Linux · macOS · Windows via WSL2

Kin

A native AI peer for the terminal. Kin works alongside you in the project itself, with explicit tools and permissions, durable sessions, and the model provider you choose.

Kin terminal timeline showing an AI peer reasoning, calling tools, and reporting results

Install Kin

One command handles both product users and authorized contributors. It installs uv first when needed, never uses sudo, never edits your shell files, and verifies the finished command with kin --version.

bash <(curl -fsSL https://get.kinra.ai/install.sh)
The public wheel path needs no Git package. If this machine is an authorized development seat, install Git first with sudo dnf install git so the installer can choose the editable checkout path.
The public wheel path needs no Git package. Authorized contributors can install Git with sudo apt install git to let the installer select an editable checkout.
The public wheel works without Xcode tools. Authorized contributors can run xcode-select --install first; an existing Homebrew uv is reused.
Kin runs inside WSL2 today. Run wsl --install from an elevated Windows terminal, open the Linux shell, then use the installer above there. Native Windows support remains a tracked follow-up.

Existing checkout

An existing ~/kin or legacy ~/kin-textual always stays an editable checkout. Kin updates it safely and never switches it to a wheel.

Authorized fresh seat

When Git and authorized GitHub SSH access are available, Kin clones the private repository, syncs its locked environment, and installs the checkout editable.

Public product install

If Git is absent, SSH authorization is denied, or GitHub SSH is unreachable, the same command installs Kin from a manifest-selected, SHA-256-pinned Kinra wheel. PyPI is used only for public third-party dependencies.

Dry run: add --check to the installer. It inspects an existing checkout and describes both fresh-install branches without probing SSH, writing known_hosts, or changing the machine.

Verify and begin

Wheel installs leave settings untouched. Checkout installs retain the shared starter baseline. Either way, provider keys are entered in Kin through /providers, not embedded in the installer.

Verify the command

kin --version
kin doctor

kin doctor checks the install, provider resolution, settings, and model endpoint. Use kin doctor --offline when the network probe is intentionally unavailable.

Choose a provider

  1. Run kin in a project directory.
  2. Enter /providers to select a local or hosted model endpoint and add any needed key.
  3. Ask Kin to inspect, explain, diagnose, or change the project. Tool permissions remain explicit.

Read the Kin first-run guide

Update and recover

Kin detects whether it came from an editable checkout or the public wheel and applies the matching update path. Wheel operations pin Kin’s immutable release bytes while using PyPI only for public third-party dependencies.

Apply or check an update

kin update
kin update --check

Repair or roll back

Re-run the installer to repair the current shape. A wheel can install a retained release explicitly:

uv tool install --reinstall "kin @ https://get.kinra.ai/kin-VERSION-py3-none-any.whl" --index https://pypi.org/simple/ --default-index https://get.kinra.ai/simple/

An editable contributor checkout uses its reviewed Git tag or commit as the rollback point.

Contributor checkout

The source repository and its history remain private. Authorized contributors can make the installer choose the development path by configuring a GitHub SSH key, or check out the repository directly.

git clone git@github.com:kinra-ai/kin.git ~/kin

Inside the checkout, use uv run … for every Python command so the pinned environment is authoritative. The public product install does not require repository access.

Deploy the Outpost

Outpost is Kin’s optional always-on deployment surface—not a third Kinra product. Its installer checks Docker and Compose, stages the required secret files, and prints the remaining identity and routing work honestly.

bash <(curl -fsSL https://get.kinra.ai/outpost-install.sh)

Read the Outpost guide

Release artifacts

Same-origin SHA-256 files detect incomplete or corrupted downloads; they do not make the origin itself trustworthy. The scripts are deliberately readable, and version.json is published only after every artifact it advertises.