AI Modes Guide

Start With The Job

Do not start by picking a model. Start by deciding what the AI should do for you.

Job	Use	Best For
Public chat bot	bot.html	On-stream bot replies, optional TTS, public chat responses.
Private helper bot	chatbot.html	Testing prompts, one-on-one help, private bot behavior.
AI cohost	cohost.html plus cohost-overlay.html	A visible or audible assistant controlled by the streamer.
AI overlay builder	aiprompt.html plus aioverlay.html	Generating and previewing overlay HTML from prompts.
Moderation or RAG	AI settings in the popup	Blocking/flagging messages or answering from your knowledge base.

1. Choose A Provider

Social Stream Ninja can use local or cloud providers. Local can be private and cheap, but it can be slower or harder to install. Cloud providers are usually easier, but accounts, API keys, and costs are controlled by the provider.

• Ollama Native API: good first local option if Ollama is already installed.
• Custom API: use for OpenAI-compatible local servers such as llama.cpp, LM Studio, vLLM, or similar tools.
• Hosted providers: OpenAI, Gemini, DeepSeek, xAI, Groq, OpenRouter, Bedrock, and similar services can change pricing and model names outside Social Stream Ninja.
• Browser/local model pages: useful for advanced local experiments, but model assets and browser limits can make setup heavier.

2. Chat Bot Basics

1. Configure an AI provider in the popup.
2. Enable Chat Bot - Primary when the AI should evaluate captured live-chat messages.
3. Use bot.html when you want an overlay or TTS output; use chatbot.html only for the separate private one-on-one bot.
4. Use the same session as your dock and overlays.
5. Test the Primary bot in overlay-only mode before letting it answer platform chat.
6. If the bot sends messages back to a platform, confirm that platform supports send-back and that the account is allowed to post.

Step-by-step setup and troubleshooting: AI Chat Bot Setup Guide.

3. AI Cohost

The cohost has a control side and an output side. The streamer controls it from the dock or cohost.html. OBS captures cohost-overlay.html when you want the cohost on stream.

cohost.html?session=YOUR_SESSION

cohost-overlay.html?session=YOUR_SESSION&tts

More detail: AI Co-host Guide.

4. AI Overlay Builder

The builder is for making an overlay, not for capturing chat by itself. Use aiprompt.html to describe and test the overlay. Use aioverlay.html as the OBS output page after it is saved.

1. Open the builder and describe the overlay you want.
2. Preview it with sample chat, follow, sub, donation, or raid events.
3. Save the overlay.
4. Add the matching aioverlay.html URL to OBS with the same session and overlay name.

5. Moderation And Knowledge Search

AI moderation and RAG search are support tools, not magic. Keep a human review path for anything that could ban, delete, block, or publicly answer viewers.

• Use non-blocking moderation first so you can see what the AI would flag.
• Keep knowledge base text short, current, and specific to your stream or product.
• If answers are wrong, fix the source material before changing random prompt wording.
• If a provider times out, test with a smaller model or a simpler prompt before debugging the whole overlay.

Common Fixes

• No response: confirm the AI provider, API key, model name, and endpoint.
• Wrong session: make sure the dock, bot, cohost, and overlay all use the same session.
• Cohost overlay blank: open cohost-overlay.html and check the label if multiple overlays are running.
• Overlay builder result not showing: confirm the builder page saved the overlay and OBS is pointed at the matching aioverlay.html page.
• Bot cannot post: check platform send-back support, login state, permissions, and rate limits.

Full reference: Commands & API AI Integration.