Skip to main content

Why Keel

Keel is a turn-based board engine for human/AI delivery teams. It is designed for teams that want planning, execution, and verification to live on one visible operating surface instead of being split across chat, tickets, scratchpads, and private memory.

This documentation starts in plain language on purpose. You do not need to know Keel's native vocabulary before you can understand what the system is trying to accomplish.

Turn-Based

Work advances through explicit moves

Instead of throwing tasks into a general pile, Keel expects each move to change the board in a visible way.

Human And AI

The operating surface is shared

People and agents read the same planning artifacts, work the same states, and close the same acceptance criteria.

Evidence-Backed

Completion requires proof

Stories close with recorded evidence, not just confidence. Keel treats missing proof as drift, not style.

What "turn-based" means in practice

Most tools assume that more activity means more progress. Keel assumes the opposite: when the board is blurry, the team slows down even if everyone is busy.

A turn is a bounded move. It begins with orientation, advances one scoped slice, and ends with proof. That rhythm is what keeps human and AI collaboration legible.

Translate first, then learn the board terms

Keel uses a native vocabulary because it wants different types of work to mean different things. A mission is not just a bigger story, and a routine is not just a repeating ticket.

The docs will translate from familiar concepts into Keel terms as you go.

What Keel does and does not own

Keel owns the board contract:

  • what work exists
  • what state each unit is in
  • what constraints were planned up front
  • what proof is attached before closure
  • what recurring work may be materialized automatically

Keel does not try to replace the rest of your toolchain:

  • your editor or IDE
  • your preferred AI harness or model vendor
  • design tools
  • source control hosting
  • deployment runtime

That boundary is deliberate. Keel is the operating layer, not the entire stack.

That rule still holds when Keel is embedded inside a larger runtime. Keel continues to own planning truth. A runtime such as Keeper can own provider ingress, routing, execution, and replay, but it should lower external work through a formal mission-request boundary instead of inventing board state out-of-band.

If you are integrating Keel with GitHub issues or a multiplayer runtime, read Keeper Boundary And Mission Requests.

Read the docs in this order

  1. Install the CLI and verify your local path.
  2. Scaffold a project and take the first turn.
  3. Learn the CLI interface, pacemaker contract, and scene surfaces.
  4. Learn the board model, Keeper boundary, lanes, and verification loop.
  5. Move into the persona track that matches your job.