WebRTC Room
Hosts, guests, mentors, and members join from web or mobile using Decoded TURN/STUN and a server-side media edge.
Creator Infrastructure
Run streams with guests, destinations, and realtime AI co-hosts.
Organizations, professionals, mentors, and users can host Decoded streams, push one program feed to RTMPS destinations, and invite a voice agent that understands the topic, audience questions, and grounded knowledge base.
The voice agent joins through WebRTC or the FreeSWITCH bridge, streams audio through a self-hosted realtime model, speaks with a consented cloned voice, listens to member questions and ideas, checks the room knowledge base first, and can use policy-gated web tools for live fact checking.
Workflow
How the stream is assembled
Program Feed
A media worker composes scenes, screen share, voice, captions, agent audio, and correction prompts into one encoded feed.
RTMPS Fanout
The composed feed is pushed to YouTube, LinkedIn, Facebook, X, or any custom RTMP destination.
Grounded Agent
The agent uses uploaded PDFs, added URLs, transcripts, and live questions to answer, challenge claims, and keep the conversation on topic.
Live Studio Workflow
Run a multistream with a realtime grounded voice agent
1. Room
Create the WebRTC room and FreeSWITCH bridge that hosts, guests, mentors, users, and agents join. All members can ask or bounce ideas to the agent.
2. Destination
Add RTMPS destinations. The stream key is converted to a secret reference and is not returned raw.
3. Voice Agent
The agent joins via WebRTC, streams audio in/out through the self-hosted realtime model, checks the room knowledge base first, then uses web tools when uncertain.
4. On-stream design
Lower thirds and logos are rendered into one composed mezzanine feed before RTMPS fanout, so every destination receives the same branded program output.