Will it delete mail or bury things in spam on its own?

It never deletes and never moves mail to Spam or Trash. Archiving only removes the inbox label, only for labels you mark archive-eligible and only above your archive-confidence threshold.

Why classify metadata first instead of the full message?

Metadata plus snippet is cheaper and sufficient for clear cases. Ambiguous or archive-sensitive mail gets a second pass on full text, and every decision is saved before Gmail is updated.

How do I keep costs predictable across thousands of messages?

Pricing is per input tokens with free output, and the dashboard shows cost, time and progress. You set a maximum run spend checked before each request, plus a provider-side key limit.

Can I rely on it for a non-English inbox?

Other languages are accepted but English is the primary training language with the strongest accuracy. Test criteria and thresholds on representative non-English mail before automating.

Where does it run, and what does it have access to?

It runs from a Google Apps Script project in your own Google account, so there is no jevMail account and no hosted mailbox database. You choose the scope, processing the Inbox or All Mail with optional limits, and each message is evaluated independently. Successfully processed messages get a technical marker so later runs skip them, which is what keeps repeat runs cheap.

How is the run budget enforced?

The budget is checked before each request using a conservative estimate, and when a request times out the behaviour is described as an error rather than a silent retry. Because Jev is priced on input tokens with no charge for output, high-volume classification stays practical, and the UI shows processing cost, elapsed time, progress and errors alongside the decisions themselves.

What data does TypeSafe say it does not do with my mail?

The README cites TypeSafe's own position that Jev is not fine-tuned on customer data and that customer requests and responses are not used to train the model. That is a vendor statement rather than an audit, and the same README is direct that confidence is a control signal and not a guarantee that any individual decision is correct, which is why it tells you to start in preview and tune criteria against your own inbox.

CommunityJev Tools & CLIs

Jev Gmail AI Spam Filter And Labeling

Self-hosted Google Apps Script that classifies Gmail by meaning with Jev Choice over your own dozen-label taxonomy, checking metadata first and full content only when confidence demands it.

See highlights

Open github.com

Stars: 23
Language: JavaScript
License: MIT
Published by: Community
Author: ilyamk
Last verified: Oct 2026

ilyamk/jev-gmail-ai-spam-filter-and-labeling

User repository on GitHub

View on GitHub

JevMail - Self-hosted AI email classifier for Gmail powered by Jev. Create custom labels, organize your inbox, and filter spam with confidence and cost controls.

Forks: 7
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

Semantic email classification powered by Jev, with confidence-aware automation, cost controls, and no jevMail-operated backend. Why jevMail · How it works · Label playbooks · Visual guide · Setup · Privacy · FAQ.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Moderation

Used inJev for Software & DevTools

Languages: JavaScript

Highlights

Up to twelve plain-English labels with boundaries plus a safe review bucket for ambiguity.
Two-stage Jev Choice flow with confidence-gated full-content review and resumable markers.
Preview mode, per-label archive eligibility, run budgets and visible decision provenance.
Personal Apps Script deployment with no operator mailbox database and explainable actions.

Watch out

Confidence guides automation but does not guarantee correctness; start in preview, keep archiving narrow, and tune criteria on your own mail. English gives the strongest accuracy and provider charges can differ from estimates. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-recall

Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.