sync_replay_project/docs/PROTOCOL.md

Protocol (proto = 16)

Wire format only. Behavior (gate, clock, timeouts) is in ARCHITECTURE.md. Source of truth: packages/core/lib/src/protocol.dart and packages/server/lib/src/server_hub.dart.

Any wire change bumps RelayConstants.protoVersion and edits this file in the same commit. Versions are ranges: a server accepts [protoMin, protoMax] (RelayConstants.protoMin / protoMax, both 16 today), a client sends its own range (hello.protoMin .. hello.proto), and the connection uses the highest version both have (welcome.proto). Later additions raise protoMax only, so they stay additive. Fields marked reserved are parsed and documented but nothing produces them yet; the restructure step named next to them does.

Transport

• WebSocket at /sync, JSON text frames, each an object with a t field.
• The first client frame must be hello within 10 s. Server pings every 5 s.
• Match incoming frames by t, never by order: welcome can race state.
• A socket without any frame from its client for 15 s (RelayConstants.silenceSecs) is closed by the server (close code 4015) and its member is gone (see "Presence"). Clients send a report every second and a ping every second anyway.
• Failures: err (see "Errors and close codes"); a refused hello is followed by a close.
• GET /health returns JSON (unauthenticated, no member names): ok, name, proto (= protoMax), protoMin, protoMax, version (the server's app version, e.g. 1.3.0), tls (reserved: always false; a server with TLS would serve the same endpoints over wss:// / https://), viewers, bytesServed, uptimeSecs. The app reads it before joining and does not dial a server whose range does not meet its own (the version gate, see ARCHITECTURE "Server lifecycle").
• GET /m/<sid>/<token>/<name> serves the host's file (Range table in ARCHITECTURE.md). GET /j/<sid>/<token>/<itemId> relays the current Jellyfin media and Up next's first one (see "Jellyfin media" below). /r/<sid>/<token>/... is the relay of a member's media (see "Relay" below). GET /a/<sid>/<token>/<hash> serves a member's picture (see "Avatars" below), GET /i/<sid>/<token>/<hash> the current media's art (see "Media art" below). POST and GET /l/<sid>/<token> send and read the session log (see "Diagnostics" below). A token works from the member's welcome until it leaves (through its grace, see "Presence").

Errors and close codes

{"t":"err","code":"version","min":16,"max":16,"server":"1.3.0","for":"hello",
 "message":"proto mismatch: got 15, want 16..16"}
{"t":"err","code":"bad","for":"seek","message":"bad message: position must be finite"}

• message is plain text for people; a proto 15 app shows it as is (the version refusal starts with proto mismatch: for that reason). code is for programs; for names the t of the frame that failed, when there is one. With code version: min/max the server's range, server its app version. This shape is stable from proto 16 on.
• Codes: version (no common version), secret (wrong secret), rejected (understood but refused: rejected: avatar too large, server fetch needs a server password, The server can't fetch this link: ...), bad (malformed: bad hello: ..., bad message: ... for a frame with a non-finite number or a bad queue item, bad swap media, swap needs a url or the hosted file, relay fetch needs an http(s) url), replaced (another hello of the same client took the member over; the client stops, "opened in another window"), stale (this hello is older than the member's current one), forbidden (presence in for someone else; an auto queue item from someone else than the swapper), rate (reserved; nudges over the cap are still dropped silently), full (Up next is full (100 items)).
• After err the socket stays open, except at hello and for replaced, which are followed by a close.
• Close codes (the server's; constants: CloseCodes in core):

Code	Socket	Meaning
1000	/sync	the member left (bye, or its grace ran out)
4001	relay channel	not (or no longer) the source of this relay; the source stops
4002	relay channel	the source is back with another media than the relay's first info; the relay ends, the source stops
4003	relay channel	a newer channel of the same source replaced this one; this one stops
4010	/sync	refused: err version
4011	/sync	refused: err secret
4012	/sync	refused: err bad or rejected at hello
4013	/sync	err replaced: another connection of the same client holds the member
4014	/sync	refused: err stale
4015	/sync	silent for 15 s: the member is gone
4016	/sync	the hosting app stopped sharing its server: the member has left (leave, reason closed); the client does not reconnect

