An "agent" is an automated reasoning and decision engine. It takes in a user input/query and can make internal decisions for executing that query in order to return the correct result. The key agent components can include, but are not limited to:
- Breaking down a complex question into smaller ones
- Choosing an external Tool to use + coming up with parameters for calling the Tool
- Planning out a set of tasks
- Storing previously completed tasks in a memory module
LlamaIndex provides a comprehensive framework for building agents. This includes the following components:
- Using agents with tools at a high-level to build agentic RAG and workflow automation use cases
- Low-level components for building and debugging agents
- Core agent ingredients that can be used as standalone modules: query planning, tool use, and more.
The scope of possible use cases for agents is vast and ever-expanding. That said, here are some practical use cases that can deliver immediate value.
-
Agentic RAG: Build a context-augmented research assistant over your data that not only answers simple questions, but complex research tasks. Here are two resources (resource 1, resource 2) to help you get started.
-
SQL Agent: A subset of the above is a "text-to-SQL assistant" that can interact with a structured database. Check out this guide to see how to build an agent from scratch.
-
Workflow Assistant: Build an agent that can operate over common workflow tools like email, calendar. Check out our GSuite agent tutorial.
-
Coding Assistant: Build an agent that can operate over code. Check out our code interpreter tutorial.
Using Agents with Tools
The following component guides are the central hubs for getting started in building with agents:
Building Custom Agents
If you're interested in building custom agents, check out the following resources.
Building with Agentic Ingredients
LlamaIndex has robust abstractions for every agent sub-ingredient.
- Query Planning: Routing, Sub-Questions, Query Transformations.
- Function Calling and Tool Use: Check out our OpenAI, Mistral guides as examples.
- Memory: Example guide for adding memory to RAG.
We offer a collection of 40+ agent tools for use with your agent in LlamaHub 🦙.