Live web app

AlgoRythmic

A playlist builder for YouTube Music that treats a playlist as a rule set you can edit rather than a black box. Describe a vibe in plain English and it compiles to selection rules you can actually see and change: era, length, no live versions, never the same artist twice. Or start from one song and weight what "similar" is allowed to mean, across eight dimensions. Preview the whole queue in the browser for nothing, then save it once as a real private playlist your car can play.

Open music.corticorp.com Back to Projects
A plain-English description compiles to editable rules, which generate a queue you preview for free, which you then save to YouTube Music

How it works

Describe. Edit. Preview. Save.

Rules, not a black box

A sentence compiles into a visible strategy: filters, ordering, and limits you can open up and adjust. If the queue is wrong, you can see which rule made it wrong.

Seed from one song

Give it a track and weight eight dimensions of similarity yourself: same artist, same era, same country, similar mood and so on. "Similar" stops being somebody else's definition.

Your library, mirrored locally

Your YouTube library syncs into a local database once, and every strategy runs against that mirror. Generating and previewing are deterministic, instant, and cost nothing.

Enriched tracks

Tracks pick up factual detail from MusicBrainz and a written sense of feel from a language model, so a rule about era or mood has something real to filter on.

Discovery beyond your library

Paginated discovery searches out past what you already own, and can name the artists that make up a scene rather than making you list them yourself.

Saved as a real playlist

Saving writes a genuine private playlist on your YouTube account, so it plays in the YouTube Music app with background playback, lock-screen controls and Bluetooth metadata. Confirmed working in a car.

The constraint

Why previewing is free and saving is not

The YouTube Data API gives a project 10,000 quota units a day, and writes are what hurt. That one number shapes the whole product: everything that can be done against the local mirror is free and unlimited, and the single expensive step tells you its price before you commit.

OperationUnitsWhat that means
Add one track to a playlist50The expensive one. A 50-track playlist costs roughly 2,550 units to save
Create a playlist50Paid once per playlist
One discovery search100Which is why discovery is paginated rather than fired off per keystroke
Read 50 tracks1Reading is close to free, so the local mirror is cheap to build and refresh
Roughly four playlists a day. Every unit spent is written to a ledger, and the app refuses any call that would exceed the day's budget rather than failing halfway through a save and leaving a half-written playlist behind. Updating an existing playlist inserts only the tracks that changed instead of rebuilding it.

Recently

What is new

A dashboard in three numbered steps

Sync, build, save. The remaining quota moved into the header where it belongs, and the first-run sync step is now impossible to miss.

Seed mode and enrichment

Seed-based playlists with eight weighted similarity dimensions, backed by track enrichment from MusicBrainz facts and a model-written sense of feel.

Discovery, paginated

Search past your own library without spending a search's worth of quota on every page, and ask it to name the artists behind a scene.

Your own uploads work

Tracks you uploaded to YouTube Music come back from the API as private. They cannot play in an embedded browser player, which is a rule of embeds and not a bug, but they add to a playlist and play in YouTube Music normally. All three behaviours were tested rather than assumed.

Built with

Claude API Next.js TypeScript Postgres YouTube Data API v3 MusicBrainz OAuth
Open music.corticorp.com