Client to server

{"t":"hello","protoMin":16,"proto":16,"secret":"","user":{"name":"","avatar":""},
 "media":{},"hostKey":"...","client":"<install id>","seq":7,"app":"1.3.0",
 "platform":"linux","caps":{"source":true,"resolve":true,"jellyfin":true},
 "feedAfter":41}
{"t":"ready","ready":true,"source":"manual"}      source: manual | auto
{"t":"force","target":"all"|"<memberId>","ready":true}
{"t":"seek","position":123.4,"seq":3}
{"t":"swap","media":{"name":"","size":0,"duration":0,"url":"..."}}
{"t":"swap","media":{"name":"","size":0,"duration":0,"jellyfin":{<ref>}}}
{"t":"swap","media":{"name":"","size":123,"duration":0,"url":"...","relay":{}}}
{"t":"swap","media":{...},"queueId":"q3"}         puts up an Up next item
{"t":"report","position":123.1,"buffered":6.2,"eventId":41,
 "duration":1440.5,"mediaId":2,"jf":{"<serverId>":true},"error":"Not allowed (403)"}
{"t":"ping","c":1767225600000}
{"t":"chat","text":"..."}
{"t":"presence","target":"<memberId>","state":"away"|"in","reason":"manual"}
{"t":"bye"}
{"t":"nudge","target":"<memberId>"}
{"t":"profile","name":"...","avatar":"<base64>"}
{"t":"art","mediaId":2,"poster":"<base64>","banner":"<base64>"}
{"t":"queue","op":"add"|"move"|"remove"|"play"|"skip"|"cancel",...}   see "Queue"

Field notes:

• hello.proto / hello.protoMin: the client's version range (protoMin absent: proto only). The server reads them before anything else in the hello (a hello of another version may not parse) and refuses with err version when the ranges do not meet. hello.media is ignored; a proto 16 client still sends {} so a proto 15 server, which requires it, answers with its own proto mismatch.
• hello.secret: the server password; a server with an empty secret is open. The password may change while members are in (the hosting app sets it when it opens its server to friends).
• hello.hostKey (optional, omitted when empty): the hosting app generates a random key per run, gives it to its server and sends it from its own client. The member whose key equals the server's is the host (members[].host), and its hello needs no secret. A server without a key (the dedicated server) has no host. Where the connection comes from does not matter.
• hello.client (optional): a random id per install, 16 to 64 characters of [A-Za-z0-9_-] (the app uses 22 characters of base64url), else err bad. Private: never broadcast, never logged. It names the member across connections (see "Presence"); without it every hello is a new member. Reserved: deriving it per server (an HMAC of an install secret and the dialled host:port).
• hello.seq (default 0): increases with every connection this install makes. A hello whose seq is lower than the one holding the member gets err stale and is closed; equal or higher takes the member over.
• hello.app (the app version, cut to 32 characters): shown as members[].app. hello.platform (linux, windows, macos, android): logged only. hello.caps: what the app can do (source: serve a relay; resolve: run yt-dlp; jellyfin: play with its own account), shown as members[].caps; unknown entries are ignored.
• hello.feedAfter (optional): the last feed.seq the client saw from this server; welcome.feed replays the lines after it. A client should send it only for the same welcome.instance (a restarted server counts from 1 again).
• seek.seq (optional, default 0): increasing per client; echoed in state.seekSeq so the sender recognises its own seek.
• report.duration / report.mediaId (optional, omitted when 0): the open media's length in seconds and the mediaId it belongs to. The server takes the first one whose mediaId is current while its media duration is 0. report.buffered is the current media's: a client sends 0 while its player does not hold it yet, and the server counts a report whose mediaId names an earlier media as buffered 0. A joining member's report ends its joining once it holds the current media at the server position (within 5 s) with the buffer the gate needs (see "Presence").
• report.jf (optional): Jellyfin access per ServerId, true = this client plays the item with its own account, false = through the relay. Sticky: a report without jf keeps the last one. At most 8 entries, keys up to 64 characters; others are dropped. The reference client lists every server it has a working account on as true, then the current item's server as it plays it (so a queued Jellyfin item finds someone to put it up while other media plays).
• report.error (optional, omitted when empty): short plain text saying why this client's player cannot open the media report.mediaId (e.g. Not allowed (403)). The client sends it in every report while that lasts and drops it once the media opens (a reload included). The server keeps it per member (one line, cut to 120 characters) only for the current mediaId: a report without it, or for another mediaId, clears it, and so does a swap. An errored member is out of the ready gate: out of the set, a ready or force does not put it back; when the error clears it is in the set again like a join (unless playing, then only if it is ready). It also ends a joining.
• ping.c: the client's clock in epoch ms; answered with pong.
• user.avatar: base64 PNG, at most 256 KiB of characters (else rejected: ...). name is trimmed, empty becomes ?, cut to 32 characters. A takeover applies the hello's name and picture like a profile.
• profile: changes this member's name and/or picture mid-session. A field that is absent (or null) stays as it is; avatar: "" removes the picture. Same rules as hello.user; a too large avatar gets err rejected: avatar too large (the socket stays open, nothing changes).
• art: pictures of the current media (see "Media art").
• chat.text is cut to 500 characters.
• presence: target a member id (absent or empty: oneself). state away sits target out: anyone may, on anyone (the host included); a gone target comes back away. state in brings oneself back (through joining, see "Presence"); for anyone else it is err forbidden. reason (manual default, idle, minimised, bye) goes into the feed line. Unknown targets and no-op changes are ignored.
• bye: the member leaves now (the Leave button, the app closing): feed leave (reason bye), its token stops working, the socket closes (1000). Anything else that ends a socket keeps the member (gone), except the hosting app closing its server to friends: everyone but the host leaves at once (feed leave, reason closed) and the socket closes with 4016.
• nudge: the server sends {"t":"nudge","from":<memberId>,"name":<name>} to target only (no feed). At most 3 per second and 20 per minute per sender; the rest, nudges to oneself and to unknown targets are dropped silently.
• swap.media.url is the link; name, size, duration are informational. swap.media.jellyfin makes it Jellyfin media (see below); then url is dropped. swap.media.relay makes it relayed media (see "Relay"). swap.media.ref says what it is (see "Media identity"). A swap keeps every member's ready and puts every member in the set (inSet; who counts is up to presence); clients that want to be unready for the new media send ready themselves (gate: ARCHITECTURE.md, "Ready gate").
• Frames before welcome (other than hello) are ignored; unparsable or unknown frames are ignored silently. Frames from a socket another hello took over are ignored.

