Streaming & podcast

Always-on hosts, unscripted duos, and scenes full of characters.

IAMX characters were built for more than one-on-one kiosk chats. The same pipeline that drives a single interactive host scales up into three long-form formats: a solo character that streams live for hours, a pair of characters that hold an unscripted conversation with each other, and whole scenes where several distinct characters share a level. This page covers all three, the Blueprint surface that drives them, and the operator tooling that keeps a live broadcast under control.

AI stream host

An AI stream host is a single IAMX character configured to perform continuously — greeting new arrivals, reacting to chat, filling quiet moments with its own remarks, and never dropping the persona. Because the character is always live, the important design problem is not "how does it talk" but "how do you decide what reaches it." That is what the livestream queue solves.

How the moderated queue works

Every incoming viewer message is pushed onto a queue rather than spoken immediately. The character pulls from the front of the queue when it is ready for the next turn, so audio never overlaps and the host is never interrupted mid-sentence by a flood of chat. Between the raw viewer feed and the character sits an operator console — a moderator can approve, hold, prioritize, reorder, or inject messages while the character keeps performing.

  Viewers ──▶  [ incoming feed ]
                     │
                     ▼
             ┌──────────────────┐      approve / reject
             │  OPERATOR CONSOLE │◀───  reorder / prioritize
             └──────────────────┘      inject operator note
                     │
                     ▼
             [ livestream queue ]  ──▶  Character speaks next turn
                     ▲
                     │  On Livestream Queue Changed  (UI refresh)
                     ▼
                Operator UI

This gives you a live broadcast that stays on-brand even when chat gets rowdy: off-topic, abusive, or duplicate messages never reach the character, and the moderator can hand-pick which questions get answered next.

Blueprint surface

Everything is driven from four Blueprint members on the character component. Wire your chat ingestion (a websocket, a platform chat API relay, a REST poller — whatever feeds you messages) into Enqueue Livestream Message, and build your moderator panel around the change event.

1. Ingest — Relay each viewer message into Enqueue Livestream Message with the viewer's name and text.
2. Moderate — Bind On Livestream Queue Changed to redraw your operator list; expose approve/reorder controls that manipulate the queue.
3. Steer — When you want the host to say something now, call Enqueue Operator Message so it leapfrogs the viewer backlog.
4. Reset — At segment or ad boundaries call Clear Livestream Queue so the host starts the next block fresh.

Podcast mode (NPC to NPC)

Podcast mode turns two IAMX characters toward each other and lets them talk — unscripted. You set a topic and press go; from there each character responds to what the other just said, in character, with its own personality, its own voice, and its own opinions. Nothing is pre-written. The result is a back-and-forth that can run for as long as you like, that you can steer at any moment, and that your audience can join live.

Two distinct minds

Each participant is a full character in its own right — a separate persona, a separate voice, and its own stance on the subject. Because their personalities and knowledge differ, they naturally disagree, build on each other, and change direction. Give one a skeptic's brief and the other an enthusiast's and the same topic produces a genuine debate rather than two voices agreeing politely.

Blueprint surface

Event BeginPlay
   └─▶ Start Podcast(
          Topic       = "Will kiosks replace call centers?",
          HostA       = Character_Ava,
          HostB       = Character_Rook )

// later, live, from your chat relay:
Event OnViewerQuestion
   └─▶ Inject Audience Comment(
          From = "mert_92",
          Text = "But what about older customers who hate touchscreens?" )

1. Cast two characters — Place two IAMX characters in the level, each with its own persona, voice, and knowledge. See /docs/multi-character for per-character setup.
2. Set the topic — Call Start Podcast with a subject line. Keep it a single, opinionated prompt; the hosts do the rest.
3. Let it run — The characters alternate automatically. No timers to manage; each turn is triggered by the other finishing.
4. Bring in the audience — Route live chat into Inject Audience Comment so viewers can nudge the discussion in real time.

Multi-character scenes

A level is not limited to one character. You can place as many IAMX characters as your scene needs, and each is fully independent — its own personality, its own voice, its own language model choice (GPT, Claude, Gemini, or a local model via Ollama), and its own knowledge base. A reception desk agent, a technical specialist, and a greeter can all coexist in the same space, each an expert in its own lane.

Per-character independence

Proximity gating

When several characters share a space, the obvious risk is that everyone answers at once. IAMX solves this with proximity gating: only the character you are near hears you. Walk up to the specialist and the specialist responds; step over to the greeter and the greeter takes over. The others stay quiet until you are within their range. This keeps conversations clean without any manual "select active character" step — physical position is the selector.

          ( you )
             │  in range of Ava only
             ▼
   [ Ava ]◀──●        [ Rook ]         [ Sage ]
   hears you          idle             idle
   & replies          (out of range)   (out of range)

Characters talking to each other

Multi-character scenes aren't limited to human-to-character exchanges. The same NPC-to-NPC capability that powers podcast mode works between any characters in the level, so your cast can hold conversations with one another — a specialist can hand a customer off to a colleague, two greeters can banter during quiet periods, or a panel of characters can discuss a topic while a visitor listens in and occasionally chimes in via proximity.

Putting it together

These three formats are layers of the same system, and they stack:

• Solo host — one character plus the livestream queue and operator console for a moderated 24/7 broadcast.
• Duo — two characters plus Start Podcast and Inject Audience Comment for an unscripted call-in show.
• Ensemble — many characters, proximity-gated for walk-up interaction, with NPC-to-NPC conversation for ambient life, handoffs, and live panels.

Because every layer is exposed through plain Blueprint calls and events, you can prototype in an afternoon and wire it into whatever front end you already run — a stream relay, a kiosk, an event stage, or the hosted panel at iamx.live.

Next steps