====== AI assistant: what the one key gets you ====== The optional AI configuration in **Settings -> AI assistant**, or in **Guided setup**, is shared by the public AI features below. Configure a provider, your own API key and a suitable model once. It is not a StreamHelpers account key, Pro pairing key or an included AI allowance: usage is charged directly by the provider under its own terms. AI is optional. A rig can stream, switch cameras, use overlays and run manually prepared audience games without it. ===== Features using the shared configuration ===== ^ Feature ^ Where to use it ^ What it does and sends to the provider ^ | **Tell me about this place** | GPS | Explains the current location or a selected nearby place in short, read-aloud points. Uses GPS, the chosen subject, local date/time and available nearby-place context. Shows confidence rather than guaranteeing the location or facts. | | **What am I looking at?** | Trivia, Chat, Companion and the optional header chip | Answers about a camera still, optionally with a question. Sends the picture, GPS, local time and available nearby-place context. Answers appear for the operator first; **Put on screen** is the separate choice to air one. Recent answers and their stills are retained on the rig for reuse. | | **Trivia writing and place suggestions** | Trivia | Writes question/answer batches about the current place, a typed subject or selected nearby rides, shows and dining. **What is here?** can supplement known places with AI suggestions. Sends the selected context, writing options and existing questions to reduce duplicates. Generated questions go into the bank; the operator chooses how to run them. | | **Joke writing** | Jokes | Writes jokes about a subject or current location into the queue, with the selected rating, style and language. Sends writing options and existing jokes; location mode can include GPS, nearby places and cached weather. Check generated material before using it on air. | | **Parade arrival estimate** | Timer's parade controls | Estimates the minutes between the parade's published step-off and its arrival at the operator's position. Sends the selected parade, location and available route context. It can decline to estimate; the manual minutes field remains available. This is an estimate, not live tracking. | | **Read the answer off it with AI** | Closest guess, including its Companion workflow | Reads a number from an answer picture such as a price tag, menu or scale when no answer has been typed. Sends the answer image and game question. The number remains the game's secret answer until **Reveal**; inspect it first and type a correction if needed. The same reader is used when a macro prepares a closest-guess answer picture. | These are separate uses of one configuration, not one combined request. They do not automatically make every generated answer or picture public. Game queues, reveal controls and on-screen commands still determine what appears on the stream. ===== Models and feature overrides ===== Text features need a model that supports the selected provider's text interface. **What am I looking at?** and reading a closest-guess answer picture require a **vision-capable model** that accepts image input. A working text model or API key does not prove that pictures are supported. Select a suitable model in Settings; the provider's default may be text-only. **Trivia** and **Jokes** normally inherit the dashboard's AI configuration. Each can instead have its own provider, key, model and compatible API URL. An explicitly selected feature provider needs its own working key; it does not silently borrow the dashboard key. Switch back to the dashboard settings to use the shared configuration again. An **Other (OpenAI-compatible)** provider requires a compatible API URL and model. Model availability, image support, rate limits, pricing and retention are controlled by that provider. Loading a provider's model list also contacts the provider. ===== Privacy, cost and review ===== The saved key stays on the router and is used to authenticate requests to the configured provider. AI requests can disclose precise GPS, selected subjects, pictures and contextual data described above. Avoid private documents, faces, credentials or other sensitive material in submitted images or questions, and review the provider's privacy and retention terms. Trivia answer matching, scoring and joke queues do not need to send viewer chat to the AI provider. Video encoding, bonding, destination sign-in, GPS collection, map imagery, weather collection and ordinary overlay rendering are not services purchased by this AI key. Joke delivery/voice playback is separate from AI joke writing. Requests consume provider usage when an AI action is invoked, including AI place suggestions and opted-in answer-picture reading. Repeated generations or retries can cost more. Missing keys, incompatible models and provider failures need attention; manual subjects, questions, jokes, answers and parade offsets remain useful alternatives. AI can misidentify a picture, invent a fact or return an unsuitable joke or timing estimate. Review generated material and the secret closest-guess answer before using it on air. Rating filters and confidence labels are safeguards, not guarantees. ===== Public demo ===== [[https://streamhelpers.com/demo/|The interactive demo]] shows controls and fictional results without calling an AI provider or accepting a real key. **What the one key gets you.** in its Guided setup opens this guide in a separate tab. The guide describes real-rig features; it does not make AI functional in the demo. See [[dashboard:settings#ai-assistant|AI settings]], [[dashboard:audience-games|Audience games]], [[dashboard:gps-and-weather|GPS, weather and timers]] and [[companion:help|Companion help]].