Server to client

{"t":"welcome","proto":16,"token":"...","memberId":"m1","resumed":false,
 "instance":"<random per boot>","grace":600,"mediaUrl":"...","relayUrl":"...",
 "state":{<state>},"feed":[<feed with "replayed":true>,...]}
{"t":"state","eventId":7,"transport":"playing"|"held"|"finished","position":12.3,
 "serverTime":"<ISO-8601 UTC>","mediaId":1,
 "media":{name,size,duration,url?,jellyfin?,relay?,ref?,swapBy?,art?}|absent,
 "members":[{"id","name","avatarHash"?,"ready","inSet","buffered","presence",
   "goneSince"?,"jf"?,"host"?,"error"?,"rate"?,"app"?,"caps"?}],
 "seekBy":"m1","seekSeq":3,"relay":true,"relayNext":true,"queue":[<item>]?,
 "queuePlay":"q3"?,"hold":{"reason":"joining","members":["m3"],"until":"<ISO>"}?,
 "advance":{"id":"q3","at":"<ISO>","secs":10}?}
{"t":"pong","c":1767225600000,"s":1767225600012}
{"t":"feed","seq":12,"actor":"","actorId":"m2","action":"","target":"","targetId":"m3",
 "reason":"","at":"<ISO-8601>"}
{"t":"chat","text":"...","from":"<name>","fromId":"m2"}
{"t":"nudge","from":"m2","name":"Alice"}
{"t":"err","code":"...","message":"...","for":"..."}

