jotharbourLocal document processing
Reference files are read in the browser. AI is off; no document upload service, API-key entry or hidden model fallback is connected.
OUR APPROACH · CONTROLS AND LIMITS
Trust should be visible.
Security is a design priority. This early public beta keeps the notebook on your device and makes its limits visible. These are current controls, not a claim of certification or perfect protection.
Constrained content
File size, document count and extracted-text limits help prevent overload. DOCX parsing rejects document type/entity declarations. Imported notebooks validate drawing and image fields. Rendered text escapes raw HTML.
Browser protections
The site uses HTTPS. Content Security Policy restricts scripts, document-processing connections and embedded content to the site; data/blob resources support local images, workers and exports. External searches require a separate choice.
Clear data controls
Notebook and solution-sheet exports let you keep a copy. The app can clear stored notebook records. Persistent-storage requests and a seven-day export reminder help you keep a backup routine; neither guarantees recovery. No payment details or API credentials are collected here.
Install from our website
Use our download page to install through your browser. No development source bundle or native executable is distributed here. Keeping code out of a project ZIP does not make frontend code secret: browsers receive the interface code needed to run it. Credentials and private server logic must stay outside public app files.
Know the limits
Local browser storage is not an encrypted vault. Another person using your browser profile could access the notebook. The browser can clear data; make backups. Use your device lock and a separate browser profile on shared computers. Files can still be malformed or misleading; this is not a malware scanner. Don’t use the beta for confidential or highly sensitive records.
There is no AI model in the public browser release. Optional model connections are being tested privately. Models can make factual and safety mistakes. Local processing changes where data is processed; model choice alone cannot guarantee safe output.
Before a wider AI or children’s launch
Planned work includes mailbox triage procedures, safeguarding and privacy assessment for the intended audience, real-device accessibility testing, verified reporting, independent security review, model evaluations and secure release signing. These tasks are not represented as complete.
Report responsibly
Create a local report draft without attaching secrets or original documents. It is not submitted automatically. Send the reviewed draft to jotharbour.support@gmail.com. This remains an early beta; school and child-facing deployment need further review.
Try the local notebook ↗