Call Coach
Floating live-call coach that transcribes locally and asks Jev after each sentence what to do next, shown with a confidence gauge.
Overview
Call Coach listens to a live conversation, transcribes it locally with Whisper, sends the text to TypeSafe Jev after every sentence and shows what to do next with a confidence score. It ships nine coaches: five for live calls (sales, customer service, job interviews, social cues and anxiety) and four for solo practice including dating, and you can build your own in Settings.
It runs as a small floating window over your call on Windows, or as a local server and website where visitors use their own key, which passes through without being stored. Only text goes to TypeSafe. Practice mode scores spoken answers and names one thing to work on. The author calls it a work in progress and a demo of Jev's capabilities.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What should the rep do next?(inferred) | coach-defined next steps |
| Score | How well is the call going / how good was this answer?(inferred) | — |
How it works
- Transcribe speech locally with Whisper
- After each sentence, send the text to Jev with the coach's questions
- Show the suggested step and confidence gauge
- Repeat throughout the call
Strengths
- Audio stays local; only text is sent
- Custom coaches
- Per-sentence guidance
Caveats
- Work in progress
- Needs a TypeSafe key
Getting started
Run the Windows installer or TYPESAFE_API_KEY=your_key node server.mjsAt a glance
- Type
- Project
- Status
- prototype
- Language
- JavaScript
- Runtime
- Windows desktop overlay / Node server
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Listens to a live sales call and, after each sentence, tells the rep what to do next with a confidence score.
MrJev/awesome-jev
Frequently asked questions
Is my audio sent anywhere?
No, transcription is local; only text goes to TypeSafe.
What coaches are included?
Sales, customer service, interviews, social cues, anxiety and practice coaches.
Can I build my own coach?
Yes, in Settings.
Related projects
Marketing and voice-agent platform with an experimental TypeSafe module that picks specialisation labels for agent roles.
ProjectprototypeSelf-hostable Reddit buyer-intent finder that has Jev judge every post and comment it reads, with a written reason and cost per lead.
ProjectproductionTypeScriptScores HeyReach LinkedIn leads and first-message variants with Jev, picks a message per lead and writes draft campaigns back.
ProjectbetaPythonmacOS WeChat helper where an LLM drafts replies and Jev classifies intent, rates risk and ranks the candidates; you press send.
ProjectprototypeSwiftCombines Jev judgments on row text with a tabular foundation model like TabPFN to predict outcomes from few labels.
LibraryprototypePythonProof of concept that scores replayed support-call transcripts live against a weighted scorecard using OpenAI Decisions or Jev.
AppprototypeTypeScript