Building Your Own Agent

cortx allows users to build their own custom AI agents to address specific business needs, reduce hallucinations, and eliminate generic outputs. Instead of relying on all-purpose agents, users can create domain-specific agents that act as subject matter experts.

🔹 Example Use Cases:

• A coding agent trained on programming knowledge.
• A design agent trained on design systems and principles.

These agents provide more relevant, context-aware responses tailored to the user’s domain.

cortx allows you to create custom AI agents in two ways:

• Manual (UI-Driven): Full control via structured setup
• Prompt-Based (Conversational): Guided agent creation through a Q&A chat flow

Both methods result in deployable, reusable agents across your workspace.

Manual Agent Creation (UI-Driven):

Use this method when you want full control over your agent’s components — ideal for workflows that require specific logic, compliance, or tool integration.

Step 1: Navigate to Agent Catalog

• Navigate to Left Nav > Agents > Agent Catalog
• Click on Create Agent button, beside the search bar
• This launches the Agent Builder interface

Step 2: Define Basic Agent Details

• Fill out the initial details:
  • Agent Name
  • Base Agent
  • Industry
  • Brief Description

Step 3: Connect Knowledge Sources

Give your agent access to the information it needs:

• Select files, folders, data sources, or documents from cortx
• Choose multiple sources as needed

Step 4: Add Tools (Action Layer)

Tools define what actions your agent can perform.

How to add tools:

• In the Agent Builder > Click on Add Tool button

• To select a tool from pre-built library:
  • Select Pre-Built Tools tab
  • Select tool or browse by key word in search bar
  • Click Map Tool button

• To create a new tool:
  • Click on Create Tool
  • Fill the text fields, proceed to next
  • Fill the advanced fields as well and proceed to add.
  • You have created a new tool, which you can find in the My Tools Section

Step 5: Choose a Model (Intelligence Layer)

Choose a reasoning engine for the agent:

• Click on Add Model Button
• To select an existing model, navigate to pre-built models tab.
• Select a model and click on Map Model Button to add and proceed to next.

• Add any applicable guardrails like Business rules, Domain Guidance, Regulatory Guidance, Q&A and Examples.

• Click on Test Agent button on the top, and confirm

• Click on the Chat icon to test the agent in chat. Or click on Publish to directly publish the agent.

• The Agent is now published now and can be seen in the “My Agents” section and available to use in:
  • Conversations
  • Workflow steps
  • Programs or templates

Prompt-Based Agent Creation (Conversational Flow)

Use this method when you want to create an agent quickly by describing it in natural language — no configuration panels needed.

1. Start a New Chat
   • Start a New Chat session
   • From the dropdown, choose Agent Builder Agent
2. State Your Intent
   • Type a plain prompt describing the agent you want.
     • Example prompts:
       1. “I want an agent that sends weekly updates to my stakeholders
       2. “Build me an agent that summarizes all meeting notes and pushes them to SharePoint
       3. “Create a roadmap planner agent for engineering sprints”
3. Answer Builder’s Questions

The Agent Builder will now begin a guided Q&A flow. It may ask:

• “What should we name your agent?”
• “What do you want this agent to do, specifically?”
• “Do you want this to run on demand or on a schedule?”
• “Should it send emails, generate a document, or push data somewhere?”
• “What tools or integrations should it connect to?”
• “Do you want a specific LLM to power the logic?”

Respond naturally — cortx will extract structured data from your answers.

4. Confirm Agent Configuration

Once you’ve answered all required questions, cortx will show a summary:

• Name
• Description
• Connected tools
• Model selected
• Task logic written by the system

You can:

• Approve and save
• Ask to modify something
• Or cancel and restart

5. Finalize and Save

Once confirmed:

• The agent is automatically saved under My Agents
• You can now trigger it in future conversations, workflows, or templates
• All metadata (model, tools, logic) is visible and editable later if needed