See your taste. Edit it. Have it find you new music you'll love.
Every streaming service watches what you play — none let you see, steer, or own the model they build from it. MaxJam hands you an editable, queryable model of your own taste, lets you correct it in plain language, and aims fresh discovery wherever you point it.
Situational playlists land in the library you already use. Any device — no hardware required.
The gap
Your recsys knows you listen to a lot of one artist — but you can't ask it what it thinks you're into, can't tell it to drop a phase you've moved past, and can't aim its discovery. Your only correction tool is the skip button.
No incumbent has an editable memory. You can't read the model, so you can't fix it.
Your taste on each service is a separate model that never reconciles. Nobody owns the union.
"Daily Mix 3" and a canned "Focus" station — mass-market templates, not your 90-minute deep-focus set.
A weekly black-box drop you take or ignore. Can't ask why, can't dial novelty, can't aim it.
How it works
MaxJam reasons over a durable, editable taste profile and your live context, then calls real tools. It doesn't train a model — it assembles good tools, good context, and a taste store you own.
Every track you play becomes a timestamped event in your store — across the services you actually use, with a full history backfill.
An editable, queryable taste model — recency-decayed, reliability-weighted, and overridable by plain-language rules that win immediately. "What have I been into this month?" gets a real answer.
"Make me a 90-minute deep-focus set." A situational playlist with a warm-up and a cool-down — written straight into your real library.
"Find me new stuff like this, but mellower." Fresh, unfamiliar picks tuned for novelty and surprise, deduped against everything you've already heard — with a weekly digest and a one-line why for each.
It's the only design that gives you a taste model you can read and correct, a discovery dial you can aim, and situational playlists shaped to your listening — across the union of your services.
"Stop recommending that artist." "I'm done with that 2023 workout phase." Plain language, applied immediately — the killer feature every reviewer named first.
Turn novelty up or down. Point it at a seed. Tell it to stop surfacing what you already play daily. Ask why any pick is there.
Your listening history and derived taste model live in your own store. The catalog is rented and swappable; the part that matters is yours and portable.
A dual-LLM quarantine keeps untrusted catalog text out of the control plane; every pick is grounded against a real catalog before it lands. No destructive actions exist.
Roadmap
Ship the brain first — prove it on any device with no hardware. Then give it a where.
The editable taste model, situational playlists into your real library, and aimed weekly discovery. No speakers, no LAN — just the brain, anywhere.
The same brain, routed across a house of Sonos speakers. Per-room reasoned sets, energy curves, follow-me — a delivery channel on the proven brain, not a second product.
For agents
MaxJam is agent-first — the tool surface is the product; this page is the layer on top. Point your agent at the machine-readable entry points.
Every pick resolved against a real catalog; untrusted text kept out of the control plane.