Choose between a voice cohost, live-chat bot, private helper, browser companion, moderation, translation, knowledge search, and AI-built overlays.
Choose the Result You Want
I Want To...
Use
API Key?
Talk to an AI that can also hear live-chat questions and answer aloud
AI cohost: cohost.html
Depends on provider
Use an open ChatGPT browser tab as an experimental companion
ChatGPT browser source
No
Automatically answer selected YouTube, Twitch, Kick, or other chat messages
Primary AI chat bot
Local or cloud provider
Show or speak AI answers without posting them back into chat
Primary bot + overlay-only + TTS
Local or cloud provider
Have a private one-on-one helper
chatbot.html
Local or cloud provider
Filter, translate, summarize, or answer from documents
Moderation, translation, summaries, or RAG
Local or cloud provider
Describe an overlay and have AI build it
aiprompt.html + aioverlay.html
Local or cloud provider
Two Different Meanings of “ChatGPT”
Option
What It Is
Conversation
ChatGPT API provider
SSN sends prompts to OpenAI using your API key and selected model.
An SSN/API conversation, separate from chatgpt.com.
ChatGPT browser source
The extension reads new User and ChatGPT turns rendered in an open chatgpt.com tab.
The browser conversation you opened; no API key is used.
Important: connecting the OpenAI API does not attach SSN to an existing ChatGPT web or voice conversation. OpenAI's API manages its own conversation state.
This is the closest option when you already use ChatGPT in Chrome and do not want an API key. SSN can capture new text turns from the page as source type openai, just like it captures a chat source.
Capture a ChatGPT Conversation
Install and enable the Social Stream Ninja Chrome extension.
In the popup, enable the opt-in ChatGPT source.
Open https://chatgpt.com/ and reload the page after enabling the source.
Open the SSN Dock with the same session.
Send a new ChatGPT message. New User and ChatGPT turns should appear in the Dock; existing history is intentionally ignored.
Send Live Chat Into the Open ChatGPT Tab
Test with a private or unlisted stream and a regular ChatGPT text conversation.
Under Management, enable Relay all messages.
Set Pick relay destinations to openai so captured viewer messages go only to the ChatGPT tab.
Keep the ChatGPT tab open. Its new replies return to SSN as openai messages.
Use ChatGPT audio or SSN TTS/overlay output to put those replies on stream.
Experimental: web-page controls can change, relay-all can flood a conversation, and ChatGPT Voice mode may not accept injected text like the regular text page. Use filters or a test stream before sending every viewer message.
AI Website Capture Status
Website
Current Support
ChatGPT
Experimental capture of new text turns; existing history is ignored. Generic browser relay can target its prompt composer.
Claude
No conversation capture. The packaged Claude script is a page-style helper only.
Gemini, Copilot, Perplexity, and other AI sites
No dedicated website capture source currently ships with SSN. Their APIs may still be usable where listed as an SSN provider.
Option 2: Configure an AI Provider Inside SSN
Use a provider when SSN should generate replies, moderate messages, translate chat, search a knowledge base, or build overlays.
Provider Type
Options
Notes
Local
Ollama; Local Gemma; Local Qwen
No cloud API key. Performance depends on the computer and model assets.
Custom/local server
Custom OpenAI-compatible API
For llama.cpp, LM Studio, vLLM, and compatible gateways. Ollama should use its native option.
Cloud
OpenAI, Gemini, DeepSeek, xAI, AWS Bedrock, OpenRouter, Groq, OpenCode Zen
Keys, access, billing, quotas, and model names are controlled by each provider.
Trial
SSN Hosted Trial LLM
Experimental and may be disabled or rate-limited.
Open Chat Bots and AI services in the SSN popup.
Open Configure LLM Service Provider and choose the provider you actually use.
Enter its endpoint, model, API key, or local asset settings.
Select Test selected chat bot. A Connected result proves only that the provider answered the test.
Enable the separate feature—Primary bot, Private bot, cohost, moderation, translation, RAG, or builder—that should use it.
