Skip to content

kin

A native Python agent harness with a Textual terminal UI. This site is the user guide: how to install it, point it at a model, and drive it day to day.

What kin is

kin is a native Python agent harness — the model client, agent loop, tools, permissions, and modes — that runs in-process, with a Textual terminal UI as one client on top of it. The harness lives in kin.harness; the UI lives in kin.tui and talks to the harness over an in-process event seam rather than a subprocess or a wire. The identity is calm and cool by design: a graphite ground where kin's live voice is cyan and its thinking is a recessive periwinkle, so a precision tool reads as one.

Quickstart

From a checkout, three lines get you from nothing to a running session:

uv sync     # create the venv and install the harness deps + the `kin` command
task run    # launch the Textual UI against the in-process harness
# uv run kin   # the raw equivalent of `task run`

You still need to point kin at a model. Start with Install, then First run for the model recipes, then Configuration to make your setup stick.

The shape

The harness is src/kin/harness/, the Textual UI is src/kin/tui/, and an in-process event seam carries everything between them — see Architecture for how the seam and the turn worker fit together.

Where to next

Three starting points — pick whichever matches where you are:

  • Just installed? First run walks the model recipes (vLLM / Anthropic-direct / OpenAI-direct / preset). Install for prerequisites; Configuration to make your setup permanent.
  • First time driving the TUI? First session is a scripted 10-minute tour (welcome banner, composer, slash history, cursor mode, planning freeze, sessions).
  • Setting up an operator surface? Models & providers for the wires + heuristic, Modes & permissions for the mode × tool gate, MCP servers for stdio / HTTP / OAuth / elicitation / resources, Outpost for the browser-terminal + scheduled-jobs surface.

The rest is in the left sidebar — but the most-clicked pages are: