Roadmap

Published phases, honest status.

We ship in numbered phases and publish the state of each one. A phase is finished when it runs on the deployed service, on a phone viewport, with a test that fails loudly if it breaks.

3phases shipped
1in progress
3planned, plus one in research
Q2 2027target for the public release

Status as of 9 November 2026.

Phases

ShippedPhase A · Data foundation

Complete history, imported and reconciled

Extended streaming history is parsed, deduplicated and loaded into a relational core with vector indexing. The reference history — seventeen years, 271,712 plays — imports end to end, and the analytics API reads from the same tables the product will use.

ShippedPhase B · Assistant

Questions answered in plain language

An intent router over thirteen handlers, streaming answers as they are computed, with the figures that produced them attached. Model access sits behind a provider abstraction with an ordered fallback chain, so one vendor outage degrades the assistant instead of stopping it.

ShippedPhase C · Memory

A taste model that persists

Track vectors, listening context and explicit feedback are stored so the model improves with use. Durable notes about a listener's taste survive between sessions — the difference between a query box and an assistant.

In progressPhase D · Mobile shell

The phone client

An installable progressive web app at a 380 px design width: dashboard, explore and chat in one shell, with the insight suite built directly into it. Everything after this phase is judged on a phone screen, so this is where the product stops being a backend.

PlannedPhase E · Insight suite

Recaps, streaks and trajectories

Year recaps, listening streaks, mood trends, genre drift and artist trajectories — the long view, rendered as something you can read in a minute.

PlannedPhase F · Playlist studio

Tools for playlists you already own

Reordering, deduplicating and re-scoring existing playlists against the taste model, with a preview before anything is written back.

PlannedPhase G · Discovery

Recommendations that explain themselves

A discovery engine that returns the nearest neighbours in feature space and says why — the artist, the record, the reason.

ResearchPhase H · Cost ladder

Affordable enough to leave running

A measured query budget per phase and a proven cheaper equivalent before the expensive one becomes a problem. Cost is a product requirement here, not a finance surprise.

How we decide what ships

Three rules decide the order of work. They are deliberately boring, because they are the reason the product keeps moving.

  • Deployed or it did not happen. A feature is done when it runs on the live service and a smoke test proves it — not when the branch is green locally.
  • Phone-first, not phone-also. 380 px is the design width. If it is not usable one-handed, it is not finished.
  • No silent failure. Every phase carries at least one acceptance criterion that fails loudly, so “looks fine” is never the standard.

What we are not building

No social feed, no follower counts, no AI-generated cover art. The product is a private instrument for one listener at a time; anything that turns it into a stage works against that.

Follow the build.

Beta invitations go out with each phase. Ask for access and we will tell you where the product is when your turn comes.