jev-mail-classifier
Config-driven IMAP inbox classifier with a terminal UI where Jev answers yes/no per category to tag, move, flag or notify.
Overview
jev-mail-classifier connects to a mailbox over IMAP and lets you define categories in plain language. For each email it asks Jev a yes/no question per category and runs the configured action (tag, move, flag or webhook) when the probability crosses that category's threshold. Everything is configured through a Textual terminal UI and a setup wizard, without writing prompts or code.
It is aimed at people who want rule-like inbox automation without paying chat-model prices or parsing JSON. The README argues that per-category thresholds on calibrated probabilities replace second prompts asking a model whether it is sure. It cites TypeSafe's own benchmark figures for this class of task (about 193x faster and 238x cheaper than a general LLM; 0.114 s and $0.000081 versus 8.566 s and $0.014). Keys from TypeSafe, OpenRouter or Vercel AI Gateway all work.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this category apply to the email (for example, is this an invoice)? | yes · no |
How it works
- Connect an IMAP mailbox and paste one Jev API key in the setup wizard
- Define categories in plain language with a threshold and an action
- Each email gets a yes/no question per category
- Actions (tag, move, flag, webhook) run when the probability passes the threshold
Strengths
- Per-category thresholds, for example 0.7 for urgent and 0.9 for spam
- No prompt engineering or JSON parsing
- Low per-email cost and sub-second latency as reported in the README
- Works with three key providers
Caveats
- Benchmark figures quoted are TypeSafe's own, not measured by the project
- Depends on a hosted Jev API
- Licence not stated in the list data
Getting started
git clone https://github.com/parth-kp/jev-mail-classifier && cd jev-mail-classifier && ./install.shAt a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- Terminal UI (Textual)
- Has code
- yes
- Stars
- 21
Use cases and integrations
How curated lists describe it
Classify your inbox with Jev (TypeSafe's System One model) — tag, move, flag, and notify, all config-driven.
wh000wh000/awesome-jev-liveClassify your inbox with Jev (TypeSafe's System One model) — tag, move, flag, and notify, all config-driven.
hellogumbo/awesome-jevConfig-driven inbox classifier that tags, moves, flags and notifies from typed answers.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide here?
For each email and each user-defined category, the probability that the category applies.
How is this different from an LLM email filter?
Jev returns calibrated probabilities directly, so you set a threshold per category instead of parsing generated text.
What do I need to run it?
An IMAP login and one API key from TypeSafe, OpenRouter or Vercel AI Gateway.
Related projects
Gmail organiser where readable rules do most filing and Jev judges only what rules leave, with receipts and undo for every change.
DemobetaTypeScriptMac app with AI bots running business ops on cloud computers, using TypeSafe for small yes/no calls like risk and which bot takes a job.
ProjectproductionLocal open-source work assistant that turns mail and tickets into agent tasks, with an optional Jev condition-check module.
ProjectbetaPythonOpen-source AI email assistant; listed as using Jev as an optional classifier for email intent and triage.
ProjectproductionTypeScriptGlue that turns one Linux machine into a team of coding agents, with Jev making typed calls like who builds a task and whether it may merge.
ProjectbetaTypeScriptAgent harness for long-running jobs that survive across sessions; listed for a Jev-backed table that routes requests to models.
ProjectbetaTypeScript