• welcome.token is private to its owner and appears only in that member's media URL. mediaUrl is omitted when there is no media; it is a built /m/... URL for a host file (host part = what the client dialled), otherwise the shared link. welcome.state has the same shape as state. relayUrl is present only when state.relay is: this member's /j/... URL for the current Jellyfin media (private like token). welcome.proto: the version of this connection. welcome.resumed: the hello took over this client's member (same memberId and token, everything it had kept); false for a new member, also after a server restart. welcome.instance: random per server boot (a client that held a member of another instance knows the server restarted, not that its grace ran out). welcome.grace: seconds a gone member is kept. welcome.feed (omitted when empty): the feed lines after hello.feedAfter, at most the last 50, oldest first, each with replayed: true (log them, do not toast them).
• state is sent every second and on every change. position is the value at serverTime (server clock); clients extrapolate while transport is playing, on the server's clock (see pong). finished: the media reached its end (position = duration); it stays until a seek or swap. eventId increases on seek, swap and transport changes. mediaId increases on swap. media.relay is present for relayed media (see "Relay"). media.duration is 0 until a client reports it. media.swapBy: member id of who swapped it in (omitted once that member left, and for the media the server started with). media.art: see "Media art". media.ref: see "Media identity". seekBy (member id) and seekSeq are present only while the current eventId came from a seek. relay (omitted when false): the server can relay the current Jellyfin media. relayNext (omitted when false, proto 13): it can relay Up next's first item too (a Jellyfin ref on the same login), so viewers without access preload it from /j. members[].jf is the member's last report.jf, omitted when it never sent one. members[].avatarHash (omitted when the member has no picture) names the picture's bytes (see "Avatars"). members[].host (omitted when false): the member sent the server's hostKey (see hello.hostKey). members[].presence and goneSince (ISO, only while gone): see "Presence". members[].inSet: the member counts for the ready gate right now (in the set, no player error, and its presence counts). members[].error (omitted when empty): the member's player cannot open the current media (see report.error). members[].rate (omitted when 0): bytes per second the server sent this member over the last 5 s (/m, /j and /r together), two significant digits, under 10 kB/s counts as 0. Members that fetch the media themselves read nothing from the server. members[].app / caps (omitted when empty): its hello.app / hello.caps.
• state.hold (present while held with media up): why. reason, most pressing first: joining (a joiner within its bound holds the start; until is when that bound ends, server clock), ready (someone in the set is not ready), buffer (someone's buffer is short of bufferNeed), source (the same while the relay's source is gone), start (nobody counts for the gate: someone has to ready up). members: the member ids the gate waits for (empty for start).
• state.advance (reserved, step 6): the shared countdown to Up next (see "Queue").
• pong.c echoes the ping's c; pong.s is the server's clock in epoch ms (the clock serverTime uses). With t3 the local receive time, the offset is s - (c + t3) / 2, accurate to half the round trip.
• feed: seq increases by one per line for the server's life (a restart counts from 1 again); actor / target are display names as they were then, actorId / targetId the member ids (server for the server's own lines; omitted when none): match people by id, not by name. action values: ready, unready, force-ready, force-unready, force-play, force-hold, join (a new member only), leave (for good: reason bye, grace or closed), gone (10 s after the member went gone; a quicker return says nothing), return (back within the grace after a gone line), away (target the member set away, absent when the actor set itself; reason idle/minimised/bye, empty for manual), in (back from away), swap, seek, pause, resume, finished, chat, rename, avatar, queue-add, queue-remove, queue-skip, advance-cancel (target the dropped continuation's title, if any), fetch-fail (the server could not fetch the current media, target its name). reason carries auto/manual for ready. rename: actor is the new name, target the old one. avatar: the actor changed only the picture. queue-*: target is the item's title, reason (skip) why it could not play (see "Queue").
• Chat is delivered twice: a chat frame with the text (fromId the sender's member id) and a textless feed with action:"chat". Render the chat frame only; the reference client drops the textless feed entry.

Presence

The member is lasting: one per client, kept across reconnects. Gate rules and timings in detail: ARCHITECTURE.md, "Sync model".

• The server maps hello.client to one member. A hello with a known client id (and a seq not older than the member's) takes that member over: memberId and token stay, so /m /r /j URLs keep working; the ready flag, presence, queue items, relay, picture all stay; an older socket still open gets err replaced and closes (4013). welcome.resumed is true.
• A socket close (or 15 s of silence) is not a leave: the member goes gone (goneSince) and is kept for the grace, 10 min (welcome.grace), with everything it had; a relay it serves goes on (see "Relay"). After the grace it leaves (feed leave, reason grace), its token stops working; a hello with its client id after that is a new member. bye leaves at once.
• members[].presence: joining (new, back from gone, or back in from away, while media is up: it loads in the background until its report holds the current media at the server position with bufferNeed buffered, or reports an error, or 60 s pass; then in), in (watching), away (sat out: still in the server, never counted for the gate, the ready counter or watched tracking), gone (disconnected, in its grace). A member away when it dropped comes back away. With no media up a joiner is in at once.
• What each presence counts for is one table, PresencePolicy in core (presence.dart), used by the server and the app: the gate counts in, and joining only while held and for at most 15 s after it started joining (never while playing: a joiner never pauses anyone); the ready counter and preloads count joining and in; watched tracking and full social notifications count in (joining too for the notifications).

Media identity

media.ref = {"kind":"episode","title":"Show","anilistId":1,"anilistEp":6,
             "tvdbId":2,"season":1,"ep":6,"absolute":30}

• In swap.media, state.media and queue[].media; passed through as is (the server drops non-integer ids and cuts title to 200 characters). kind episode, movie or extra (else no ref); every other field optional; season absent for absolute numbering. At most 2 KiB of JSON, else err bad. Reserved: nothing produces it yet (step 4).

