AI models

Overview

This document aims to guide you through configuring your AI models in Powell Buddy to be able to use those models into your agents. We will take the example of an Azure OpenAI model but it works for any kind of AI provider supported by Azure.

Prerequisite

To proceed with this guide, you will need to configure your Azure OpenAI Service in your Entra ID Portal by following the instructions in this article : FOCUS: Create your Azure OpenAI service 

Access

Once logged into Powell Buddy administration, the left-hand menu allows you to navigate between the different functionalities of the solution.

Managing the models is accessible from the left menu via Administration > AI models.

Click on the “AI models” entry to view and manage your general-purpose models used by agents.

Create a new AI model

1. Click on the create button.

2. Fill in the form fields:

• Title: A name to identify the model. Example: "AI for Intranet".
• Description: You can detail the model if you wish. This is optional.
• Type: The type of AI provider.
• Consumption limit per month: To control the cost of usage of your model in your agents.
• Consumption limit per user per month: To control the cost of usage per user so that all of your employees can use.

3. Selecting the type allows you to enter the connection information for the agent. Here the example of an Azure OpenAI configuration:
- Resource name: The name of the resource when creating the service in Azure.
- API key: The connection key to the service that you received during its creation.

Note: the list of supported model is available below

4. Validate the form with the save button.

Your AI model is created.

To edit/delete an AI model

From the Ai model page, you can simply use the "..." button to the right of each AI model to choose to edit or delete it.

The edition page allows to change any information already configured during the creation process.

 

List of supported agent models

Powell Buddy supports a range of Azure-hosted LLMs to ensure enterprise-grade security and integration. Some models unlock specific features such as Plugins & Knowledge Base support, image generation, or OCR.

 

🔌 Plugins & Knowledge Base Support

• gpt-4o

• gpt-4o-mini

💬 Additional Prompt Only (No plugin or knowledge base supported)

OpenAI

• gpt-4o

• gpt-4o-mini

• gpt-35-turbo-16k

• gpt-35-turbo

• gpt-4

• Not tested: o3, o3-mini, o3-pto, o4-mini, o1, o1-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4-32k

DeepSeek

• DeepSeek-R1

• Not tested: DeepSeek-V3, DeepSeek-V3-0324

Mistral

• Mistral-small

• Mistral-large-2407

• Mistral-large-2411

• Not tested: mistral-small-2503

Microsoft

• Phi-3-small-8k-instruct

• Phi-3-small-128k-instruct

• Not tested: Phi-3-mini-4k-instruct, Phi-3-mini-128k-instruct, Phi-3-medium-4k-instruct

Meta

• Llama-2-7b

• Llama-2-13b

• Llama-2-70b

• Meta-Llama-3-8B

• Meta-Llama-3-70B

• Not tested: Llama-2-7b-chat, Llama-2-13b-chat, Llama-2-70b-chat

🖼️ Image Generation

• dall-e-3, GPT-image-1

• Not tested: GPT-image-1-mini

🔎 OCR (Optical Character Recognition)

• gpt-4o

Specialized Models

This tab lets you configure AI models dedicated to specific advanced tasks such as file attachments processing or image recognition (OCR). These models are called when these features are activated in agents or knowledge bases.

You can access this via the “Specialized models” tab in the AI Models page.


 

Use Cases

These models are required for:

• File attachments: Allowing agents to analyze and understand document content. (Activated in agent configuration)
• OCR: Indexing and interpreting text embedded in images. (Activated in knowledge base configuration)
• Speech to Text: Converting spoken language into written text for voice interactions. (Activated in agent configuration)
• Text to Speech: Converting written text into natural-sounding spoken language for audio responses. (Activated in agent configuration)

To ensure performance and isolation, it is highly recommended not to reuse models already assigned to standard agents.

Attached Files (Embedding)

Purpose: Used to process and embed file attachments so agents can retrieve and reason over their content.

• Recommended model: text-embedding-3-small

• Provider: Azure OpenAI

To configure:

1. Go to the “Specialized models” tab.

2. Locate “Attached files (embedding)”.

3. Click the edit icon ✏️.

4. Fill in the connection info:

   • Resource URL

   • API key

   • Deployment model (recommended model, Open AI text-embedding-3-small)

This model will be used whenever file upload is enabled in an agent.

Image Recognition (OCR)

Purpose: Used to index and understand images inside knowledge bases.

• Recommended model: gpt-4o

• Provider: Azure OpenAI

This model is automatically triggered when OCR is enabled for a knowledge base containing image files.

To configure:

1. Go to the “Specialized models” tab.

2. Locate “Image recognition (OCR)”.

3. Click the edit icon ✏️.

4. Fill in the connection info:

   • Resource URL

   • API key

   • Deployment model (e.g., gpt-4o-test)
     
      

Speech to Text

Purpose: Used to convert spoken language into written text, enabling voice interactions with agents or knowledge bases.

• Recommended model: whisper
• Provider: OpenAI (or compatible)

To configure:

1. Go to the “Specialized models” tab.
2. Locate “Speech to Text”.
3. Click the edit icon ✏️.
4. Fill in the connection info:
   • Resource URL
   • API key
   • Deployment model (e.g., whisper-1)

This model will be used whenever voice input or transcription features are enabled in an agent.

Text to Speech

Purpose: Used to convert written text into natural-sounding spoken language, enabling audio responses or voice synthesis.

• Recommended model: tts
• Provider: OpenAI (or compatible)

To configure:

1. Go to the “Specialized models” tab.
2. Locate “Text to Speech”.
3. Click the edit icon ✏️.
4. Fill in the connection info:
   • Resource URL
   • API key
   • Deployment model (e.g., tts-1 or tts-1-hd)

This model will be used whenever text-to-speech features are enabled in an agent or knowledge base.