dacular

A private document vault on your Mac.

Index your own PDFs, CSVs, and notes and ask open-ended questions — "how much did I spend on travel last year?", "when does my insurance renew?" — answered on Apple Silicon, with the documents never leaving the machine.

The trick is headgate: the powerful frontier model is an untrusted code generator, not a data processor. It sees only the shape of your vault and writes a program; the program runs locally in a sandbox that can't phone home, and a small on-device model is the only thing that ever reads the real text. Your data never reaches the frontier model.

⚠︎ Experimental. dacular is an early research project — the design and APIs are still moving and it isn't ready to trust with data that truly matters yet.

get started install

Install dacular on your Mac, index a folder of PDFs/CSVs/Markdown, and ask your first question — one command per step.

walkthrough example

A start-to-finish example: drop a folder of bank statements on your Mac and ask "how much did I spend on travel last year?" — answered locally, the data never leaving the machine.

headgate the harness

The privacy harness underneath: a frontier model is treated as an untrusted code generator, not a data processor. It writes code; the code runs locally in a sandbox that can't phone home.

deep dives under the hood

How a question becomes a sandboxed program the frontier model writes but never gets to run against your data — with the actual generated Mojo.

the engine millrace ↗

dacular runs on millrace — a from-scratch, pure-Mojo GPU inference engine for Apple Silicon. The local model that reads your real data is served there.