> For the complete documentation index, see [llms.txt](https://imbrace.gitbook.io/imbrace-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://imbrace.gitbook.io/imbrace-documentation/ai-agent.md).

# AI Agent

An **AI Agent** in IMBRACE is a virtual assistant you set up to handle a specific job. It can answer questions, follow your instructions, and work with your organisation's information. Each agent is built for a clear purpose, such as answering customer enquiries, helping staff find information, or supporting a particular workflow.

Once an agent is set up, users can interact with it in plain language. The agent understands the request, draws on the knowledge and instructions it has been given, and responds or takes action.

#### What an AI Agent is made of

Every AI agent has a few core parts that together define how it works.

1. **Identity.** This includes the agent's name and a short description of what it does. A clear name and description help users pick the right agent and understand what to expect from it.
2. **Instructions.** These tell the agent how to behave: what its role is, what tone to use, what it should and shouldn't do, and how to handle different kinds of requests. Well-written instructions are the biggest factor in how useful an agent is.
3. **Knowledge.** This is the information the agent can draw on, such as documents or data sources you connect to it. The agent uses this to give accurate answers based on your organisation's content rather than general knowledge alone.
4. **Agent definition.** All of the above is saved together as the agent's definition. This is what gets packaged into a .zip file when an agent is exported, and what you upload when you import an AI agent.

#### What you can do with AI Agents

Create a new agent from scratch for a specific task or team. Import an existing agent from a .zip file to reuse it or move it between workspaces. Update an agent's instructions or knowledge as your needs change. Test the agent with sample questions and refine it until its answers are consistently accurate.

#### Tips for a good agent

Give each agent one clear purpose rather than trying to make it handle everything. Keep the description short and specific so users know when to use it. Test with real questions your users are likely to ask, and adjust the instructions based on what you see.