Browser and SFU validation
Broaden target-browser, codec, NAT, TURN, separate-host, and failure coverage for mediasoup and the controlled first-party media path.
Expand the participant fabric without blurring what exists today.
The roadmap builds outward from a stable room, media, identity, and operations core. Priorities change when production validation and design-partner requirements reveal better sequencing.
priority
nowProduction confidence
nextMedia workflows
thenAI voice + intelligence
proof
validateBrowsers · SFU · TURN
extendRecording · output · replay
completeVoice · DSP · agents
Direction, not datesProduction validation firstLimits stay visibleDesign partners inform orderStatus is source of truth
Increase confidence in production paths.
The highest-value work improves reliability, evidence, and operational confidence in the paths customers already use.
Complete live capture operations
Finish live-room worker wiring, durable artifact storage, recovery, retention, and production acceptance paths around the existing recording foundations.
Durable multi-instance state
Turn room-registry, admission, revocation, monitoring, and regional-routing abstractions into validated shared-state deployment paths.
Native media parity
Continue Swift, Kotlin, React Native, Flutter, and C++ media integration and target-device verification without overstating cross-platform parity.
Turn bounded media slices into coherent product paths.
The frame bus, scene runtime, batch/live composition, and broadcast control/composited-feed APIs are real today. The next work completes stock end-to-end broadcast validation, portability, semantic replay, and deployment hardening.
Scene runtime and composition
Demand-driven frame taps, scene request/update/release, batch and live workers, and grid, speaker, PiP, and custom layouts.
Broader portability and output support
Live composition supports mediasoup and the integrated engine; batch remains first-party-only. More codecs, GPU backends, cross-platform seed calibration, declarative scenes, per-viewer outputs, and broad production validation remain roadmap work.
Declarative scenes
A higher-level scene description, overlays, captions, policies, and application SDK methods instead of direct low-level protocol envelopes.
Broadcast control and semantic replay
RTMP, SRT, HLS, and LL-HLS control-plane and composited-feed APIs have shipped. Stock consumer-receives-output validation, SIP integrations, multi-aspect replay, and context-aware recording renders remain roadmap work.
Complete the AI voice and media-intelligence value chain.
Local TTS and presence
Browser-native speech, Piper-WASM, publishable tracks, streamed fragments, speech lexicons, VAD events, and agent presence surfaces already provide reusable pieces.
Conversational voice agent
Server-side turn-taking, barge-in, cancelable streaming STT→LLM→TTS, room audio publication, transcript state, latency events, and per-turn traces.
Realtime surfaces
Interactive canvas, shared-object scenes, publishable tracks, semantic sidecars, replay, and headless commands are shipped. A dedicated PMG adapter remains roadmap work.
Media DSP engine
Shared Rust/WASM and native signal-processing primitives for explainable audio, video, screen, and network readiness events tied to room context.
What earns roadmap priority.
Production validation before breadth
Hardening existing browser, SFU, TURN, recording, and operational paths takes priority over presenting architecture sketches as complete products.
Demand-driven media processing
The SFU remains the default transport plane. Decode, compose, analyze, or broadcast only when a requested workflow needs it.
Provider and deployment choice
AI, storage, telemetry, and cloud services remain pluggable so customers can choose local, private, managed, or custom integrations.
One honest room model
New SDKs, participants, surfaces, and workflows should share identity, permissions, presence, state, telemetry, and lifecycle instead of creating parallel systems.