Option 3: Voice AI Cohost
The cohost is designed for the exact “talk to AI while it also sees live chat” workflow. Your microphone and selected live-chat messages feed the same cohost session, and the response can be spoken on stream.
Choose the provider, microphone, response type, instructions, and live-chat mode from the cohost page.
Open cohost.html?session=YOUR_SESSION.
Select a voice/realtime provider or the LLM configured in the SSN popup.
Explicitly select your microphone and Audio Response, then press Start Co-host.
With OpenAI Realtime, leave Live Chat on Context only - answer when asked when you want the AI to remember recent viewer messages without interrupting. Choose Questions and mentions or Every message for spoken automatic replies; these modes do not post into platform chat.
For OpenAI Realtime, capture the audio from the co-host browser/app in OBS and use cohost-overlay.html?session=YOUR_SESSION for the stage avatar and speech bubble. The overlay does not receive the same WebRTC voice stream. Use the co-host output-device selector for a virtual cable when supported; &forcetts deliberately adds a separate overlay voice.
Optional stream tools are off by default. Enable Spotify, an allowlist of OBS scene names, or featured-chat control in the SSN popup, then arm the same tools under Co-host Stream Controls on the co-host page.
For lower OpenAI latency, choose the Mini model, Low reasoning, and Fast turn-taking. Full quality and higher reasoning can improve difficult answers at greater delay and API cost; Fast turn-taking can interrupt natural pauses. Diagnostics shows measured first-output latency.
OpenAI camera and Screen Share input is optional and defaults to No Video. When selected, SSN sends one current image with each direct spoken or typed request; it is not a continuous video stream and can add API cost and latency.
Keep the speech bubble visible, or provide captions, so viewers do not need to hear the AI voice to follow the conversation.
OpenAI Realtime uses the OpenAI project key configured under ChatGPT API in the SSN popup. The standard key stays in the SSN extension/Desktop background, which creates a short-lived Realtime client secret for the co-host page. Microphone and AI voice audio then travel over WebRTC. Replies are capped at 512 output tokens and Diagnostics shows cumulative usage and latency. SSN checks quiet sessions and reconnects failed transports; a recovered session reapplies its instructions and performance settings but cannot restore the previous spoken conversation. This creates an SSN cohost session; it does not take over an existing conversation in the ChatGPT website or ChatGPT desktop app.
chatbot.html is a private helper and prompt-testing page. Enable Chat Bot - Private Interface, open the page with your session, and use the configured LLM provider.
The private bot is separate from the Primary live-chat bot and currently does not share its RAG dataset or other chat history.
Option 6: AI Processing Tools
Tool
What It Does
Start Safely
Censor bot
Classifies, sanitizes, or blocks incoming messages.
Start in non-blocking mode and review results.
Translation
Translates incoming chat and optional outgoing host messages.
Test one language and avoid block mode until reliable.
RAG knowledge base
Lets the Primary bot retrieve relevant text from uploaded resource files.
Use current, focused text or Markdown files.
Chat summaries
Allows users to trigger an AI summary with !summary.
Check history size, cost, and private content.
Option 7: AI Overlay Builder
Use aiprompt.html to describe, generate, preview, and save an overlay. Use aioverlay.html as the OBS output page after it is saved.
Simulate normal chat and alert events before adding a generated overlay to OBS.
Connected, but no live reply: enable the Primary bot or the specific AI feature; provider setup alone does not process chat.
Cohost sees chat but stays quiet: Context only intentionally waits for the streamer to ask about chat. Switch to Questions and mentions or Every message for automatic replies.
ChatGPT browser source is silent: enable the ChatGPT opt-in, reload chatgpt.com, and test with a new turn instead of existing history.
Bot skips most messages: check trigger words, moderator-only mode, rate limits, and the model's instruction to reply selectively.
Text appears but no sound: confirm the TTS-producing page is open and OBS is capturing that page's audio.
