Syntopia BOX Guide
You can connect an AI model to your live host in Syntopia BOX by pasting a single API key. That one step is what turns a good-looking AI avatar into a host that actually thinks — one that writes its own scripts, rewrites every line so nothing repeats, and answers viewers’ comments in real time.


The key comes from an LLM provider you choose: DeepSeek or OpenAI. Syntopia BOX doesn’t ship its own closed brain; you plug in the model you want and keep full control of the account behind it.
This guide covers what the model powers, exactly where you paste the key, how to choose between DeepSeek and OpenAI, and roughly what it costs. It goes deeper on this one feature than the complete Syntopia BOX guide, which covers it in a single step.
- Your AI host has three parts: a look (the AI digital human), a voice (the voice model), and a brain (the connected AI model). This post is about the brain.
- You connect it under AI model by pasting an API key from DeepSeek or OpenAI — no coding required.
- The model does three jobs: writes scripts, rewrites every line so nothing repeats, and answers comments intelligently from your knowledge base.
- The provider bills you directly, pay-as-you-go. A live host’s text is short and bursty, so usage is light — but always confirm the current rate on the provider’s own pricing page.
What “connecting an AI model” actually means
In Syntopia BOX, your AI host is built from three separate layers. The digital human is how it looks, trained from a short video of a presenter. The voice model is how it sounds — cloned from a few seconds of clear audio, with V10 supporting 600+ languages (you train a voice per language you sell in). The AI model is the brain: the large language model that decides what the host says when there is no fixed script to read.
Without a connected model, your host can still read scripts you typed in by hand. Connect one, and the same host can generate those scripts for you, keep them fresh across a long stream, and reply to the chat on its own. It is the difference between a host that recites and a host that converses.
- Look — the AI digital human (an avatar model, not a real person).
- Voice — the cloned or ready-made voice model (V10 supports 600+ languages; train one per language).
- Brain — the connected text AI model (an LLM such as DeepSeek or OpenAI) that writes and answers.
How to connect an AI model (step by step)
This is the whole setup — it takes a couple of minutes and you only do it once.
- Create an account with your chosen provider: DeepSeek or OpenAI.
- Generate an API key in that provider’s dashboard. (A key is a long string of characters that lets BOX call the model on your behalf.)
- In Syntopia BOX, open the host setup and go to AI model.
- Paste your API key into the field and save.
That’s it. From that point on, the AI Script, AI Rewrite, Variation and Smart Live Chat features all run on the model you connected. If you ever want to switch providers, you simply paste a different key.
- The key is yours — created in your provider account and billed to you, not routed through a Syntopia subscription.
- Treat it like a password: keep it private, and revoke and regenerate it from the provider’s dashboard if it is ever exposed.
What the connected model powers
Once a key is in, the model does three distinct jobs across your live room.
It writes your scripts
In Standard Mode you still control every word, but the model gives you AI Script, AI Rewrite, Continue and Vary to draft and polish lines fast. In AI Stream Mode you skip scripting entirely: import the product’s name, price and selling points, and the model writes the whole script for you — which is how you go live quickly across many products.
It rewrites every line so nothing repeats
A host that loops the same five sentences for three hours sounds like a recording — and platforms notice. The model’s Variation feature rewrites a script into many versions so no two runs are identical. Because large models aren’t perfectly predictable, the method is to split a script into segments of 300–500 characters and vary each one, rather than varying a 5,000-character block in one go (where parts can get lost).
During the stream, BOX picks one sentence at random from each set of {curly brackets} and one word from each set of [square brackets], so every pass through the script comes out worded differently. You can also steer the rewrite with plain-language instructions — for example “professional but easy to follow, friendly, clear, credible, no exaggeration” and “avoid internet slang.”
It answers comments intelligently
This is the part viewers feel. Turn on Smart Live Chat in the AI Chat tab and the model answers chat questions in real time, drawing on a custom knowledge base of up to 20,000 characters — your prices, sizes, shipping, returns and best answers to common questions. The better the knowledge base, the better the answers.
