Features

Multi-agent debate

Two or three perspective agents debate visibly before a moderator synthesizes the final answer.

Multi-agent debate

Trigger a debate thread where multiple agents argue assigned perspectives. A moderator agent posts a synthesis when rounds complete.

Console

Agents → Debate (/agents/debate)

Seed default roles (optimist, skeptic, moderator) or create custom roles.
Configure trigger pattern (regex on user prompt), system prompt, and max rounds (default 2).
Run debate from console or trigger from chat when the pattern matches.

In chat, debates render as grouped agent_step events in DebateThreadPanel with participant_role: "debate".

API

GET  /admin/agent-debate/roles
POST /admin/agent-debate/roles
POST /admin/agent-debate/roles/seed
PATCH /admin/agent-debate/roles/:roleId
DELETE /admin/agent-debate/roles/:roleId
GET  /admin/agent-debate/sessions?roomId=
POST /admin/agent-debate/run

Run body: { roomId, prompt, maxRounds? }.

Behaviour

Token cost telemetry per debate session (latencyMs, step count).
Timeout produces a partial consensus message from the moderator role.
Events stream on the room WebSocket like normal agent steps.

When to use

High-stakes decisions, policy review, or UX where users benefit from seeing reasoning paths before the final answer.

Empathy layer (voice)

Opt-in prosody detection adapts agent tone and escalation without telling the user they seem stressed.

Ambient agents

Event-driven agents that react to webhooks and room events within policy bounds, not only @mention.

On this page

Multi-agent debate Console API Behaviour When to use Related