We’re embedding Convai in a browser-based tool where multiple users will chat with the same character at the same time. We want to use the Evaluation API to give each user feedback. Support@Convai have suggested to post in forums.
Key questions:
Multi-User Handling
Does the “session_id” keep evaluations linked to the right user when lots of sessions run at once?
Does the API response include the “session_id” so we can track it?
Is the “session_id” the same across the Evaluation API and the Chat History API?
Custom Scoring
Can we set custom scoring (like 1–5 scale) or weighting for different skills (e.g. empathy, open questions)?
Can we prompt the Evaluation API to use a specific scale for scoring participants answers?
Rate Limits
-Are there any limits or best practices for using the Evaluation API with lots of users?
Example Output
-If custom scoring is supported, could you share an example of the request, prompt and response?
For multi-user handling, you include a session_id in the request body, and each user can have their own session. For your third question about consistency across the Evaluation API and Chat History API, I wasn’t quite sure what you meant, could you clarify?
As for custom scoring and prompting, yes, that’s supported. You can define specific criteria, scales (e.g., 1–5), and prompts within your request. Please refer to the documentation link above for request format examples.
We’re building a browser-based educational platform using Convai. Conversations will be held via webplugin or directly on your site. Our structure is:
Page 1: User speaks to a Convai character.
Page 2: User gets feedback from character using the Evaluation API.
We want to confirm:
Will the session_id consistently link the chat and its Evaluation API feedback to the correct user, even when many users are interacting simultaneously? (and because this will be on different webpages) Our main concern is that because it is not a single interface, how can we be sure the Evaluation API corresponds to the user and their earlier conversation?
This depends on how you structure and manage your application.
For example, if you pass -1 as the session_id when a user opens the experience and starts interacting with the character, a new session_id will be automatically generated. You can then store this generated session_id and use it.
Okay, I’m not super technical but I imagine this is more challenging if hopping between webpages to access the conversation and then the Evaluation API Feedback coach?
Page 1: conversation with character
Page 2: evaluation of previous conversation
All of this will be done via convai embedding.
I imagine doing this via Unity or unreal means that the session ID is clear. I’m looking for confirmation that session ID can be passed from one to the other smoothly for our application.
If you can answer that it would be greatly appreciated.
Convai Pixel Streaming Embed is currently accessible only with the Enterprise plan. At the moment, retrieving and passing session IDs across pages is not supported. However, for enterprise users, we can offer solutions tailored to your project needs.
If this functionality is essential to your application, we recommend reaching out to our sales team at sales@convai.com to discuss upgrading to an Enterprise plan.
The Evaluation API does not track sessions automatically. As outlined in our documentation, it requires you to provide session ID manually in the API request. (session_id, character_id, and the prompt).
This means session tracking is entirely managed on your end, and you can assign or reuse session IDs as needed to align with your application’s structure, whether in Unity or a web-based setup.