jev-document-classification
Local-first app that files documents into category folders using Jev decisions on category, confidentiality, subject and injection risk.
Overview
JEV Document Classification is a local-first application that sorts root-level documents in a folder into configured category folders. It extracts readable text locally, then calls Jev through Vercel AI Gateway for typed decisions on category, confidentiality, prompt-injection risk and subject.
Unsupported, unreadable or empty files go to a Not processable folder, category decisions below 0.75 probability go to Need review, and suspicious injection scores go to Suspected prompt injection. The audit records both the actual destination and Jev's original suggestion so reviewers can decide quickly. Files are never overwritten, moves can be previewed and undone, and each run reports duration, input tokens and estimated cost. It requires a Vercel AI Gateway API key.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which configured category does this document belong to? | — |
| Choice | What is the document's confidentiality level? | — |
| Score | How likely is the document to contain a prompt injection? | — |
How it works
- Choose a source folder and extract text locally
- Ask Jev for category, confidentiality, injection risk and subject
- Route low-confidence (<0.75) to Need review and suspicious files to quarantine
- File the rest; preview and undo via the audit
Strengths
- Confidence threshold with a human review folder
- Undoable, non-destructive moves
- Per-run cost and token reporting
Caveats
- Requires a Vercel AI Gateway key
- When probability is unavailable, files go to the selected category without a confidence check
Getting started
`npm install && npm run dev`, open http://localhost:5173, enter a Vercel AI Gateway key and choose a folder.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Local web app with Node server
- Has code
- yes
Use cases and integrations
How curated lists describe it
JEV Document Classification enables the rapid and cost-effective classification of text-based documents using AI, leveraging TypeSafe's "System One" model.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Each document's category, confidentiality, injection risk and subject.
What happens to uncertain documents?
Below 0.75 probability they go to a Need review folder with Jev's suggestion recorded.
Can I undo it?
Yes, the audit supports preview and undo, and files are never overwritten.
Related projects
Chinese-language Zotero AI reading assistant that uses Jev judgments to classify papers in your library.
LibraryproductionJavaScriptPrivacy-first personal knowledge management system, listed as integrating Jev for document classification and tag routing.
PluginproductionTypeScriptClassifies tax PDF pages into 261 IRS forms and 7 page kinds with one Jev request per page and a confidence gate.
ProjectproductionTypeScriptRust tool that asks Jev which PDF pages genuinely need OCR, extracting text pages locally and sending only the rest to OCR.
ProjectbetaRustGo terminal app that files documents using five Jev questions: category, sensitivity, urgency, whether action is needed and action state.
ProjectbetaGoCalibre plugin that asks Jev yes/no questions to suggest genre and subject tags for books, with per-tag thresholds and review.
PluginbetaPython