Terminal-native AI peer

Kin

Work with an AI peer in the files and tools you already use.

Kin is a terminal-native agent harness for real work in a real filesystem. It keeps the person close to the work while giving the peer enough context, tools, and bounded authority to contribute well.

Available now

Install Kin Read the docs

Kin

The working environment stays visible: workspace, model, session, posture, and the next move.

Where it fits

The working peer

People who want an inspectable, local-first way to do serious work with AI without moving the work into a separate chat product.

Capabilities

What it makes possible.

Works where the work lives

Read, search, change, run, and verify inside an ordinary filesystem and toolchain.

Keeps sessions durable

Resume work with its decisions, context, and consequences intact.

Delegates bounded work

Use skills, subagents, workflows, and scheduled work without dissolving supervision.

Makes authority visible

Modes, permission gates, sandboxing, and review surfaces keep action legible.

Working sequence

A clear path through the system.

01
Enter a real project
Start Kin from the directory that contains the work.
02
Ask for an outcome
Give direction and context, then let the peer make a concrete shape.
03
Inspect and react
Review what changed, name what is wrong or alive, and cycle again.

Boundary

What this page is not claiming.

Peer is an engineering stance, not a claim about sentience.
Respect does not lower the evidence bar; important claims and changes still require verification.
Outpost is an optional always-on companion, not a second window onto the same running Kin.

Go deeper

Use the product page for orientation. Use the docs for decisions.

Read Kin documentation