when testing with core ai realtime models like gpt-realrime-mini(beta) the NPC doesn’t follow commands and doesn’t preform actions.
Recent character changes
changed to realtime core ai model and changed to a compatible GCP voice disabled the LTM and disabled the knowledgebank that was attached to the NPC. all to try and reduce the latency as much as possible.
User message or prompt
put your hands up.
Current response
in different session the character answers “action hands up” and preform nothing.
Expected response
to preform the actions when commended to.
Steps to reproduce
change to realtime models
talk to the NPC.
I checked the relevant documentation, resources, and similar forum posts before posting.
on
I confirm that I did not include API keys, tokens, passwords, secrets, private character data, private conversation logs, or customer data.
Realtime models do not currently support Actions. This is a known limitation.
For your use case, I recommend using the Gemma models instead. They are very fast and work well for Actions, response speed, and general conversation quality.
Thank you for your response. I have another small question please I thought that enabling long-term memory will help reduce response time is it not true? Based on your new UI? It increases response time.
Do you have a session ID so we can investigate further?
If Knowledge Bank and Narrative Design are also enabled for your character, LTM, KB, and ND run in parallel. The overall response time is determined by whichever process takes the longest, so enabling LTM may increase latency in some cases.
As mentioned on the new UI calculations of milliseconds in the new Core ai panel. it reduces the milliseconds by 500 ms after I disabled long-term memory
As mentioned in the previous message, it is not something I noticed it is only something written in the new dashboard of the core AI settings I will upload a picture. If you do need a session ID you can use the one uploaded in this post.
OK now I understand because in one of your previous tutorials, you mentioned that the long-term memory can help reduce latency because the NPC remembers repetitive conversation and questions.
I’m not sure which blog post, tutorial, or content you are referring to.
Could you please share the link or screenshot of the content where this was mentioned? It is also possible there may have been some confusion around the intended meaning.
