Jevthoven
Prompt-first music studio where Jev makes every compositional choice, from plan to each bar pattern, rendered as editable MIDI.
Overview
Jevthoven is a symbolic music studio. You describe a piece in one sentence, such as a wistful waltz with soft keys over round bass, and Jev decides every musical unit in sequence with a bounded rolling context of recent bars, motif and harmony plan: the overall plan (meter, tempo, form, lanes), instrument choice, harmony, phrase intent, per-lane groove parameters and complete-bar patterns, one Choice call each. Code renders each pick into notes. There is no offline composer, audio model or hidden song library; every note traces back to a stored Jev decision or a labelled manual edit.
The result is an editable multitrack composition with a piano roll and arrangement tools whose deterministic edits never call the model. Plain-English revisions such as 'make the bass busier in the second half' are routed through Jev, and regions or lanes can be regenerated as resumable jobs. Playback uses Tone.js, and projects export as JSON or Standard MIDI. A fixture mode allows development without credits; live mode uses roughly 75 to 80 requests for a 16-bar, 4-lane piece.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What plan (meter, tempo, form, lanes) fits the prompt? | — |
| Choice | Which instrument should this lane use? | — |
| Choice | Which complete-bar pattern comes next for this lane? | — |
How it works
- User writes a one-sentence music prompt
- Jev chooses plan, instruments, harmony and phrase intent
- For each lane and bar, Jev chooses a pattern given recent context
- Code renders choices into notes; user edits deterministically
- Revisions and regenerations are routed through Jev as resumable jobs
Strengths
- Every note is traceable to a decision
- Deterministic editing without model calls
- MIDI export and local SQLite persistence
- Fixture mode for credit-free development
Caveats
- Live generation costs about one request per lane-bar
- Output limited to the pattern vocabulary offered
- Requires Node 22.16+ and a TypeSafe key for live mode
Getting started
cd app, cp .env.example .env (set TYPESAFE_API_KEY or use fixture mode), npm install && npm run build && npm start, then open http://127.0.0.1:4318. One-click install via Pinokio is also offered.At a glance
- Type
- Project
- Status
- beta
- Language
- TypeScript
- Runtime
- Node server and browser app (Tone.js)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Turns a music prompt into editable multitrack MIDI by asking Jev to choose instruments, harmony, and bar patterns.
logicrw/awesome-jev-projectsSymbolic-music studio where Jev chooses plans, instruments, and bar patterns, and code renders editable music and MIDI.
cobanov/awesome-jev
Frequently asked questions
Does Jev generate audio?
No. It chooses musical units and patterns; code renders them into notes and MIDI.
How many API calls does a piece take?
About 75 to 80 requests for a 16-bar, 4-lane piece, per the README.
Can I edit the result?
Yes, with a piano roll and arrangement tools that do not call the model.
Related projects
Emoji autocomplete that asks Jev eight parallel questions per typing pause: emoji choice, emotion, mood, urgency, sarcasm and more.
ServiceprototypeTypeScriptExperiment where Jev turns a one-sentence brief into a coherent design system for a whole dashboard console.
AppprototypeTypeScriptGenerative painting canvas where Jev chooses paint materials and palettes for regions of 105 procedural sketches.
ProjectprototypeTypeScriptFigma plugin that splits a one-sentence design instruction into small Jev choice questions, then code draws the result.
PluginprototypeTypeScriptCan a System One model steer music? Jev picks only enum labels for a composition plan; code writes, engraves and plays the notes.
AppprototypeTypeScriptWeb demo that turns one sentence into six room settings, with Jev choosing each setting and the app rendering the room.
Appprototype