Stage Agent Runtime

Product brief: a reusable web framework for live event co-hosts — “Hi Lisa” on a dedicated display, operator-gated, vault-grounded

← CBS 2026 research · Anam · OpenClaw · Multi-agent harness · CBS GTM · Headless Empire OS

Status

Brief / design, not shipped software. Event research lives on /cbs2026avatar. Build handoff for agents lives in-repo at procedures/stage-agent-build-brief.md (pass that file into Grok Build /goal or a coding agent).

One-liner

Stage Agent Runtime is a small web product that puts a named agent persona on a stage display, with a host call-in, operator kill switch, event vault, and a pluggable brain (OpenClaw first — e.g. Lisa Yuson).

Problem

We have agent brains (OpenClaw personas with voice and character) and avatar/TTS vendors, but no event-grade shell: fullscreen presence, operator confidence UI, segment modes, offline fallbacks, and a reusable config for each conference.

Solution (product shape)

Surface	Who uses it	Job
/display/:persona	Stage monitor / projector (kiosk browser)	Avatar video + speaking state; no chrome
/console	Stage operator (tablet, authenticated)	Segments, preview line, green light, kill, switch persona
Gateway	Host mic / push-to-talk / optional wake word	STT → brain; never free-roam without gate in v1
Brain adapter	OpenClaw (v1), later Hermes/etc.	Same interface: prompt in, speech text out
Persona pack	e.g. Lisa Yuson	Face ID, voice ID, soul.md, style rules
Event vault	CBS 2026 instance	Agenda, bios, sponsors, rulings, run-of-show

Naming

Product: Stage Agent Runtime (codename stage-agent)
Category: stage agent / live event co-host
Persona: Lisa Yuson (first pack) — not the product name
First event instance: CBS 2026

MVP (what “done” means for v0.1)

Operator opens console, clicks once to unlock audio on display machine.
Operator selects segment: Welcome | Intro speaker | Sponsor read | Idle.
Agent produces text from vault-locked context; console shows confidence line.
Operator green-lights → TTS + avatar (or portrait + TTS in v0) plays on display.
Kill switch mutes audio and freezes/hides avatar in <1s.
Fallback: play pre-rendered file for same segment if brain/network fails.

Out of scope for v0.1: full free conversation, multi-persona live switching mid-bit, unattended all-day MC, public unauthenticated console.

First instance: CBS 2026 + Lisa

Persona pack: Lisa Yuson — lisayouson.com
Brain: OpenClaw session scoped to CBS vault (continuity with GTM agent story)
Moments: 3 rehearsed segments only for first public use
Research detail: /cbs2026avatar

How to build (agent handoff)

Do not implement from this page alone. Use the private build brief:

procedures/stage-agent-build-brief.md

In Grok Build (or any coding agent):

# 1) Open the monorepo / new repo as agreed in the brief
# 2) Paste or @ the file procedures/stage-agent-build-brief.md
# 3) /goal with done-means from that file (Phase 0 scaffold first)

Prefer Morgan-style flow: wide design already on this site → /goal only the next phase (workflows + goal).

Non-goals

Replacing human hosts or production managers
Live financial/legal advice from stage
Cloning real humans without contracts
Depending on venue Wi‑Fi with no fallbacks

CBS 2026 avatar co-host research
Anam + OpenClaw avatars
OpenClaw · CBS GTM (Lisa in stack)
Multi-agent harness map
Knowledge base / vault · HEOS
CLI for agents — task-shaped tools under the brain