Rooms and lifecycle
Join, leave, presence, liveness, reconnect, cursor replay, media repair, quality-gated recovery, and stale-participant cleanup.
People, AI agents, robots, devices, applications, and services join one room model. Identity, media, state, commands, recovery, and operations stay together.
Adopt the stable media and signaling paths first, with explicit alternatives for room size, infrastructure policy, and deployment requirements.
Join, leave, presence, liveness, reconnect, cursor replay, media repair, quality-gated recovery, and stale-participant cleanup.
Mesh audio/video for small rooms and a production mediasoup SFU path for larger topologies, with screen share and file-backed media sources.
First-party relay infrastructure with tenant-aware metering, dynamic credentials, regional configuration, and relay validation tooling.
Chat, receipts, typed signals, schema-validated data channels, chunked files, compression, and participant-aware routing.
Coordinate participants and product interfaces without funding and maintaining a separate real-time state system for every workflow.
Ephemeral typed events and per-participant last-write-wins state with late-joiner snapshots for cursors, device status, UI state, and workflow events.
Transport-neutral room objects for whiteboards, annotations, controls, lightweight application state, and agent-visible context. Server snapshots and persistence remain roadmap work; the model is per-key LWW, not a collaborative-text CRDT.
Turn room state, chat, telemetry, and external data into deterministic canvas surfaces that can be interactive, published as tracks, recorded, and replayed semantically. A dedicated PMG adapter remains roadmap work.
Audio-reactive AI personas, capture-stream virtual cameras, speech lexicons, browser-native speech, and self-hosted Piper-WASM output with a publishable audio track.
Workers, devices, native applications, and robots gain the same governed identity and room context as human participants.
Node and Python worker adapters for transcription, translation, moderation, analytics, and custom AI; signaling/data clients for Go, Rust, and .NET.
MQTT v5, serial, BLE, and USB sidecars translate external device traffic into the same typed protocol envelope used by rooms.
React Native, Flutter, Swift, Kotlin, and C++ paths share generated protocol types and signaling APIs. Media maturity varies by runtime.
ROS2 bridges, teleoperation controls, fleet rooms, telemetry, command topics, services, and WHIP/SFU camera integration for guided deployments.
The SFU remains the transport plane. Selected streams branch into explicit processing only when a product experience requests a scene or workflow.
Room-scoped scene request, update, and release messages start pluggable workers on demand and tear them down when no longer needed.
Grid, active-speaker, PiP, and caller-defined layouts run through the Rust/GStreamer compositor path, including live layout updates and composited producer output.
Bounded live composition supports mediasoup and the integrated engine with persistent pipelines, no-restart track changes, scene updates, and output published back into the room. Batch composition remains first-party-only.
RTMP, SRT, HLS, and LL-HLS control-plane and composited-feed APIs have shipped, but stock end-to-end consumer validation is still open. Broader codecs, GPU backends, declarative scenes, per-viewer fan-out, SIP, and durable semantic renders remain roadmap work.
Identity, observability, policy, support context, and usage data remain part of the platform rather than becoming downstream operational debt.
JWT, JWKS, OIDC, OAuth introspection, RBAC mapping, waiting-room tickets, moderator approval, abuse limits, and targeted token revocation.
Peer state, ICE path, topology, quality trends, redacted support bundles, consented connectivity checks, local bundle evaluation, and repeatable RTC test matrices.
OpenTelemetry, Prometheus, tenant-labeled usage records, a reference rollup consumer, and a region-aware illustrative metering dashboard. Durable billing stays customer-owned.
Room and participant controls, worker visibility, regional capacity, process pressure, live relationships, SFU diagnostics, and TURN totals.