Avatars

• The picture is sent as base64 in hello.user.avatar or profile.avatar; state carries only its hash: the first 16 hex characters of the sha256 of the decoded bytes. An avatar that is not valid base64, or whose bytes are not a PNG, JPEG, WebP or GIF (by magic bytes), counts as none.
• GET /a/<sid>/<token>/<hash>: the bytes with their image content type, cache-control: private, max-age=31536000, immutable (a hash names fixed bytes; clients fetch each hash once, from the host they dialled). Errors: 401 bad token, 404 no current member has that picture, 410 unknown session. Pictures go with their member (leave, change, removal).

Media art

client  {"t":"art","mediaId":2,"poster":"<base64>","banner":"<base64>"}
state   media.art = {"poster":"<hash>","banner":"<hash>"}

• A portrait poster and a wide banner for the current media, so every member (guests without Jellyfin included) can show them. Sent by the member who made the current swap (media.swapBy) or the host, for the current mediaId; from anyone else, or for another mediaId, it is ignored. Each field is optional; one that is not a picture (same check as avatars) or larger than its cap (poster 256 KiB, banner 768 KiB of decoded bytes) counts as none. A later art replaces the earlier one whole; any swap clears it.
• media.art carries hashes like avatarHash; a key is omitted when that picture is absent, art when both are. The same picture as poster and banner (one hash) is a web video's thumbnail.
• GET /i/<sid>/<token>/<hash>: as /a/ (see "Avatars"); also the pictures of queued items (queue[].thumb). 404 when neither the current media nor a queued item has it.

Jellyfin media

<ref> = {"serverId":"<Jellyfin ServerId>","itemId":"<32 hex>",
         "mediaSourceId":"...","title":"Pilot","subtitle":"Show - S01E01",
         "runtimeSecs":1440.5}

• serverId, itemId required; mediaSourceId, subtitle, runtimeSecs optional (omitted when absent); title always sent (may be empty). runtimeSecs is informational: the sync clock still uses the duration a client reports.
• The server keeps only the ref: for Jellyfin media no URL or token is stored or broadcast (a url sent along is dropped). serverId, itemId and mediaSourceId must match [A-Za-z0-9-]{1,64}, otherwise bad swap media. media.name is the swap's name, else the ref's title; names are sanitized like link names.
• Each client resolves the ref with its own account on that server. A client without one plays the relay URL when state.relay is true: welcome.relayUrl, or after a swap built like the /m one from the host it dialled: http://<host>/j/default/<token>/<itemId>.
• GET /j/<sid>/<token>/<itemId>: for the current item or, when state.relayNext, Up next's first one, the server fetches /Videos/<itemId>/stream?static=true&mediaSourceId=<id> (the original file) with its own credential and streams it through. Range is forwarded as sent; status, Content-Range, Content-Length, Content-Type and Accept-Ranges come back as Jellyfin sent them. Errors: 401 bad token, 403 neither of those items, 404 no relay for either, 410 unknown session, 502 Jellyfin unreachable or rejecting the server's credential. Bytes count toward bytesServed.
• Pasted Jellyfin links sent as plain url stay plain links.

Relay

Media whose bytes go through the server's buffer: a file on a member's computer, a link passed on by a member, a link the server fetches itself, or a Jellyfin item for viewers without an account. Behavior (buffer, flow control, cases) is in ARCHITECTURE.md, "Relay".

swap  media.relay = {}               the swapper's machine serves the bytes
      media.relay = {"tracks":[{"kind":"video"},{"kind":"audio"}]}
                                     the same, as two tracks played together
      media.relay = {"tracks":[{"kind":"video","as":"pieces"}]}
                                     the same, as one segmented playlist
      media.relay = {"fetch":true}   the server reads media.url itself
state media.relay = {"source":"m2","live":true}   or {"fetch":true,"live":true}
      media.relay = {"fetch":true,"live":true,"tracks":[{"kind":"video"},{"kind":"audio"}]}
      media.relay = {"source":"m2","live":true,"rate":4100000,"need":2300000}
      media.relay = {"source":"m2","live":true,"queueId":"q3"}   started in Up next (proto 14)
      media.relay = {"source":"m2","live":true,"sourceGone":true,"sourceGoneSince":"<ISO>"}
                                     its source is gone, in its grace
      queue[0].media.relay = {"source":"m2","live":true,"tracks":[...]}   its warm relay
      queue[0].media.relay = {"fetch":true,"live":true}

Tracks: a relayed media is one file, or tracks that viewers play as one (a page's separate video and audio). tracks lists them in order, up to 4 entries of kind video or audio, each a Range-readable byte stream, or with "as":"pieces" a segmented stream (an HLS media playlist or DASH fragments) relayed piece by piece and served as a VOD media playlist. A piece is one segment, init section (EXT-X-MAP) or AES-128 key, fetched upstream whole with one GET. At least 2 entries, or 1 pieces entry (a muxed playlist, kind video); absent, an unknown kind or as, or anything else means one file.

