Three layers, one governance model, any agent framework underneath.
{{ l.desc }}
Composable objects
{{ o.def }}
e.g. {{ o.example }}
Capability layer
{{ b.desc }}
The adapter layer
AgentSpec, CoordinationSpec, StateAdapter, HITLAdapter, and GuardrailAdapter sit under ELIZON's own object model — normalizing agent definition, coordination, state, human approval, and guardrails across whichever framework you build in.
Modalities
Any Agent, Team, or Workflow can answer the phone. Real-time voice over LiveKit, reachable by SIP trunk or WebRTC, with STT/TTS providers (ElevenLabs, Cartesia, Deepgram, Google, Azure) resolved as typed slots — no provider lock-in.
Video face agents extend the same pipeline with a rendered avatar on the stream — same agent, same governance, sub-700ms conversational target.
Deployment
{{ d.desc }}
Design partners get hands-on before the September 1 launch.
Get early access