AI
AI in BPShowServer is practical show help on the same show graph as Action Deck — not a side chatbot product. Keep the five stories separate; do not mash them into one AI hero. Guided walkthrough: Getting Started · 06.
Real Community 1.0.190 Demo Grok stills live on Getting Started · 06.
Five stories
| Story | What operators get | Path (1.0.190) |
|---|---|---|
| Voice LLM | Voice on intercom — page crew, ask room status, confirm cues you already trust | AI → Assistant and/or I/O → Add Endpoint → Voice LLM (voice-llm); Action Deck Add button → Press → intercom / Voice LLM listen; park off page 1 until trusted |
| ElevenLabs SFX / TTS | Short generated audio that lands in Library / cartwall | I/O → Add Endpoint → ElevenLabs (API key), then Control → Audio → SFX Generation → Generate → Library / Add to Cartwall |
| On-prem MCP | Local MCP assistants drive UI / OSC / HTTP / CLI / Action Deck on the LAN box | AI → MCP (/mcp) — title MCP Server; preferred for Community Free |
| Cloud MCP | Remote MCP against a Cloud-capable ShowServer you already operate | Same MCP field family; see API / MCP and Remote / Cloud studio |
| Plugin Builder | Ask AI to script a bridge; review under Automation | AI → Plugin Builder (/plugin-builder) |
Voice LLM (intercom)
Enable Voice LLM from AI → Assistant and/or I/O → Add Endpoint with provider Voice LLM (voice-llm). Bind an Action Deck tile: empty cell → Add button → Label → Actions → Press → intercom / Voice LLM listen action.
Use it to:
- Ask the room status
- Page operators
- Fire or confirm cues you already trust on Action Deck
Sample (Blue Lamp Theater / Blue Lamp — Theater Show when built or loaded): AI / Voice tile + Cartwall Alerts → Intercom Chime. Fresh Default does not guarantee those tiles — add them in Getting Started · 06.
ElevenLabs SFX / TTS
- I/O → Add Endpoint → ElevenLabs (API key).
- Control → Audio → SFX Generation — this is not Cloud SFX.
- Generate, then send output to Library and/or Add to Cartwall.
First-timers still upload via Control → Audio → Library → Upload.
On-prem MCP vs Cloud MCP
On-prem MCP (preferred for Community)
Open AI → MCP (/mcp). Page title MCP Server. Configure:
- Enable MCP server
- MCP URL
- Transport
- Require bearer token
- Access token
- Tool policy: No tools / Code Mode / Full catalog / Custom
Point a LAN MCP client at the MCP URL on that page. Prefer tools that fire existing Action Deck / Automation cues.
Cloud MCP
Already operate a Cloud-capable ShowServer? Connect Cloud MCP against that deployment’s MCP endpoint — same show graph as on-prem. Prefer on-prem MCP for Community Free from Download.
Thin reference: API / MCP overview.
Plugin Builder
Open AI → Plugin Builder (/plugin-builder). Capability inside the suite — extend the show graph when you need a custom bridge.
- Ask AI to script a bridge for your site.
- Review and own the result under Automation.
- Keep Twinkly + WLED as first-class RGB(W) endpoints in I/O; use Plugin Builder when you need something else (for example a Light-O-Rama bridge).
Seasonal path: drive Twinkly + WLED from I/O today; open Plugin Builder when a custom controller path belongs on the same graph.
Guardrails
- Keep AI / Voice tiles off Action Deck page 1 until operators trust them.
- Prefer assistants that call existing cues over inventing parallel cue lists.
- Run the box locally, then walk Getting Started — prefer Demo Grok stills on Getting Started · 06 over inventing new AI UI.
- Prefer on-prem / LAN MCP for Community; Cloud MCP only if you already operate a Cloud-capable deployment.