Type a prompt.
Get a music track.
Describe the music you want in a sentence — the API returns an original track you can play, keep, and publish.
Price: a 3-minute track ≈ $0.18 — billed per second of generated audio ($0.001/s; $1 buys 1,000 credits). A failed job costs nothing. Accounts are invite-only and start with a free $1 bonus (≈ 16 minutes of audio).
Music generation is live today. Images can be uploaded and arranged alongside audio, but image generation isn't served yet.
Agent quickstart →
A single self-contained prompt: paste the whole file into a coding
agent in an empty directory, hear generated music in < 5 minutes.
API reference →
Everything, with examples and response shapes. Agents: fetch
/docs/api.md or /llms.txt directly.
Public feed →
Published posts, playable lanes, comments — no account needed.
Examples →
curl walkthroughs, a studio, a feed reader — all against the live
API.
Novel →
Write a book chapter by chapter against the live API: a cap that
bounds the whole cost, one permit at a time, commit each chapter to the
library. A chapter that never lands is free.
Reader →
The reading half of the same book: contents, chapter, semantic
search — and nothing on the page that can generate anything.
Console →
A hand-operated probe over the live API.
In one breath
- Generate music from a prompt (async jobs, per-second credits; keep only the takes you like).
- Store content-addressed: identical bytes stored once, your row is your projection of them.
- Arrange into posts — opaque JSON the server never interprets; one post can render as a multitrack editor, a gallery, an article.
- Publish & discuss — anonymous reads for public posts; comments follow the post's visibility.
- Invite-only — registration spends a referral code; every account is accountable through its chain.