The most common objection we hear to workplace AI is "we can't send our data to the cloud." We built MonitorMe to show that, mostly, you don't have to.
MonitorMe
Your working day, journalled by AI — without your data leaving the machine.
The Problem
Knowledge workers lose the thread of their own day. Timesheets are fiction written on Friday; standup updates are archaeology; "what did I actually do on the Henderson project?" has no reliable answer. The obvious fix — software that watches your screen — usually means streaming screenshots of everything you do, including client data and credentials, to a vendor's cloud. For most businesses that's a non-starter, and it should be.
- —Recall gap: real work is scattered across apps, tabs and interruptions nobody logs.
- —Privacy wall: cloud screen-recording tools are unshippable in any business with confidentiality obligations.
- —Trust deficit: "AI assistant" too often means "data leaves the building".
What We Built
A local-first Electron app for Windows 11 that treats privacy as an architecture, not a settings toggle:
- —Screen context capture — periodic captures plus clipboard and browser-URL context, stored in day-partitioned folders on your machine.
- —On-device OCR — Windows' built-in WinRT OCR reads the text out of captures locally. No cloud key required for any of the capture pipeline.
- —Encryption at rest — captures, OCR text, clipboard data and API keys are AES-256-GCM encrypted on disk. Lose the laptop, lose nothing readable.
- —AI journalling — an LLM (via OpenRouter, called only from the app's main process) turns the day's context into a structured journal. The one deliberate cloud touchpoint, under your key, with your choice of model.
- —SQLite storage — everything queryable, everything yours, nothing phoning home.
The Discipline
- —Security-audited — a formal static security audit is part of the repo, and its findings drove the encryption and IPC design.
- —Honest about maturity — this is a working demonstrator we run ourselves, not a boxed product. Capture, OCR and encrypted storage are complete; the inference layer is the active development front.
Delivery timeline
Electron shell, capture-core package, SQLite, encrypted settings.
screen harvest, WinRT OCR, context collection, day-partitioned encrypted artefacts, soak-tested capture guardrails.
OpenRouter model registry, journal generation, review UI.
What it proves
"Local-first" is our answer to the most reasonable objection in AI adoption. MonitorMe demonstrates the pattern we sell to privacy-bound clients: keep capture and storage on-device, encrypt everything, and spend cloud tokens only on the one step that needs them — under your key, on your terms.
We design local-first AI systems for businesses whose data can't leave the building.
Book a Reality Check →