Weclank is a local-first streaming studio with AI co-hosts. The app moves from splash to local account login to the studio, where a user builds scenes, adds participants, produces a broadcast canvas, and optionally sends that broadcast to local recording or RTMP destinations.
Participant — a local source that can appear in scenes. Participants include camera, screen, mic, voice agents, avatar agents, and text assistants.
Scene — an ordered composition of participant placements. A scene decides which participants are on the broadcast canvas, their geometry, their visibility, and their z-order.
Source placement — a participant's normalized rectangle inside a scene. Placement coordinates are ratios of the broadcast canvas so scenes survive resolution changes.
Scene composition — the rules for source placement, z-order, visibility, hit-testing, layout presets, and backstage projection.
Backstage — participants that are not currently visible on the active scene, either because they are absent from the scene or present but hidden.
Participant runtime — disposable resources attached to a participant at runtime: media streams, TTS providers, audio mixer inputs, renderers, and banter sessions.
Agent turn — one decision cycle for an AI co-host: consume an event, apply gates and context, call the LLM/tools, emit a reply, and optionally speak.
Broadcast capture — the MediaRecorder pipeline that captures the broadcast canvas plus mixed audio for local recording or RTMP egress.
Broadcast overlay plane — the ordered graphics pass drawn on top of participant tiles before capture: chat, generated overlays, QR codes, and captions.