LlamaIndex agents: FunctionAgent & ReActAgent

A plain query engine answers a single question from your data. An agent can decide when to query, query multiple sources, call other tools, and loop until it has a complete answer. LlamaIndex makes that jump small: wrap your index as a tool, hand it to an agent, done. This is how RAG becomes agentic.

Your index becomes a tool

The bridge is the QueryEngineTool — it turns any query engine into a callable tool with a name and description the agent reads to decide when to use it.

Two agent types

LlamaIndex gives you two agents, differing only in how they decide on actions:

• FunctionAgent — uses the model’s native function/tool calling. Cleaner, more reliable, and the default for any model that supports tool calling (most do now). Reach for this first.
• ReActAgent — uses text-based ReAct (Thought / Action / Observation written as text). Useful for models without native tool calling, or when you want the reasoning trace visible as text.

Both are Workflows under the hood — the agent loop is a Workflow with steps for “call the LLM,” “run the tool,” and “check if done.” So everything you learned about Workflows applies: you can add custom steps, stream events, and trace the run.

from llama_index.core.agent.workflow import FunctionAgent
from llama_index.core.tools import QueryEngineTool

# Wrap two indexes as tools the agent can choose between
docs_tool = QueryEngineTool.from_defaults(
    docs_index.as_query_engine(),
    name="product_docs",
    description="Answers questions about product features and pricing.",
)
policy_tool = QueryEngineTool.from_defaults(
    policy_index.as_query_engine(),
    name="policies",
    description="Answers questions about refunds, SLAs, and terms.",
)

agent = FunctionAgent(tools=[docs_tool, policy_tool], llm=llm)
# response = await agent.run("Is the Pro plan refundable after 20 days?")
# → the agent picks 'policies', queries it, and answers — citing the source nodes.

The agent reads each tool’s description to decide which to call — so write descriptions like you’re writing for a smart colleague: clear, specific, and honest about what each tool does and doesn’t cover.

Quick check

Next

Agents are only as good as the documents they index. LlamaParse solves the messy-PDF problem that wrecks naive ingestion.