Why Kagisecure

Two weeks, about $300 a month of AI, and my first open-source project

For years I admired open source from the sidelines. This is why I finally shipped something.

Note, October 2026: this was written for the first release. Since version 0.6.0 Kagisecure starts in Open mode, which asks for no approvals; approving every use is now Protected mode, one switch away. Newer versions are published as signed apps only while Kagisecure is in development.

I told myself I was too busy — and “too busy” was an excuse. I had never released an open-source project. What changed is AI-assisted coding: for the first time, finishing something solid on my own felt possible.

Why a password manager

So I asked what I could contribute that would matter most. The answer was a password manager. Every time an AI agent helps me code, it eventually needs a database URL or an API key — and pasting a .env into the chat sends the secret straight into the transcript and the model’s context. Convenience, paid for with the one thing I most wanted to protect.

Kagisecure lets agents know which secrets exist and run commands with them, without ever seeing the values. Every use is approved on your Mac.

Offline by design. The only network request is the update check.

Kagisecure works entirely offline. No cloud. No account. No sync server. No telemetry. The one thing the app asks the internet is whether there is a newer version, and you can turn that off. Your passwords live in one encrypted file on your Mac, and nowhere else — there is no server to breach.

Don’t take my word for it: read the source. That is exactly why it is open.

Easy isn’t the same as anyone-can

“Vibe coding makes that easy,” some will say. Partly true. But easy to produce and built properly are different things. I ran several AI agents in parallel on a $200 Anthropic plan, alongside a $100 OpenAI plan, and it still took two weeks.

The design calls — the cryptography, what an agent may and may not see, how approval works — were mine, recorded in 31 architecture decision records, with more than 600 tests on the Rust side and an app notarized by Apple.

It’s early

There will be bugs. So the threat model is public, vulnerabilities can be reported privately, and every line is on GitHub. Try it before you trust it with your most important production keys.

If we grow it together, it can save a lot of people a lot of time — and a lot of worry. Stars, issues and pull requests all help.

brew install --cask itsucara/tap/kagisecure