Changing prompt for summaries with own GPU transcription Server

Hexanet

Platinum Partner
Joined
Aug 6, 2020
Messages
23
Reaction score
8
Hello everyone,

Is ther anywhere a parameter ti change prompts for summaries when you do not use the 3CX Cloud Transcription ?
Do I have to use the OpenAI or Google Summary parameter ?

Thanks for your help.
 
  • Like
Reactions: telcocentricu1
Yes it is in the custom parameter table.
Admin > Advanced > Parameters
GOOGLE_SUMMARY and OPENAI_SUMMARY
 
Thanks for this answer.
Just to be sure I have to modify both with the same prompt ?
Can you explain me what's the difference between Google or OpenAI Summary?
 
You are welcome
No you don't need to have the same prompt. They can be completely different. They are 2 different labs so they use different models.
Anyway, you cannot use both at the same time. So you need to choose one.
In short, there is not much difference in this particular topic. Both simply get audio and transcribe it. They both have the same limitations (no stereo support and no inbuilt diarization). We had to make diarization in the prompt to make it work.

That's where we come in with 3CX Cloud Transcription. We take stereo and have diarization. And most importantly, cheaper.
 
I undestand, but which one is used, or do I have to personnalize, if I use an on premise GPU Translation Server ?
 
If in transcription you select Google, the google prompt is used
If in transcription you select OpenAI, the openai prompt param is used
If you select 3CX Cloud, we use a hardcoded system prompt which you cannot change.
If you select 3CX Onboard, wait until next week and you will have the ability to customize it.
 
Do you have any info for this update ? Can I modifiy the prompt in the latest Beta Version? And where ?
I Use this config :

1777562288467.png
 
Thanks for your quick reply.Our Client want to modify the resume prompt to have a little more words and some specific point of interest like :

"Summarize the following call transcript in the context of a Customer Service department specializing in pharmacy training. If the transcript contains fewer than 50 words, reply only with “N/A”. The summary must be clear, concise (maximum 300 words), and accurately reflect the key points mentioned by both the client and the advisor. When the terms “xxx” or “yyy” appear, summarize them using “x” and “y” respectively. The summary must be structured into the following 6 sections: 1. **Call Context** - Reason for the call - Initial client situation (e.g., subscription status, usage context, needs…) 2. **Client Needs / Issues** - Questions raised - Obstacles encountered - Expectations expressed 3. **Information, Support, and Solutions Provided by the Advisor** - Explanations given - Actions performed during the call (platform guidance, checks, configurations…) 4. **Identified Objectives** - Client objectives (e.g., completing a action, developing for the team, solving a connection issue…) - Internal objectives if mentioned (quality, value creation, retention) 5. **Action Plan / Next Steps After the Call** - Advisor's actions (follow-up, email sent, monitoring, technical escalation…) - Actions expected from the client (documents to provide, steps to complete, activation, consultation…) 6. **Key Elements for Client Follow-up** - Relevant contextual information - Points of attention - Opportunities for value creation or retention (if applicable) Do not add any introductory or concluding text. Reply only with the summary structured in these 6 sections."
 
Hello,
I've searched in the last RC but don't have any new summary for private transcript. I've found Grok but that's all.
Is there any other place to search ?

1778829326323.png
 
Agreed
Currently it is hard coded - but we internally agreed to implement this.
Coming soon as part of a new improved product launch!! Stay tuned.
 
Is it hardcoded on all configurations of AI? We have premise server using OpenAI and I have been trying to adjust the prompt with no luck.
 
Yes currently it is hardcoded. But we are going to release a new version and in the new version, you will be able to change it.
Please bear with us a little while longer.
 
Hi, have you some good news for us for making own prompts on own AI servers ? Is there a chance to see also our own AI servers be used as AI Agents or receptionnist ?

Thanks for your feedback.
 
  • Like
Reactions: nikolascx
Hi, have you some good news for us for making own prompts on own AI servers ? Is there a chance to see also our own AI servers be used as AI Agents or receptionnist ?

Thanks for your feedback.

Working on it - we need to wait for update 10 to be ready. the new 3CX AI Server needs update 10.
Bear with us a little while longer. But it is 100% guaranteed. You can change the prompts from the custom parameters table.
 
  • Like
Reactions: Evolute IT
BTW can you elaborate more on this

Is there a chance to see also our own AI servers be used as AI Agents or receptinist?

What AI Agents do you have? Give us some more details to understand what you want to do and also how you want to achieve this - (Architecture wise)
 
Today only OpenAI is available to create Agents (Receptionnist or simple Agent).
Is there a way to have another configuration, using a private server, like for the translations ?
 
Is there a way to have another configuration, using a private server, like for the translations? ->

Still very very vague.
Lets try like this:

Call comes in to 3CX.
Then what you want to do?
What entity should answer the call and what you want to do with the call during the call and after it has ended.
 
Sorry for the confusion, let me explain my use case more clearly.

Today, when creating a 3CX AI Receptionist or AI Agent, the only available option is to connect it to an OpenAI project. As far as I understand, there is currently no alternative AI backend that can be used.

What I would like to know is whether 3CX plans to support private/self-hosted AI models in the future, similarly to how custom/local AI servers can already be used for features such as transcription, translation, or sentiment analysis.

My goal is to keep all voice and conversation data inside the customer's infrastructure and avoid sending any information to external cloud services.

The architecture I have in mind would be:
  1. A call arrives in 3CX.
  2. 3CX sends the audio/transcript to a customer's private AI server (on-premises or private cloud).
  3. The AI server acts as the Receptionist/Agent, understands the caller's request, and determines the appropriate action (routing, transferring, collecting information, etc.).
  4. 3CX executes the actions returned by the AI server.
We already have an internal AI server handling translation and sentiment analysis successfully. My question is whether such a server could eventually be used as the AI backend for AI Receptionists and AI Agents instead of OpenAI.

Is this currently possible, or is support for custom/self-hosted AI providers planned for future releases?
 

Latest Posts

Members Online Now

Forum statistics

Threads
111,832
Messages
589,278
Members
164,662
Latest member
DejanMDS