# LLM Agents and Integration

# LLM Agents and Integration

Curiosity Workspace enables you to build sophisticated AI agents that interact with your data and perform tasks using Large Language Models (LLMs).

# What is an LLM Agent?

An LLM agent is an AI-driven workflow that can:

• Reason: Understand user intent and decompose tasks.
• Act: Call workspace endpoints or external APIs to retrieve information or perform actions.
• Observe: Process the results of its actions and adjust its plan.

# Building an Agent

To build an agent in Curiosity Workspace:

# 1. Define the Tools

Create AI Tools that the agent can use. These tools might include:

• search_docs: Search for relevant documents in the workspace.
• get_user_info: Retrieve details about a specific person from the graph.
• update_status: Change the status of a project or task.

# 2. Configure the LLM

Select an LLM provider (e.g., OpenAI, Azure OpenAI, Anthropic) and configure the model settings in AI Integrations → LLM Configuration.

# 3. Design the Prompt Pattern

Use Prompting Patterns like ReAct (Reason + Act) to guide the agent's behavior.

# 4. Grounding and Guardrails

• Grounding: Always provide the agent with context retrieved from the graph or search to reduce hallucinations.
• Guardrails: Implement checks to ensure the agent's outputs are safe and comply with organizational policies.

# Example: A Customer Support Agent

• Trigger: A new support ticket is created.
• Action: The agent searches for similar past tickets and relevant documentation.
• Reasoning: The agent synthesizes a draft response based on the retrieved information.
• Action: The agent posts the draft response for human review.

# Next Steps

• Learn about Graph Reasoning
• Explore Multimodal Search