langchain-ai/langgraph
1
0
Fork 0
mirror of https://github.com/langchain-ai/langgraph.git synced 2026-07-15 12:35:43 -04:00
Code Issues 314 Packages Projects Releases 17 Wiki Activity

[PR #2477] [rfc] langgraph: add agent node wrapper and handoff tools #2732

New Issue
Closed
opened 2026-02-20 17:47:28 -05:00 by yindo · 0 comments
yindo commented 2026-02-20 17:47:28 -05:00
Owner
Copy Link
Copy Source

Original Pull Request: https://github.com/langchain-ai/langgraph/pull/2477

State: closed
Merged: No

This change introduces few building blocks for multi-agent:

  • make_agent_node - a wrapper around create_react_agent that returns a graph node function
    • allows users to define custom input/output processing functions (handle what information is passed to/from agent as well as handle different state schemas). by default all of the message history is shared by all of the agents via messages (each agent receives full message history and and adds its internal history to the full history). TBD if this is the best default
    • handles handoff tools that return GraphCommand.goto and returns a GraphCommand.goto to dynamically route to a different node (agent)
    • handles state updates from tools when a tool in subgraph agent wants to update parent state
  • AgentRouterState (TBD on name, would love to improve) -- react agent state + active node (node) for routing on follow-up interactions
  • add_entrypoint_router - helper that adds the agent nodes to the StateGraph and adds a top-level entrypoint router based on the active node in the state. TBD if this can be replaced with human-in-the-loop
  • ToolNode is now aware of tools returning GraphCommand and can apply state updates from GraphCommand.update. this allows general support for tools that can update state (only internal to the graph)
  • handoff - a shorthand that helps create tools that return GraphCommand with goto and update

Basic usage:

from langgraph.prebuilt.handoff import handoff
from langgraph.graph import StateGraph, GraphCommand, START, END

# define handoff tools
bob_tool = handoff(goto="bob", tool_message="Transferred to Bob", name="transfer_to_bob", description="Transfer to Bob")
alice_tool = handoff(goto="alice", tool_message="Transferred to Alice", name="transfer_to_alice", description="Transfer to Alice")

# define agent nodes
alice = make_agent_node(model, [add, bob_tool], state_schema=AgentRouterState, state_modifier="You're Alice.", output_processor=keep_last_message_as_human)
bob = make_agent_node(model, [multiply, alice_tool], state_schema=AgentRouterState, state_modifier="You're Bob.", output_processor=keep_last_message_as_human)

# define graph
builder = StateGraph(AgentRouterState)
add_entrypoint_router(builder, route_to=[("alice", alice), ("bob", bob)], default_start_node="alice")
memory = MemorySaver()
graph = builder.compile(checkpointer=memory)

# pass control between the agents
config = {"configurable": {"thread_id": "1"}}
for chunk in graph.stream({"messages": [("user", "hi! transfer me to bob")]}, config):
    ...
for chunk in graph.stream({"messages": [("user", "back to alice please")]}, config):
    ...

Here are some usage examples:

**Original Pull Request:** https://github.com/langchain-ai/langgraph/pull/2477 **State:** closed **Merged:** No --- This change introduces few building blocks for multi-agent: * `make_agent_node` - a wrapper around `create_react_agent` that returns a graph node function * allows users to define custom input/output processing functions (handle what information is passed to/from agent as well as handle different state schemas). by default all of the message history is shared by all of the agents via `messages` (each agent receives full message history and and adds its internal history to the full history). TBD if this is the best default * handles handoff tools that return `GraphCommand.goto` and returns a `GraphCommand.goto` to dynamically route to a different node (agent) * handles state updates from tools when a tool in subgraph agent wants to update parent state * `AgentRouterState` (TBD on name, would love to improve) -- react agent state + active node (`node`) for routing on follow-up interactions * `add_entrypoint_router` - helper that adds the agent nodes to the `StateGraph` and adds a top-level entrypoint router based on the active node in the state. TBD if this can be replaced with human-in-the-loop * `ToolNode` is now aware of tools returning `GraphCommand` and can apply state updates from `GraphCommand.update`. this allows general support for tools that can update state (only internal to the graph) * `handoff` - a shorthand that helps create tools that return `GraphCommand` with `goto` and `update` Basic usage: ```python from langgraph.prebuilt.handoff import handoff from langgraph.graph import StateGraph, GraphCommand, START, END # define handoff tools bob_tool = handoff(goto="bob", tool_message="Transferred to Bob", name="transfer_to_bob", description="Transfer to Bob") alice_tool = handoff(goto="alice", tool_message="Transferred to Alice", name="transfer_to_alice", description="Transfer to Alice") # define agent nodes alice = make_agent_node(model, [add, bob_tool], state_schema=AgentRouterState, state_modifier="You're Alice.", output_processor=keep_last_message_as_human) bob = make_agent_node(model, [multiply, alice_tool], state_schema=AgentRouterState, state_modifier="You're Bob.", output_processor=keep_last_message_as_human) # define graph builder = StateGraph(AgentRouterState) add_entrypoint_router(builder, route_to=[("alice", alice), ("bob", bob)], default_start_node="alice") memory = MemorySaver() graph = builder.compile(checkpointer=memory) # pass control between the agents config = {"configurable": {"thread_id": "1"}} for chunk in graph.stream({"messages": [("user", "hi! transfer me to bob")]}, config): ... for chunk in graph.stream({"messages": [("user", "back to alice please")]}, config): ... ``` Here are some usage examples: * basic usage examples: https://github.com/langchain-ai/langgraph/blob/vb/prebuilt-agent-components/examples/multi_agent/experimental/simple_multi_agent_examples.ipynb * hierarchical agent tutorial: https://github.com/langchain-ai/langgraph/blob/vb/prebuilt-agent-components/examples/multi_agent/experimental/hierarchical_agent_teams.ipynb * customer support example: https://github.com/langchain-ai/langgraph/blob/vb/prebuilt-agent-components/examples/multi_agent/experimental/customer_support.ipynb
yindo added the pull-request label 2026-02-20 17:47:28 -05:00
yindo closed this issue 2026-02-20 17:47:28 -05:00
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
agents
awaiting author revision
bug
change
chat models
checkpointer
ci
code quality
code quality
dependencies
documentation
drawing
duplicate
enhancement
good first issue
help wanted
invalid
maintainer
meta
missing information
not maintained by langchain
open-swe-auto
open-swe-max
open-swe-max-auto
pending
performance
pull-request
Mirrored from GitHub Pull Request
 question
requires design
sdk
server
stale
studio
tests
type checking
usage question
v0.5
v1.1
No Label pull-request
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langchain-ai/langgraph#2732
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?