Discussions

Hello everyone,

I am planning to use Heygen to run an Interactive Avatar for a structured video interview. The interview consists of a predefined set of questions (in the range of 10–20), which should be asked in a fixed order.

After speaking with the Heygen support team, I was told:

• For this workflow, I would need to use the Streaming SDK or API (with TaskType.repeat method) and implement my own application logic, where the questions are stored in my own database and then passed to the Interactive Avatar.
• The interviewee’s (human) responses currently cannot be stored directly in Heygen – I would also need to use my own database to record and store the conversation flow and answers.
• Follow-up questions could be generated using the built-in Knowledge Base, but if I wanted to use a specific language model such as GPT-4.1, I would have to connect my own LLM process.
• At present, Heygen does not offer a built-in feature to automatically store and process the conversation flow, nor the option to freely select an LLM in the standard workflow.

They told me to post here the following questions:

1. Is it correct that I cannot simply place the full set of interview questions in the System Prompt (similar to ElevenLabs conversational voice agents), and instead must use the SDK or API to pass the questions over to the Interactive Avatar? If yes, what is the reason?
2. Is it correct that I currently always need my own database to store the conversation questions and answers (transcript)?
3. Which language models can I currently use natively in Heygen – and is GPT-4.1 indeed only available through a custom LLM integration?
4. Are there plans to support conversation recording and storage, direct in-system processing, and free LLM model selection in the future?

If this topic requires a more in-depth technical review, is there a paid support or priority assistance option available for implementation guidance?

Thanks in advance for any clarification.