Thinking out loud.
On audio AI, music technology, and software that listens.
2026-09-05
Don't poll. We'll call you when it's done.
Polling is the wrong default for async audio. Here's why completion webhooks are the ergonomic that changes how you build on audio infrastructure.
Read →
2026-09-02
One platform, two front doors: chat for people, an API for developers
Most audio tools split into two brands for two audiences. We did not. One engine, two ways in: a chat for people and an API for developers.
Read →
2026-08-28
Design a pipeline once, run it 500 at a time
Recipes turn a repeated audio workflow into one reusable pipeline, then run it in bulk with fair concurrency, retry, cancel, and honest errors.
Read →
2026-08-24
You should never pay twice to analyze the same file
Owner-scoped content-hash caching plus transparent per-second prepaid pricing: understand a file once, and don't get charged to understand it again.
Read →
2026-08-20
Storage that keeps your sound, not just a tool that touches it
Most audio tools are stateless: you feed a file in, get a result, and it forgets. Scarleta is where your audio library actually lives, versioned and always there while your account is active.
Read →
2026-08-16
Give any AI an audio toolbelt
An AI other AIs can use. Connect your agent to Scarleta's MCP server and it can separate stems, transcribe, and detect key and tempo, as native tool-calls.
Read →
2026-08-12
Understand a song: key, tempo, and structure in one pass
What does an AI ear actually hear? Key and chords, tempo and beat, song structure, notes, and stems: outcomes you can read, from a single track.
Read →
2026-08-08
We chose not to generate sound: here's why
Everyone's racing to make AI that produces more audio. We deliberately went the other way: an AI that helps you understand the sound you already have.
Read →