Prebuilt agents and tools

AgentFlow includes prebuilt building blocks for common agent patterns. Use them when the default shape matches your workflow, and drop down to StateGraph, Agent, and ToolNode when you need custom control.

Prebuilt agents

Prebuilt agents live in agentflow.prebuilt.agent.

Agent	Use case
ReactAgent	Standard model and tool loop.
RAGAgent	Retrieval-augmented generation with retriever and synthesis steps.
PlanActReflectAgent	Plan, execute, and critique across multiple passes.
StructuredOutputAgent	Constrain the final answer to a schema.
SupervisorTeamAgent	A supervisor routes work to named workers (WorkerConfig).
SwarmAgent	Peer agents hand control to each other (SwarmMemberConfig).
AudioAgent	Realtime audio-to-audio sessions over a live provider socket.

Reranker base classes ship alongside the RAG agent:

Class	Use case
BaseReranker	Interface for custom reranking of retrieved documents.
CohereReranker	Rerank with the Cohere Rerank API.
CrossEncoderReranker	Rerank with a local cross-encoder model.

Everything above except AudioAgent is also re-exported from agentflow.prebuilt. AudioAgent must be imported from agentflow.prebuilt.agent:

from agentflow.prebuilt.agent import AudioAgent
from agentflow.prebuilt import ReactAgent, SupervisorTeamAgent, SwarmAgent

Prebuilt tools

Common tools are exported from agentflow.prebuilt.tools.

Tool	Use case
safe_calculator	Evaluate simple math expressions safely.
fetch_url	Fetch web content from a URL.
file_read, file_write, file_search	Work with files where enabled.
google_web_search, vertex_ai_search	Search integrations.
memory_tool, make_user_memory_tool, make_agent_memory_tool	Store or retrieve long-term memory.
create_handoff_tool, is_handoff_tool	Transfer control between graph agents.

Handoff tools

Handoff tools let the model transfer execution to another graph node by calling a tool with the transfer_to_<agent> convention.

from agentflow.prebuilt.tools import create_handoff_tool
from agentflow.core import ToolNode

tools = ToolNode([
    create_handoff_tool(
        agent_name="RESEARCHER",
        description="Transfer to the research agent.",
    ),
])

Use handoff tools for agent-to-agent delegation. Use Command(goto=...) when routing should be explicit code rather than an LLM tool choice.

Rules

Rule	Why it matters
Prefer prebuilt agents for common patterns	They encode the standard AgentFlow graph shape.
Keep handoff targets aligned with node names	The graph jumps to the target node.
Use tools for capabilities, not routing policy	Routing policy often belongs in graph edges or Command.
Check reference docs for constructor details	Prebuilt surfaces evolve as new patterns stabilize.

Related docs

• Agents and tools
• Callbacks and Command
• Command and handoff reference