Creating and Mapping a Custom Model to an Agent

This use case walks you through how to create a custom model in Cortx, configure its inputs and outputs, and then map it to a custom agent. This workflow is especially useful when you want to test and verify new models for specific industries or specialized tasks.

Step-by-step guide

1. Navigate to Models

1. Go to the Agents section from the left navigation.
2. In the side blade, select Models.
3. You will see three tabs:
   • My Models – Displays all custom models you have created.
   • Prebuilt Models – Ready-to-use models provided by Cortx.
   • Default Models – System-level default models.

---

2. Create a Custom Model

1. Click Create Model in the top-right corner.
2. A form will appear where you need to fill in details:
   • Provider (dropdown) – Select from providers like Azure OpenAI, Groq, OpenAI, VLLM, nvidia, Bedrock, Deepseek, Anthropic, etc.
   • Model – Choose from the provider’s available models (e.g., GPT-4.1, Nano-4.1, etc.).
   • Display Name – The name you want to assign to the model.
   • Access Key – Your authentication key for the model provider.
   • Description – Briefly describe the model’s purpose.
3. Click Next.
4. Select Input/Output types by checking relevant boxes (Text, Image, Audio, Video).
5. Configure Advanced Capabilities if required (Reasoning, Tool Calling, Parallel Tool Calls).
6. Save the model:
   • Save → Model is stored in Draft state.
   • Test Model → Opens a new chat window for direct testing.

---

3. Publish the Model

1. Once tested, go back to My Models.
2. Open the model’s 3-dot menu.
3. Click Publish to make it available for agent mapping.

---

4. Create a Custom Agent

1. From the left navigation, go to Agents.
2. In the side blade, select Agents.
3. Choose My Agents (for your custom ones) or Prebuilt Agents (Cortx-provided).
4. Click Create Agent in the top-right corner.
5. Fill in the form:
   • Base Agent – Select a prebuilt agent as the foundation.
   • Name – Title for the custom agent.
   • Industry – Choose the domain or vertical.
   • Description – Provide details about the agent’s function.
6. Click Save. The agent will now appear in My Agents.

---

5. Configure the Agent

1. Edit the agent persona:
   • Description & Instructions – Add details about the agent’s goals and rules.
2. Add Knowledge Sources – Select the types of knowledge the agent can access.
3. Add Tools – Map the required tools for execution.
4. Go to the Models section within the agent editor.

---

6. Map Custom Model to Agent

1. In the Models tab, select your newly created model from My Models.
2. Click Map Model to attach it to the agent.
3. Save the configuration.

---

7. Test the Agent

1. Open the agent in a new chat.
2. Interact with it to validate responses.
3. Use the 3-dot menu beside the agent tray to switch between models if needed.
4. Experiment with different inputs and outputs to confirm the agent behaves as expected.

---

Best Practices

• Always test a custom model in draft state before publishing.
• Keep model descriptions and access keys updated for easy tracking.
• Use base agents aligned to your industry for better results.
• Verify outputs against real use cases before deploying agents to production.

---

✅ With these steps, you can create, verify, and deploy a custom model mapped to a custom agent in Cortx, giving you full control over specialized workflows.