File local documents with typed safety lanes

01 · Role in the system

What Jev does here

Text extraction happens on the machine; only the readable text goes to Jev through Vercel's AI Gateway as typed questions. Each document gets a Choice for the best configured destination folder - validated against the configured list, so an out-of-list answer throws - plus Choices for confidentiality, subject, and a prompt-injection risk score on a 0-100 scale in increments of ten, with criteria that refuse to flag ordinary quoted instructions or technical discussion lacking adversarial context. Outcomes are lanes, not actions: unsupported or unreadable files go to Not processable, low-confidence category answers to Need review, suspicious scores to Suspected prompt injection. Files are never overwritten; an audit log supports preview and undo, and every run reports duration, input tokens and estimated cost.

02 · Control boundary

Where Jev sits

Jev as the filing clerk for a local document pipeline: typed Choices decide folder, confidentiality and injection risk against configured criteria, validation rejects out-of-list answers, and uncertain or suspicious documents route to review lanes with full audit and undo.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Folder categories are operator-configured; classification quality tracks the criteria you write.
• Runs through Vercel's AI Gateway rather than the native endpoint.
• Small project (6 stars at review time); desktop scope, no bulk server mode.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• Charlyhno-eng ↗Community · github · public · checked 2026-09-30