Two weeks of quiet watching beats six months of workshops.
Here's exactly what Jottings does to the machines on your team's desks - and what it refuses to do.
- Phase 01
Install.
A consultant or an internal lead ships a signed installer to consenting employees. A menu bar icon lives in plain view with a pause button and a visible countdown to the auto-uninstall date.
Code-signed for Mac (Apple Developer) and Windows (EV cert). Auto-updates via electron-updater. Silent installs are refused by design.
- Phase 02
Observe.
For two to four weeks the agent logs which apps are open, for how long, what sequences repeat, and where attention actually goes. Not keystrokes. Not content. Workflows.
Capture modules: active-window, app name, URL domain, duration, app-transition, aggregate click and keystroke counts, and screenshot OCR. Configurable per engagement in the consent flow.
- Phase 03
Redact.
Every observation runs through a local redaction pipeline before a single byte leaves the device: regex for SSN / CC / phone / email / address / DOB / plates / routing numbers, on-device NER for everything else, OCR redaction that discards raw bytes, and a per-app allow/blocklist that ships pre-configured to block 1Password, banking sites, personal Gmail, and medical portals.
≥99.5% recall on our synthetic PII suite. Full redaction audit log is generated per engagement and handed to the client.
- Phase 04
Synthesize.
Redacted observations get fingerprinted, clustered, process-reconstructed into a work graph, scored for dollar impact and complexity, and matched against the Skill Library.
Fingerprinting is structural; clustering is embedding-based; sequence mining uses long-context models on redacted transcripts only. Raw bytes never reach a model.
- Phase 05
Report & Purge.
A branded, editable PDF lands in the consultant's workspace - ready for client delivery in under ten minutes of editing. Observation data is purged within 90 days of engagement close. The agent uninstalls itself on the end date.
Every deleted record is hashed into an attestation log you keep. If a lawyer ever asks what we had, you can prove it no longer exists.
SidebarWhy desktop, not a browser extension.
Real work lives in Excel, Outlook, Salesforce, Figma, forty internal tools, PDF attachments, and six chat apps. Browser extensions see maybe thirty percent of it - and not the thirty percent where the dollars are.
If your observation layer cannot see the work, your recommendations have no evidence underneath them. That is the entire reason log-based process-mining recommendations feel thin: the evidence is thin.
Desktop observation is harder to build and harder to sell. It is also the only surface where the full work graph is visible. We picked hard on purpose.
Five guaranteesWhat we never do.
- We never capture keystroke content.
We capture aggregate counts and app context. The words you typed never leave your device.
- We never upload raw screenshots.
OCR runs locally, text is redacted locally, and the pixels are discarded before anything hits the network.
- We never track individuals.
The unit of analysis is the workflow. Individual identifiers are hashed and unrecoverable by the time the report is rendered.
- We never run outside the engagement window.
The agent has a hard end date in its own menu bar. On that date it uninstalls itself. No pet-door subscription.
- We never share your data across engagements without k-anonymization ≥5 orgs and explicit Skill Dojo opt-in.
Your raw observations stay yours. Only aggregate patterns, with explicit consent, flow into the public Skill Library.
One PDF. Ten opportunities. Ranked by dollar impact.
Every scan produces the same report format. Cover, logo, and palette swap for each consultant or team. Top-ten opportunities each get their own page: evidence links, dollar math, complexity score, recommended tool stack, estimated payback.
Anatomy of the reportYour first scan tells you more than your last six months of workshops.
Code-signed · auto-uninstalls on end date · opt-out any time.