Jev Voice Agent
Run a real-time voice agent from prewritten replies and Jev choices.
What Jev decides
Which fixed reply label fits the latest caller turn.
Deepgram handles speech input and output while Jev chooses one approved reply label from a fixed script without an LLM in the loop.
- Jev role
- Choose the next prewritten spoken reply.
- Input
- The latest transcript, conversation state, and allowed reply labels.
- Output
- A reply label and its Choice probabilities.
- Decision frequency
- After each completed caller turn.
- Fallback
- The agent can only speak human-authored lines; open-ended answers are outside this design.
How it works
Overview
Deepgram Flux detects spoken turns and transcribes them. Jev reads the transcript and selects a reply from human-written lines, then Flux TTS speaks that line. This keeps replies inside a fixed script but limits open-ended conversation; the repository includes offline recorded scenarios and a section on when this design stops fitting.
- Language
- TypeScript
- License
- MIT
- Author
- dg-coreylweathers
Sources
Related projects
Textured
Turn a sound description into synthesizer controls through Jev judgments.
What Jev decides
Seven ordered sound-quality scores and probabilities over eight sound families.
typesafe-computer-use
A native macOS computer-use loop built from local OCR, accessibility data, and bounded Jev actions.
What Jev decides
Which computer action, visible item, site, or off-screen accessibility control to use next.
Jev Chat Assistant
An Android conversation copilot that uses Jev to interpret chats and rank reply candidates without sending for the user.
What Jev decides
The likely intent, conversation risk, reply need, best next action, unresolved tension, and which of three candidate replies fits best.