Agents

Learn what agents are, how they work, and how they access information within Chocolate Factory.

What is an Agent?

Agents are AI-powered assistants that you can customize for specific roles and workflows. Each agent has its own identity, model configuration, system prompt, and optional tools, allowing you to build specialized assistants for tasks like customer support, content writing, data analysis, and more.

Agents belong to an organization and can be scoped to a specific project or shared across the organization.

How Agents Work

When a user sends a question to an agent, the agent processes the request using a foundation model and can invoke connected tools to complete tasks before returning a response.

1. A user sends a question to the agent.
2. The agent invokes a foundation model (such as Amazon Bedrock) to understand the request and plan next steps.
3. If needed, the agent calls tools, such as built-in capabilities, MCP servers, databases, or workflows, to gather information or perform actions.
4. The agent returns a final response to the user.

How Agents Access Information

Agents combine multiple knowledge sources and automatically select the most appropriate information for each request to generate accurate and relevant responses.

The available knowledge sources include:

• Tool Knowledge → Documents, messages, databases, and other data sources that the agent has permission to access.
• Web Knowledge → Real-time information retrieved from the internet when current or external information is required.
• LLM Knowledge → The foundation model’s built-in knowledge acquired during training.

By combining data sources, real-time web information, and the model’s general knowledge, agents can provide more complete, context-aware responses while respecting organizational permissions and access controls.