Skip to content
Spec Kitty

Decision layer for autonomous software delivery

Your AI agents write the code. Your people make the calls.

Coding agents now make thousands of small decisions at machine speed. Spec Kitty makes sure they build with what your experts know and what your team decided.

Start free with the CLIBook a demo
  • Free and open source (MIT)
  • Works in Claude Code, Codex, Cursor, Copilot and 20+ more
  • Your models, your keys
GitHub stars
1,677
CLI downloads in the last 6 months
28.1k
Open-source contributors
100+

As of 8 Oct 2026. Sources: GitHub, PyPI.

The problem

Red button or blue button?

Every real system is thousands of red-or-blue calls, usually made by whoever is at the keyboard, and now by AI agents at machine speed. Deciding them well is still a human job, and most teams have no system for it.

  • Problem 1

    Agents never read the wiki

    Your standards and past decisions live in people's heads and in wikis agents never read. The same ticket comes out differently depending on who picks it up.

  • Problem 2

    Decisions made in chat get lost

    When behavior changes, nobody can show who decided it, why, or what it was checked against. The reasoning stayed in a chat window.

  • Problem 3

    Agents grade their own work

    The agent that wrote the change is the one that says it's fine. Review becomes a rubber stamp at machine speed.

What Spec Kitty does

Know what you decided. Decide what applies. Build with it.

Start with the free CLI today; add your team's knowledge when you are ready.

  1. 1 · Know

    Contact us

    Brain Kitty

    Each person curates their own knowledge graph of the standards, runbooks, rules and decisions they own. Answers cite it, and people approve every change.

    See Brain Kitty

  2. 2 · Decide

    Early access

    Team Kitty

    The hosted team product. It collects the team's Brain Kitty graphs so the team decides together what applies, then serves it to Build Kitty over MCP, the standard way coding agents load outside knowledge.

    Request early access

  3. 3 · Build

    Free, MIT

    Build Kitty

    The free Spec Kitty CLI. Agents spec, plan, build and review in the coding agent you already use, with the knowledge Team Kitty serves over MCP and checks an agent can't skip.

    See how it works

Work solo for free. Bring in Team Kitty when you invite your team.

Brain Kitty

Years of how you work, in a form agents can use.

Each person curates their own Brain Kitty: a knowledge graph of the standards, runbooks, rules and decisions they own. Every answer cites its source.

  • Ask. Before the decision: what was already decided, cited, the same answer every time.
  • Check. After the work: new work reviewed against what was decided, with the evidence quoted.
  • People approve. Brain Kitty proposes and informs. It never silently changes what was decided.
Contact us about Brain Kitty
Brain Kitty knowledge graph: 31 dimensions, 65 plays, 8 decision rules and 24 rulings drawn as linked nodes, a ranked list of matching nodes, and an agent chat panel answering a question about Spec Kitty and citing the ruling it used.
Brain Kitty: your standards and rulings as a graph, with an agent answering from it.

How it works

From what your people know to merged, reviewed work

  1. 01

    Each person curates their Brain Kitty

    The standards, runbooks, rules and decisions you own, in your own knowledge graph.

    People: You decide what goes in and approve every change.

    Agents: Brain Kitty proposes updates; it never changes what was decided on its own.

  2. 02

    Team Kitty combines them

    Team Kitty brings everyone's Brain Kitty together and serves it to every coding agent over MCP.

    People: The team decides together what applies to its work.

    Agents: Agents load only what the team chose, and only the slice each step needs.

  3. 03

    Provide answers, read the spec

    Build Kitty asks what to build and how, with your team's knowledge loaded.

    People: You answer, then review and edit the spec before any code.

    Agents: Agents write the spec, plan and work packages.

  4. 04

    Agents build in parallel

    Each work package runs in its own git worktree, so agents don't collide.

    People: You schedule the work packages like any other work.

    Agents: Agents build, each in its own worktree.

  5. 05

    A second reviewer checks, you merge

    The work is checked against the spec and your team's knowledge before it lands.

    People: You merge. Nothing lands without a person.

    Agents: A reviewer that isn't the author checks the work.

From a first workshop

“A ticket our own engineers had sized at ten days went out the door in about five hours, during the workshop, on our first time using it.”
Engineering leadership, legacy Java ERP platform One team's result on one ticket, not a benchmark. See the rollout workshop

Who it's for

Pick your starting point.

Your knowledge. Your decisions. Agents that follow.

Start free with the CLI, or book a demo. Our rollout packages leave working software behind.

Start free with the CLIBook a demo

Rollout packages