See your own taste, mapped
Your listening is scattered: Last.fm has your plays, Discogs has your collection, screenshots hold the rest. taste resolves all of it to one identity per artist (a cluster_id) on the hosaka map and turns it into a dashboard: your play timeline, your collection, your artist clusters, how much resolved clean, and recommendations grounded in the documented graph.
No account, no connection required. The demo runs on a fictional listener profile and is publicly cacheable because it holds no real data.
What the dashboard shows
Plays over time, anchored to the dates your sources recorded them. See how your listening has shifted over the years.
What you own or have shelved, with resolution coverage per item. Unmatched items are labelled, not hidden.
Your artists plotted on the hosaka map — see the scene neighbourhoods your taste actually sits in.
A dial showing how much of your library resolved clean. Resolution is never 100% — roughly 47.9% of scene-booking artists lack a Discogs ID or MusicBrainz ID — so the dial shows the honest number rather than hiding it.
Next cut: graph neighbours of what you already love, with scene chips that name the city-scene an artist plays in. Since July 2026, suggestions traverse the hosaka map — roughly 1.97 million artist clusters from 14 independent source kinds across 448 scene areas — seeded by your resolved artists, not by what other listeners clicked.
Not collaborative filtering
Most recommendation systems infer taste from aggregate listener behaviour: “users like you also played X.” taste does not do that. Every recommendation traverses the hosaka map — shared bills, labels, scenes, co-appearances — seeded by the artists you have already resolved. The path from a seed to a suggestion is a documented connection in the graph, not a weight in a matrix trained on anonymous play counts.
This means suggestions change as the hosaka map grows, not as taste accumulates more listeners. There is no engagement-maximising feed; the graph does not optimise for session length.
Ways in, today
ninsei.hosaka.fm is the first app built on taste. Connect your Last.fm or Discogs from inside NINSEI and your resolved taste follows you in the player. No separate account setup — the connect flow is hosted by taste and NINSEI reads the result.
Sign-in to the taste dashboard directly (magic-link email) is early-access while taste is in dub-plate — a small-run, design-partner phase before wider release. Email delivery is limited during this phase.
No instant sign-up — we review and add accounts manually during dub-plate.
Your rights, inline
The mechanisms are documented in full at /privacy. The short version:
- Export everything — one call returns all of it: sources, items, resolutions, and the consent ledger with
rcpt_v1_receipts. - Delete everything —
DELETE /api/v1/meerases everything personal in a single database cascade. The bearer token you called it with stops working the moment it completes. - Revoke a device — each bearer token (
taste_v1_, 90-day sliding expiry) is independently revocable. Revoking one does not touch others. - Screenshots are read once and discarded — the image is OCR’d in memory and then thrown away; it is never written to disk. Nothing enters your profile until you confirm the extracted candidates.
Price
Free while taste is in dub-plate. No ads. No selling of data. No cross-app profiles. taste runs on its own encrypted stack, separate from every other hosaka service.
The pricing model is being finalised alongside the contract for general access — we are proving the pressing before cutting for scale.