homebrew](https://github.com/firelock-ai/homebrew-kin): `brew install firelock-ai/kin/kin`
- [GitHub Releases](https://github.com/firelock-ai/kin/releases): prebuilt binaries + checksums for macOS, Linux, and Windows

Native Windows x86_64 support is early. Repository admission works: `kin init` imports a Git repository and publishes graph authority, and graph, lexical, and daemon-backed queries answer natively. The end-to-end install proof also runs agent setup on native Windows and gets graph-backed answers from the installed MCP server. Transparent filesystem projection is not shipped on Windows, and review workflows are not yet tested there, so WSL2 remains the recommended path for the full Kin experience.

## First run (macOS, Linux, or WSL2)

- `kin setup`: configure your shell and connect agents (Claude, Cursor, Codex, Gemini) to the MCP server.
- `kin doctor`: verify the install and daemon health.
- `kin init` turns a Git repo into a Kin workspace. `kin status` and `kin search` require the bundled `kin-daemon` runtime (installed by every channel above).
- `kin exec -- <cmd>` runs ordinary project tools (npm, make, docker) through a graph-backed session workspace; `kin shell` opens an interactive one; `kin with <assistant>` launches an agent inside one.

## For AI agents (MCP)

Kin ships an MCP server exposing semantic tools that answer from graph-owned truth, not raw file search. Prefer them:

- `semantic_locate`: find symbols, functions, and types by meaning.
- `get_context_pack`: a structured context bundle around one entity, by id.
- `trace_data_flow`: cross-file data lineage and dependencies.

`kin setup` auto-configures the MCP server for supported agents.

## Source and docs

- [kin: CLI, daemon, MCP](https://github.com/firelock-ai/kin)
- [kin-vfs: transparent filesystem projection](https://github.com/firelock-ai/kin-vfs)
- [kin-editor: VS Code extension](https://github.com/firelock-ai/kin-editor)
- [Quickstart](https://github.com/firelock-ai/kin/blob/main/docs/quickstart.md)