• swap: relay with jellyfin keeps the ref (the swapper serves it to viewers without an account; fetch is ignored there). relay without jellyfin: no URL is stored or broadcast; fetch needs an http(s) url (else relay fetch needs an http(s) url), which the server keeps to itself, and a server with a password (else server fetch needs a server password: an open server would fetch any address it can reach for anyone). A web page url is read through the server's yt-dlp; when the server cannot read the link, the swapper later gets The server can't fetch this link: <reason>; when yt-dlp failed and the server has no cookies, the reason is it has no cookies (...) instead. Any later swap ends the relay. tracks on a member's swap names the tracks its source will serve; the server ignores it with fetch and with jellyfin.
• state.media.relay.source: member id of the source (absent with fetch). live (omitted when false): bytes can flow (the source's channel is connected, or the server could read the link). tracks: as the swap named them, or, with fetch, set once the server has read the link (together with live).
• sourceGone (omitted when false) and sourceGoneSince (ISO): the source member is gone (see "Presence"). The relay and its buffer stay through the source's grace; viewers play what is buffered. A source's leave (grace over, bye) ends the relay: live goes, sourceGone stays, its URLs answer 404. A swap ends it as before. The warm relay's queue[0].media.relay carries the same fields.
• state.media.relay.rate (omitted while unknown): bytes per second the source (with fetch, the server's own fetch) delivers while the buffer is waiting on it, smoothed over about 4 s of that time and present once 2 s of it are measured; it holds while the buffer is full and the source idles. need (omitted while unknown): the media's average bytes per second, the tracks' sizes over media.duration (a pieces track: its segments in so far, after a minute of them). Both two significant digits, updated once a second. rate under need means the source cannot keep up; clients tell the source once per media.
• Warm relay (proto 14): while Up next's first item is relayed (how server, through or file, or jellyfin when someone present has no access to that server and state.relayNext is false), the server holds a small relay for it and shows it as that item's media.relay: source (the adder; for jellyfin empty until the first member with jf true for that server dials in), or fetch, and live with tracks once bytes can flow. Its URLs use the item's queue id where a media id would be. Not for the host's own file, nor for an adder alone in the server. A swap with queueId of that item, from its source (or, fetch, from anyone) with the same tracks, makes it the media's relay: media.relay.queueId is the item's id, media.name the item's media.name, the buffer and source carry on and the queue-id URLs keep working. Any other swap of the item, or the item no longer first, drops it (its URLs 404, its source is closed with 4001).
• welcome.mediaUrl for relayed media without jellyfin: this member's /r/ URL; absent for the source itself (it plays its own copy) and for a fetched link the server has not read yet (clients open it on live). Clients build it after a swap like the /m one: http://<host>/r/default/<token>/<key>/<name> (<key> the mediaId, or media.relay.queueId when set; name with / and \ made _, then URL-encoded; the server ignores it). With tracks, each track's URL is http://<host>/r/default/<token>/<key>/<i>/<name> (a pieces track: <name>.m3u8); one track is opened as is, several as an mpv EDL of them all, each framed by its byte length: edl://!no_clip;%<len>%<track 0 url>;!new_stream;!no_clip;%<len>%<track 1 url>.
• GET /r/<sid>/<token>/<key>/<name>: the media (one file), Range table as for /m; <key> a mediaId, or a queue id (the warm relay, or the media's relay started from it). GET /r/<sid>/<token>/<key>/<i>/<name>: track i (from 0) of a media with tracks (track 0 of a one-file media too); for a pieces track (any name), its media playlist (application/vnd.apple.mpegurl): #EXT-X-PLAYLIST-TYPE:VOD, EXT-X-MEDIA-SEQUENCE, EXTINF durations, EXT-X-DISCONTINUITY, EXT-X-MAP and EXT-X-KEY:METHOD=AES-128 (with IV when given) as upstream, EXT-X-ENDLIST; every URI is p/<j>. GET /r/<sid>/<token>/<key>/<i>/p/<j>: piece j of pieces track i, whole, once it is in (a Range inside the piece answers 206). Errors: 401 bad token, 404 not the current relayed media nor the warm one (a player still on an older mediaId gets 404, never the new bytes), no such track (the one-file URL of a media with tracks) or piece (or p/ of a byte track), 410 unknown session, 503 + retry-after the source never connected or the server could not read the link (60 s), or a piece did not arrive within 60 s. A byte request whose bytes do not arrive within 60 s ends early. Bytes count toward bytesServed.
• GET /r/<sid>/<token>/source (WebSocket): the source's relay channel, dialled by the source (works behind NAT); ?q=<queueId> for the warm relay of that item (or the media's relay started from it). Only that relay's source token is accepted; anyone else, or the source once its relay is dropped, is closed with code 4001 (the source stops instead of reconnecting). A warm relay's tracks are those of its source's first info. The source's token works through its grace, so a source back from a drop dials the channel again (before or after its own hello) and sends its info again: the same as its first (track kinds, byte sizes, piece shapes, and name / duration where both give them) continues the relay: the buffer stays and the server pulls what it still lacks with new pull ids, from the first byte or piece not buffered (the pull frames below are that exchange; nothing else is needed). Another info closes the channel with 4002 and ends the relay. A second channel for the same relay replaces the first, which is closed with 4003. The source stops on 4001, 4002 and 4003 and reconnects (with backoff) on anything else. Frames:

