Bring your own stack
One config, any provider on every layer.
Pick a voice per agent. MiniMax and ElevenLabs out of the box, or any OpenAI-compatible /audio/speech server — including fully local voices. Chain them for automatic fallback.
MiniMaxt2a_v2
ElevenLabscloud
OpenAI voicescloud
Kokoro-FastAPIlocal
openedai-speechlocal
LM Studiolocal
lan scan
2 found
dante
main · chat, tools
found
mira
main · chat
found
Zero-config peers
Point it at a subnet. Get back agents.
/v1/agents/scan expands a CIDR or IP range — up to 1024 hosts — and probes every one concurrently for an A2A agent card. Only hosts that answer come back, with name, description, and skills already parsed and ready to drop into agents.yaml.
Already know the addresses? /v1/agents/discover checks an explicit peer list instead. Both read the same /.well-known/agent-card.json that OpenClaw and Hermes peers publish.