Kanna 4 for macOS

Build software you understand, at a pace you can sustain, while having fun.

Kanna is a software development platform for working with coding agents. It keeps you connected to your code, engaged in the work, and sane.

Apple Silicon Intel

Requires macOS 15 or later

Kanna's Desk with no task on it: the planning agent on the left summing up what's running and offering a quiz, and on the right the Shelf with three tasks and the backlog, one task blocked — The planning agent beside the Shelf: three tasks on it, the backlog below, one task blocked.

Better tasks, up front.

Most tools hand you a prompt box. In Kanna, the planning agent helps you write better task prompts.

Each task starts with enough detail to go from idea to production code, often in one go. You're still the bottleneck, but for design decisions, not for reviewing PRs.

The planning agent helping write a task: the user asks for a heads-up before king tides, and the agent asks how it should be delivered, with choices, beside the Shelf and backlog — The planning agent turns what you want into a task an agent can finish in one go.

See it before it's built.

Agents are great at toy examples. Hardened code takes longer. So each task starts with a model of the feature: a mockup or a prototype, with a markdown spec beside it. We call it reference-driven development.

02.1Iterate on the model.

A mockup changes in minutes, so you can try ideas fast without building them for real.

02.2Mark it up together.

Comment on the spec, the mockup or the prototype, and the agent refines it with you.

02.3Approve, then build.

Once you approve the model, the agent writes the production code to match it.

A task presenting its mockup of a tide-table day page, with a comment thread beside it: the user asks where the tide is now, and the agent replies with what it added — The model, marked up: you comment, the agent answers and refines it.

Three tasks on your Desk. The rest in the background.

Watching ten agents at once is like drinking ten shots of espresso. Jumping from task to task feels fast, but you're spinning your wheels. Real work takes real concentration, and getting into flow takes time. So Kanna keeps you on one task and keeps the switching to a minimum.

03.1A Desk of three.

The task you're on, and two more on the Shelf. That's the limit, on purpose.

03.2The backlog works while you do.

Agents work through your backlog in the background, and stop when they're done or need you.

03.3Results, presented.

Call a task up to your Desk and its agent has laid out what it did: the spec, the model, the diff, its notes. Agents choose which windows, tabs and docs to show you.

A task on the Desk, its work done: the agent's summary of the fix on the left, the diff it opened to present it on the right — Called up to the Desk: the task explains its fix, with the diff open beside it.

Stay connected to your codebase.

Even when agents write most of it, it's still your code.

04.1Explained on the lines.

Agents explain their changes with notes on the lines, shown in Kanna, never added to your code.

04.2A daily quiz.

Take a break with a daily quiz on how your codebase actually works. Your results are kept, so later quizzes focus on your weak spots.

An agent walking through code: its note above lines 35 to 44 of harmonics.ts, the rest of the file dimmed — A walk through the code, one note at a time.

The quiz window: a spot-the-bug question in the agent's chat, and the code from a missed question open beside it with the explanation on its lines — The daily quiz: miss one and the code opens at the line.

Kanna: enjoy coding again.

Download it for your Mac.

Apple Silicon Intel

Requires macOS 15 or later

Works with Claude Code, Codex and other agent CLIs. Source available under FSL-1.1-MIT: each release becomes MIT two years after it ships.

Source on GitHub →Join the Discord →