source -> server  {"t":"info","size":<bytes>,"name":"Ep 06.mkv","duration":1440.5}
                                                         first frame, one file
                                                         (name, duration optional)
                  {"t":"info","tracks":[{"kind":"video","size":<bytes>},
                                        {"kind":"audio","size":<bytes>}]}
                                                         first frame, tracks
                  {"t":"info","tracks":[{"kind":"video","as":"pieces","seq":5,
                     "pieces":[{"init":true},{"key":true},
                               {"d":6.0,"map":0,"key":1,"iv":"0x..."},
                               {"d":4.5,"map":0,"disc":true}]}]}
                                                         a pieces track
                  binary: [u32 big-endian pull id][bytes]  in order per pull
                  {"t":"end","id":1}  {"t":"fail","id":1,"reason":"..."}
server -> source  {"t":"pull","id":1,"track":0,"start":0,"end":1048575}
                                                         inclusive; no track: 0
                  {"t":"pull","id":2,"track":1,"piece":17} piece 17, whole
                  {"t":"ack","id":1,"n":65536}             bytes arrived
                  {"t":"cancel","id":1}                    stop that pull

Pull ids are unique across the tracks of a connection; track is sent only for a media with tracks. A pieces track's pieces are its pieces in order: {"init":true} an init section, {"key":true} an AES-128 key, else a segment of d seconds whose map / key name the init and key pieces it uses, iv an explicit IV (0x + hex), disc a discontinuity before it; seq (default 0) the media sequence of the first segment. A piece pull's bytes end with end like a range pull. A first info whose tracks differ from the swap's tracks (count, bytes or pieces), or a malformed piece list, closes the channel with 4001. A later info unlike the first closes it with 4002 (see above).

Join and reconnect

hello then welcome, then the app opens mediaUrl, seeks to the server position and buffers. A new member is ready=false and joining (with media up); it is in the set at once unless the server is playing, then from its first ready (members present at a swap are put in the set by it), and the gate counts it per "Presence". A reconnect is a hello with the same client and a higher seq (and feedAfter): within the grace it is the same member (welcome.resumed), its token and media URL unchanged, its ready flag and presence kept; after a non-resumed welcome (grace over, server restarted) the client sends its intent again (ready, presence).

Queue (proto 12)

Up next: what plays after the current media, held by the server for everyone, in memory (a restart loses it). Any member edits it; items stay when their adder leaves. by is the adder's lasting member id: it survives reconnects, and while the adder is gone its through/file items wait for it (it is still in members) instead of being skipped; after it left for good they cannot play. Behavior (who puts an item up, the countdown, skipping) is in ARCHITECTURE.md, "Up next".

client {"t":"queue","op":"add","at":"end"|"next","item":{<item>},"thumb":"<base64>"}
       {"t":"queue","op":"move","id":"q3","to":0}
       {"t":"queue","op":"remove","id":"q3","why":"Bob left"}
       {"t":"queue","op":"play","id":"q3"}
       {"t":"queue","op":"skip"}
       {"t":"queue","op":"cancel"}
       {"t":"queue","op":"add","item":{<item>,"auto":true,"for":2}}
       {"t":"swap","media":{...},"queueId":"q3"}
state  "queue":[<item>,...], "queuePlay":"q3"
<item> = {"id":"q3","media":{<swap media>},"how":"direct"|"server"|"through"|"file"|"jellyfin",
          "title":"...","sub":"...","duration":213,"thumb":"<hash>",
          "by":"m2","byName":"Bob","key":"<adder's handle>","auto":true?,"for":2?}
state  "advance":{"id":"q3","at":"<server time>","secs":10}   reserved

