A worker apart from the UI
Every agent run will go through a worker, and the UI never touches a provider process. Closing the app leaves the job running, and the approval policy only has to live in one place.
16 / System design · solo · no code yet
The design for a personal operating system that sits above the IDE, Obsidian and the calendar, so notes, plans and coding agent runs share one data model. Its first user is Tom, whose ideas, code, tasks and the reasons behind each decision sit in too many places. For now it is a design document. Nothing runs yet.

Agents can propose freely. Anything important waits for a person to approve it.
Every agent run will go through a worker, and the UI never touches a provider process. Closing the app leaves the job running, and the approval policy only has to live in one place.
Each run will export tracked files to a separate copy, record a hash, clear the environment and return a patch. The design says plainly that this stops accidental edits to the workspace but does not prove the CLI cannot read elsewhere, so "strictly contained" is a label earned by passing tests.
Sync will be end-to-end encrypted. Full prompts, raw tool output and source code stay on the device by default. The price is that a lost recovery kit cannot be restored from the server.
Agents will be able to propose a schedule, but nothing is written until a person approves it, and only to a calendar of its own. An approval is tied to that exact content, so a change means approving again.


Build a product together, distribute one that exists, commission software, or join in — write a few lines and NYX will get back to you.
Send an invitation →