LLM Configuration¶
Each agent is powered by a language model. On this page you choose which model each agent uses and how it should behave, expressed as a temperature. Defaults are already in place, so you only need to touch these settings if you want to fine-tune the interview.
There are two modes, switched with the toggle at the bottom of the page: Basic and Advanced.
Model¶
The model is the underlying AI that does the thinking. Larger, more capable models tend to ask better follow-up questions and make more reliable judgments, while smaller models are faster and cheaper. The list of available models depends on what your installation has access to.
Temperature¶
Temperature controls how creative or predictable a model's output is, on a scale from 0 to 1.
- Low temperature (closer to
0) makes the model more focused and deterministic. The same input will tend to produce the same output. This is ideal for agents that make factual judgments, which is why the Classification agent uses a low temperature by default. - Higher temperature (closer to
1) makes the model more varied and creative. This helps the interviewer phrase questions in a natural, non-repetitive way, which is useful for the Probing, Reformulation, and Guide agents.
If you are unsure, leave the temperature at its default.
Basic mode¶
Basic mode keeps things simple: you pick one model that every interview agent will use. This is the recommended starting point, choose a model you trust and let all the agents share it.
If the agents are currently set to different models (for example because you configured them individually in Advanced mode), the picker shows --- until you select a single model to apply to all of them.
Advanced mode¶
Advanced mode lets you configure the model and temperature for each agent separately. Use this when you want, for example, a powerful model for probing but a faster one for the lighter judgment tasks.
You can configure each of the following agents individually:
- Probing Agent
- Classification Agent
- Reformulation Agent
- Guide Agent
A convenience picker at the top, Set all agent models at once, lets you apply one model to every agent and then adjust individual agents from there.
See the Agents overview for what each agent does during the interview.