• add: item without id, by, byName, thumb (the server sets them: id q<n>, unique for the server's life; who sent it and their name then). at next puts it first, else last. thumb (optional): base64 picture, checked and capped like a poster (256 KiB decoded); item.thumb is its hash, served at /i/ (see "Media art") while the item is queued. title and sub are cleaned like media names and cut to 200 characters (an empty title becomes the media's name); duration (seconds, omitted when unknown) is informational. At most 100 items; over that the sender gets err Up next is full (100 items).
• how and what media must carry:
  • direct: an http(s) url; everyone opens the link.
  • server: an http(s) url and relay {"fetch":true}, on a server with a password (as for a fetch swap). The server keeps the URL to itself: the broadcast media has none.
  • through (a link the adder passes on) / file (the adder's file): name and size only, and key (8 to 64 of [A-Za-z0-9_-]): the adder's own handle for the link or path, which never leave its app. A url sent along is dropped.
  • jellyfin: a jellyfin ref, checked like a swap's (any media with a ref is jellyfin). Anything else: err bad message: ....
• move: id to index to (clamped). remove: drops id; with why (cut to 120 characters) it is a skip (feed queue-skip with the reason; clients show it), else feed queue-remove. add gives feed queue-add. Unknown ids are ignored.
• play: moves id first and sets state.queuePlay to it: a request for the member who can put it up to do so now. skip is play of the first item. queuePlay ends with any swap, when the item is removed, or after 30 s.
• auto (with for, the mediaId it continues): the continuation (the next episode). Only for the current media (else err bad), and only from the member who put it up (media.swapBy) or, when that member is not in, from the first member who is (else err forbidden). At most one: a newer one replaces it; it stays after the other items (an add goes before it, next puts an item first, move never passes it, it does not move itself); any swap that does not put it up drops it.
• cancel: drops the auto item and the countdown for everyone (feed advance-cancel); nothing to drop, nothing happens.
• state.advance (reserved, step 6): the shared countdown when the media finishes and an item is first; at at + secs the server would set queuePlay to id, holding while its performer or source is gone. The server never starts one yet; cancel and any swap clear it.
• swap.queueId: the swap puts that item up and takes it off the queue in the same step. A swap naming an item that is no longer queued is ignored (someone put it up first). For a server item the swap may leave out url: the server uses the one it kept. A swap without queueId leaves the queue as it is.
• state.queue (omitted when empty) and state.queuePlay (omitted when none) ride in every state, welcome.state included. The first item's media.relay shows its warm relay (see "Relay").

Diagnostics (proto 15)

Each app sends its log lines to the server, which merges them with its own into the session log (behavior, file and format: ARCHITECTURE.md, "Logging").

POST /l/<sid>/<token>   content-type application/gzip, body = gzip of:
  {"v":"Starsync 1.4.0 on linux","share":false,
   "lines":[{"t":1767225600123,"l":"WARNING","n":"mpv","m":"..."}],
   "dropped":12,"marks":[{"t":1767225600500,"note":"..."}]}
GET  /l/<sid>/<token>?hours=6

• POST: one batch. Every field is optional. v: the app and version, noted once per member (and when it changes). share:false (omitted when true): the member does not share; the server notes it once and drops any lines. lines[]: t epoch ms on the server's clock (the client converts with its pong offset), l level name, n logger ([A-Za-z0-9_.-], up to 40, else app), m the message, already redacted, an error and stack trace on the lines below (cut to 8192 characters). dropped: lines the client dropped since its last batch (its buffer was full). marks[]: "Mark a problem" at t with an optional note (one line, cut to 200 characters); it goes in even with share:false.
• Limits: a body of at most 256 KiB, unpacking to at most 1 MiB (else 413); not gzip of a JSON object: 400. Per member and minute at most 30 posts and 4 MiB of lines, else 429 with retry-after: the client keeps the batch and sends it later. Unknown token: 401. Success: 204, also on a server that keeps no session log (the batch is dropped).
• GET: the session log of the last hours (1 to 168, default 6; at most 8 MiB, the newest kept), lines the server still holds for ordering included, as gzip of UTF-8 text (application/gzip). Any member may read it (it is redacted and holds no chat). 401 for an unknown token, 404 on a server that keeps no session log.
• Clients send every 10 s and when they leave; each batch carries at most 256 KiB of lines.