Model
In the Model Tab of OVA (Odea Virtual Agent), users can configure the core AI model settings for their Virtual Agent. Here is a breakdown of the key components visible in the image:
- Provider & Model Selection:
Choose the AI provider for the Virtual Agent. In this example, OpenAI has been selected, and you can also choose the specific model from the available options (e.g., GPT-3.5 Turbo).
- Temperature:
- This slider adjusts the model’s creativity or randomness. A higher temperature (closer to 1) results in more diverse and creative responses, while a lower temperature (closer to 0) makes the AI more deterministic and focused.
- Max Tokens:
- Defines the response length. Setting a limit here ensures that the generated responses stay concise. In this configuration, the max token value is set to 250, limiting the length of replies.
- First Message:
- This is the initial greeting message the Virtual Agent will send when it starts interacting with a user. For example, “Hello, this is Ava. How may I assist you today?”
- System Prompt:
- The system prompt provides context and personality to the AI model. It defines the Virtual Agent’s role, persona, and behavior. Here, the agent “Ava” is described as an expert in customer support with a warm and engaging tone, capable of simulating diverse scenarios for training purposes.
- Knowledge Base:
- This section allows users to attach a specific knowledge base that the Virtual Agent can refer to for more accurate responses. (In the image, it appears no files are currently selected.)
- Detect Emotion:
- This toggle can be enabled if the user wants the Virtual Agent to detect and respond to the emotional tone of the conversation, adding a layer of empathetic interaction.
- Cost and Latency Indicators:
- At the top, you can see estimated costs per minute of usage and the expected response latency. In this example, the estimated cost is ~$0.09 per minute, and the latency is around 700 ms.
The Model Tab is essential for configuring the foundational behavior of your OVA, providing control over the Virtual Agent performance, personality, and response style.
