Configuring own model for Pega AI agents/Coach/GenAIConnect for cloud/On Prem

AI
1

Hello Everyone,

Could you please let us know whether there is any configuration feasibility to allow clients to select and use their own LLM models within AI Agents, Coach, or GenAI Connect, instead of relying solely on the models provided by Pega?

I am looking for both instances where pega is deployed on-prem and Pega Cloud. Are there any existing extension points for this ? Is this current limitation ? Anything coming in upcoming versions?

-Karunkumar

2

Hi, not at them moment (Pega 25). I will let product folks to comment about plans for future releases.

3

Thanks for the update.
Supporting client-owned LLM models would be a valuable capability for our use case, especially for organizations with specific compliance or governance requirements. We would appreciate any future updates from the product team regarding this enhancement.

4

Hi @karunkumarn, excellent question!
Like Kamil mentioned, at the moment this is not offered yet, and our product team is exploring options to offer this in future Pega versions.

Out of curiosity, does your organization already have their own LLM model they want to connect?

5

Hi @Tim_Straatsma ,

It’s a fine-tuned model for a specific task. They wanted to use only that model while using the Pega AI agent in one of the use cases that we are exploring.

6

Thank yo for this important voice. Pega recognize such voice and will be looking how to address such needs

7

Hi @karunkumarn ,

There will be some highly relevant announcements coming out at Pegaworld in a few weeks. If you won’t be there in person, check out the broadcasts of the keynotes.

I did find another discussion thread with a proposed approach you could investigate which leverages custom GenAI Connect steps: How to use our own LLM/AI Model instead of Pega OOTB AI models? | Support Center?

8

Thanks for sharing the announcements, they sound really interesting. I’m familiar with the GenAI Connect customization approach, but I feel it still differs from what agents can truly offer.

What excites me more about agents is the ability to have an interactive and continuous session, rather than starting fresh with every new prompt or topic. That kind of contextual engagement feels much more natural and powerful.

I’m definitely looking forward to seeing how Pega evolves and expands its capabilities in this space.

9

The approach in the post doesn’t seem to work with the agents as we do not have an option itself to create a new GenAI provider. What we could do is simply configure the REST connectors with our desired model URLs and call them. Looking forward for the related announcements in the PegaWorld.

10

Yea not at the moment but this is the most anticipated feature for the next Pega world.

11

This feature is already there where we can use a REST connector and the can be used within Pega Gen AI connect, agents etc. There are exentions points are available to override the Gen AI provider. By default, Pega is the Gen AI provider which can be overridden and use a REST integration with extension points which is currently available and appropriately map the requests and responses.