The agent architecture behind every ReactLive event

Five agents.
One conductor.

The team behind every ReactLive event.

Setup. Protect. Engage. Answer. Report. Five specialists run by one orchestrator — covering the event before, during, and after. Freeing your team to focus on what really does need their input.

Join the waitlist Meet the team

1 orchestrator · 5 specialists · 3 phases · 4 strength levels

The conductor

The Boss runs
the room.

The Orchestrator is the main event agent. It reads your SOUL.md to understand how the event should feel — tone, voice, red lines — then routes context between the specialists, holds the timeline, and decides who acts when. Each of the five agents below reports to it.

Main event agent
Orchestrator
aka Core AI
aka Conductor
aka Producer Agent
aka The Boss

Sets the tone for the whole event from a single SOUL.md — voice, vocabulary, red lines, escalation rules. Coordinates context flow between the five specialists, holds the timeline, decides which agent acts when, and escalates anything outside its remit to the human moderator.

SOUL.md → Setup → Protect → Engage → Answer → Report

The specialists

Three phases.
Five agents.

Each phase of an event has different demands. Different agents step forward at different moments — but the context is shared end-to-end.

Before
During
After
01 Before

Setup Agent

aka Launch Agent
Goal

Prepare the event.

Core skills
  • →Event configuration — Q&A, polls, moderation rules
  • →Content ingestion — docs, agenda, speakers
  • →Workflow setup — defines agent levels & rules
  • →Pre-event suggestions — best setup by event type
  • →Integration handling — streaming, tools, embeds
Dataevent type, docs, agenda, templates
Toolscreate settings, enable flows, apply presets
Rulesonly auto-apply safe defaults
Superpower

Turns event setup from hours into minutes.

02 During

Protect Agent

aka Moderate · Shield
Goal

Keep the event safe.

Core skills
  • →Content classification — spam, abuse, off-topic
  • →Risk detection — flags sensitive or harmful
  • →Policy enforcement — applies event rules
  • →Real-time filtering — blocks or flags instantly
Datasubmissions, policies, event context
Toolshide, flag, queue, escalate
Rulesonly auto-block high-confidence abuse
Superpower

No chaos, only safe interaction.

03 During

Engage Agent

aka Content · Pulse
Goal

Drive interaction.

Core skills
  • →Priority ranking — surfaces best questions
  • →Clustering — groups similar submissions
  • →Moment detection — topic shifts & drops
  • →Content triggering — polls, prompts, highlights
  • →Audience sensing — engagement in real time
  • →Timing optimization — acts at the right moment
Datatranscript, engagement signals, docs
Toolspublish poll, highlight question, push CTA
Rulesdon't interrupt key moments
Superpower

Keeps the event alive and moving.

04 During

Answer Agent

aka Echo Agent
Goal

Answer the questions.

Core skills
  • →Question understanding — detects intent
  • →Answer retrieval — docs + live transcript
  • →Answer generation — clear, contextual replies
  • →Dedup awareness — recognises repeats
  • →Confidence scoring — answers vs escalates
Datatranscript, docs, prior Q&A
Toolsreply privately or publicly, suggest
Rulesonly answer if grounded in trusted sources
Superpower

No question goes unanswered.

05 After

Report Agent

aka Insight Agent
Goal

Explain what happened.

Core skills
  • →Data aggregation — all event interactions
  • →Insight extraction — trends, topics, sentiment
  • →Summary generation — clear event recap
  • →Audience analysis — engagement, drop-offs
  • →Content highlights — best Qs, key moments
Datatranscript, Q&A, polls, engagement
Toolsgenerate report, charts, highlights
Rulesseparate facts from interpretations
Superpower

Turns event data into clear, actionable insight.

"This feels more like a team."

Dial it up or down

Off, Suggest,
Assist, Auto.

Every agent has four strength levels. Start at Suggest for your first event. Climb to Auto when you trust the team. Adjust per agent, per event.

Level
Protect Agent
Engage Agent
Answer Agent
Off
Disabled.
No AI filtering, just basic moderation.
No AI-driven engagement.
No AI answers.
Suggest
Recommends only.
Flags risky or spam content. Moderator decides.
Suggests polls, prompts, and questions to highlight.
Suggests answers to the moderator.
Assist
Auto on safe actions.
Auto-blocks obvious spam & abuse. Flags edge cases for review.
Auto-publishes low-risk engagement. Asks for approval on key moments.
Auto-answers simple or repeat questions. Escalates complex ones.
Auto ●
Full autonomous.
Fully moderates. Blocks, filters, and prioritises automatically.
Fully drives engagement. Surfaces questions, runs polls, manages flow in real time.
Answers all questions instantly.

Setup and Report agents have their own controls — Setup runs once before the event, Report runs once after. The matrix above covers the three live agents that act in the room.

1
Orchestrator coordinating the room.
5
Specialist agents, each with one job.
4
Strength levels, set per agent per event.
0
Black boxes. Every agent is configurable.

FAQ

Questions
about agents.

Want to know more? Ask us directly.

01Why split into five agents instead of one big AI? +
Specialisation gives you control. Each agent has one job, one set of data, and one set of rules — so you can dial each one up or down without affecting the others. Want to autonomously moderate but only suggest answers? Set Protect to Auto and Answer to Suggest. One model, one policy can't do that without a lot of prompt gymnastics.
02Does the Orchestrator override my moderator? +
No. The Orchestrator coordinates between agents — it doesn't outrank humans. Your moderator can override, edit, or withhold anything any agent proposes, at any strength level. Auto means the agent doesn't wait for approval; it doesn't mean the human can't intervene.
03Can I run an event without the AI agents? +
Yes. Set every agent to Off and ReactLive becomes a clean, fast Q&A + polls + reactions platform with a live transcript and human moderation tools. Flip individual agents back on for a single event whenever you want to see what they'd do.
04Where does the Answer Agent get its information? +
Only from the trusted sources you give it: documents you upload, the live transcript, and previous Q&A from the same event. It will not draw on the open web, model priors, or guesses. If a question can't be answered from those sources at the confidence level you set, it gets escalated to the moderator.
05Can I write my own agents? +
That's on the roadmap. The MCP connector will let you call any of the five agents from your own stack, and we're working on a way for advanced users to register their own specialist agents into the Orchestrator's flow. Waitlist members get early access when it ships.

The waitlist

Your next event
with an AI in the room.

Drop your email and we'll be in touch when early access opens. Waitlist members get three events free and locked-in pricing.

NO CREDIT CARD · ROLLING BETA · DOGFOODED AT OUR OWN TOWN HALLS