Trust
What you actually get.
Four honest answers.
Kajuto looks at your repos and helps you delete them — that's a matter of trust. A fair deal needs honest answers to the hard questions, not just the easy ones. Here are mine.
Why mac-only — and forever?
Kajuto sits deep in macOS APIs that simply don't exist elsewhere: SwiftUI MenuBarExtra with Liquid Glass, NaturalLanguage for the on-device embeddings behind finds, strict Swift 6 concurrency, UserNotifications for the morning greeting.
Going cross-platform would mean building three separate products — each one further away from the system. The result would be a kobold that's mediocre everywhere instead of excellent on one platform.
Your guarantee
Cross-platform is not on our roadmap, not parked in 'Future / aspirational', and not 'maybe in 5 years' — it's a deliberate non-goal. Mac-only by design. Forever.
Which data goes to AI — and to which one?
Honestly first: Kajuto is NOT 100% local. For analyses the app sends project context to an AI endpoint: README, package manifest, and a git-log summary. No full codebases, no secrets — but more than "nothing". Steering proposals and the cleanup inventory, on the other hand, run without any AI, locally and deterministically.
Which endpoint that is, you decide: by default Claude Code (Anthropic) through your own subscription via claude -p — no API key needed. Alternatively DeepSeek via jcode or any OpenAI-compatible endpoint. The call goes straight from your Mac to the endpoint — there is no moinsen.dev server in between recording anything.
Two more paths, both started by you: when you share a link as a find or a Zuruf, your Mac fetches that one page (for YouTube: title, channel and description) so the AI knows what it is about. And a Zuruf task without a project (“look into …”) is done with a web search — only the sentence you said goes to Claude, no project knowledge.
What stays local: all scores, reports, and decisions (cache in .pumuckl right inside the repo), your finds including embeddings (on-device, Apple NaturalLanguage), and the portfolio knowledge. GitHub access runs through the gh CLI with your own login — Kajuto never sees a token.
Your guarantee
The AI endpoint is switchable at any time, no lock-in. There is no Kajuto cloud collecting your project data — we can't sell or lose what we never have. And Kajuto never deletes anything in your cloud accounts itself: it only shows you what's still there.
Why isn't Kajuto open source?
In 2026, open-source maintainers are flooded with AI-generated pull requests — most well-intentioned, few well-crafted. Maintainer burnout from PR triage is the most common cause of small OSS projects dying.
Kajuto is built by a solo developer who funds the next features from the €29 one-time price. Going open source would mean spending 30% of the time on PR reviews instead of code — at the expense of what makes Kajuto better, i.e. at your expense. That said: the .pumuckl data format is plain JSON in your repo — your analysis data belongs to you in readable form.
Your guarantee
The open-source question is publicly re-asked on a regular basis (in the public community repo on GitHub). When market reality or maintainer capacity changes, the answer changes — and you'll hear it first.
What does "no telemetry" mean for an AI app?
Many apps write "no telemetry" on their landing and still ship data — crash reports, anonymous usage stats, remote feature flags, A/B tests. "No telemetry" is interpreted very generously in the industry.
At Kajuto, "no telemetry" is literal: the app reports nothing about you or how you use it. It only connects where you start something yourself — the AI endpoint you configured; the GitHub API (via the gh CLI, with your own login); web pages you share as a link; your own private iCloud database when you use the iPhone and Watch app; and TesterPayKit when you send a report through “Feedback…” yourself. There is no Kajuto account, no license-server check (the app verifies the license offline), no analytics endpoint, no Sentry. This landing page itself sets no cookies either.
Set straight, honestly: up to version 1.3.0 the app (then still called Klabauter) opened a test session with TesterPayKit at launch and reported scan errors including project names — a leftover from its own development that never belonged in the shipped app. From version 1.3.1 that is gone: without your click on “Feedback…”, Kajuto does not talk to TesterPayKit.
What we specifically did not build
No Sentry, no crash reports (you can attach crash logs to a GitHub issue manually). No analytics, no "anonymous" usage stats. No remote feature flags — what you installed is what's in the binary. No A/B testing. This list is append-only — if anything ever joins it, it will be announced publicly in the GitHub repo before release.
Still have a question?
Missing an honest answer that should be here? Open an issue on the repo — I usually reply within 24-48h.
Ask on GitHub