Questions

It stays with your account. Editions, receipts, and the notes you wrote belong to your Rovyn account rather than to the assistant that was connected when you made them — so a new assistant, granted the same permission to read receipts, can read that history from its first question. Disconnecting the old one ends its access and deletes nothing: the editions it built stay on your shelf, still credited to it, because it did build them. What you've written lives with you, not with any one assistant — Rovyn builds no profile of its own, it just hands over the receipts and your words.

Yes, and they stay out of each other's way. Each connection is its own grant with its own permissions, added and withdrawn on its own — one assistant for the long courses and another for the commute, if that's how you work. Disconnecting one ends that assistant's access and leaves the others running; reconnecting an assistant replaces its old grant rather than stacking up a second one. They write to the same shelf and read the same receipts, and every edition names the assistant that built it, so you can see whose picks you're correcting; the odd one built outside a normal connection shows up simply as “Assistant”.

Because your Rovyn account is where the editions land, and the account is simply whichever identity you signed in with. Sign in to the connector with Google and to the iPhone app with Apple and you have two Rovyn accounts — your assistant building editions into a library your phone isn't in. Accounts can't be merged afterwards, so the choice worth a second of thought is the first one: pick Apple or Google and use it at both doors. Connect says it again next to each assistant's steps.

The honest answer is what Rovyn asks for and what Rovyn hands back — a scoped brief goes out, receipts and your own words come back. The brief is what your assistant states about the request: the goal it's building for, plus whatever constraints you gave it. Rovyn doesn't ask for your chat history — the brief is a fixed shape: a goal, the time you have, a level, free-form preferences, and how long we may keep it. A field we don't recognise is dropped when the brief is validated rather than stored. Those free-text fields are roomy, though, so what lands in them is your assistant's choice, not ours; you can read the brief on the edition it produced, until a temporary-interest one is cleared. What returns is per episode: finished, partly, or skipped, roughly how far and how long, whether you saved it, and any note you wrote, passed on in your words rather than summarized or scored.

Those notes don't only travel when your assistant asks for them. If you granted the permission to read receipts, a bounded set of what you've written rides along with the tools it already calls — always when it builds an edition, and at most about once every twenty minutes on the tools it uses to look things up — searching episodes, listing shows, checking whether something it found is playable. Each note goes with the episode and show it was about, when you said it, what you did with it, and the goal it was heard under when there's one to attach; it's capped at forty notes, and long ones are cut at 500 characters and marked as cut, the whole receipt still there to ask for. Nothing is filtered to the topic at hand — deciding which note bears on what you asked for is the assistant's job rather than the server's. One box covers both: the permission that lets an assistant fetch your receipts is the same one that sends the notes along.

Raw playback events aren't exposed to assistants — no MCP tool returns them — and a connection carries the permissions you ticked at the consent screen: reading plans, writing plans, reading receipts, and writing feedback are four separate boxes. There's no way to drop one box from a live connection: changing what a connection holds means connecting again with different boxes ticked, and disconnecting the assistant takes all four away at once. The privacy policy has the specifics.

Any assistant that speaks MCP can connect, and four have step-by-step instructions on Connect: Claude and Perplexity on their paid tiers, Grok for all users, and ChatGPT through developer mode on a paid plan. Gemini is narrower — custom MCP apps are a Spark feature on the Ultra tier — and a listing in ChatGPT's directory is in review, with no date we can promise you. The assistants move their own doors often, so Connect is the page kept current; this one would only tell you where things stood the day it was written.

Because something already does, and it has a chat window where you can say what you actually want and correct it when it's wrong. Where every other podcast app puts Discover, Rovyn puts an empty shelf, waiting for what you ask for. A recommendation engine underneath that would only be a second opinion nobody asked for, quietly competing with the assistant you chose.

Yes, within its lane. Search covers both the open directory and everything you already have — editions, shows, episodes — and you can follow a show from a result or from the player itself; followed shows get a Shows tab with their latest ready to play, and anything you saved sits above them. Directory searches are answered by Podcast Index, so the words you type go to them. It's a fallback, not a feed: everything there is something you chose, and Rovyn still doesn't rank or recommend on its own. The editions your assistant builds stay the main event.

Mostly, yes. New editions can download over Wi-Fi in the background, so a queue built overnight is typically on your phone by morning, and you can download any episode or edition by hand. Downloaded episodes play without a connection; finished ones are removed after a few days so storage tends to look after itself. Streaming still needs a connection, as anywhere.

The protocol is. Cueback — the language Rovyn and your assistant speak to each other — is media-neutral and specified apart from the app, so other players and services can implement it. It's published: github.com/tinkon/cueback carries the specification, the MCP tool reference, and the @cueback/* packages, all Apache-2.0, currently at protocol 0.4. The Rovyn service and app themselves are ours, hosted and run by us. If you're building against Cueback, support@polimati.com reaches the people who wrote it.

Nothing at the moment. The iPhone app is in TestFlight until launch — join directly at testflight.apple.com/join/brhnD7Mm, no invite needed. The one cost that isn't ours: several assistants keep custom connectors on their paid tiers, noted per assistant on Connect.