You choose who delivers the reply with Live Chat Method: the Host says it out loud, a Chat message posts it as text, or the Assistant voice answers. Tick one, two or all three. Proactive Live Chat uses the same brain to welcome new viewers, ask for follows and likes, and nudge people to buy, even when no one has asked a question yet.
The voice and the look make your host believable. The connected model is what makes it useful — it writes, it varies, and it talks back.
Which provider should you connect — DeepSeek or OpenAI?
Both are standard LLM API providers, and BOX treats them the same way: get a key, paste it, done. The difference is in cost, capability and language strength, so pick based on what your stream needs.
| Provider | Good fit for | How you pay |
|---|---|---|
| DeepSeek | Keeping running costs low while still getting solid scripts and chat replies — a common pick for high-volume, always-on live rooms. | Your own DeepSeek account, pay-as-you-go by usage. |
| OpenAI | The broadest general capability and widest language coverage, when you want the most polished replies and cost is a smaller concern. | Your own OpenAI account, pay-as-you-go by usage. |
Roughly what does it cost?
The AI model is billed by the provider you choose — not by Syntopia. You pay that provider directly for what the model uses, and it is typically a pay-as-you-go amount based on how much text goes in and out. A single host’s scripts and chat replies are short bursts of text rather than heavy workloads, so for most live rooms the model is a small line item next to the rest of your setup.
Exact rates change often and differ between the two providers, so check the current pricing on your chosen provider’s own page before you commit, and set a usage limit or budget alert in that provider’s dashboard if you want a hard ceiling.
Getting the most out of a connected model
- Feed the knowledge base. The chat is only as smart as what you give it — load real prices, shipping and returns, and clean answers to the questions you get most. You have up to 20,000 characters.
- Segment before you vary. Break scripts into 300–500-character chunks and vary each one, so long scripts rewrite cleanly without losing parts.
- Use brackets deliberately. Put the lines you want shuffled inside {curly brackets} and swappable words inside [square brackets] so every run sounds fresh.
- Keep a human in reach. The model answers on its own, but a real person can step in at any moment with Human Takeover if a question needs a human touch.
Once the brain is connected, see the complete Syntopia BOX guide for building the full live room around it, or compare plans on the Syntopia pricing page.
Frequently asked questions
How do I connect an AI model to my AI live host?
Create an account with DeepSeek or OpenAI, generate an API key in that provider’s dashboard, then in Syntopia BOX go to the AI model setting and paste the key in. No coding is required. From then on the host can write scripts, rewrite lines so nothing repeats, and answer viewers’ comments in real time.
Which AI model is best — DeepSeek or OpenAI?
Both work the same way in BOX. DeepSeek is a common choice for keeping running costs low on long, high-volume streams; OpenAI offers the broadest general capability and widest language coverage. Pick based on your budget and the languages you stream in — and you can switch any time by pasting a different key.
What does the connected model actually do?
Three things: it writes your scripts (AI Script and AI Rewrite in Standard Mode, or full auto-scripts in AI Stream Mode from product info you import), it rewrites every line through Variation so no two runs of the stream are identical, and it powers Smart Live Chat to answer viewers’ questions from your knowledge base.
How much does the AI model cost to run?
The model is billed by your chosen provider, not by Syntopia — you pay DeepSeek or OpenAI directly on a pay-as-you-go basis for the text the host generates. A live host’s scripts and replies are short, so usage is usually light, but rates change often, so confirm the current price on the provider’s own pricing page and set a budget limit in their dashboard if you want a ceiling.
Do I need an AI model connected to go live at all?
No. You can still run a stream by typing scripts in yourself in Standard Mode. You only need a connected model for the features that generate text on their own — AI-written scripts, Variation to stop repetition, and Smart Live Chat replies. Connecting a model is what lets the host handle those without you.
Is my API key safe, and who is billed for usage?
The key is created in your own provider account and billed to you, so usage charges go to you, not through a Syntopia plan. Treat the key like a password: keep it private, and if it is ever exposed, revoke and regenerate it from the provider’s dashboard.