LangGraph integration

AgentWeave provides a first-class async LangGraph routing node backed only by the public AgentWeaveRuntime API.

LangGraph state
    ↓
AgentWeaveLangGraphNode
    ↓
AgentWeaveRuntime.preview_route()
    ↓
selected/permitted candidates + confidence + provenance
    ↓
normal LangGraph downstream nodes

Install

pip install 'agentweave-router[langgraph]'

Use as a graph node

from langgraph.graph import START, END, StateGraph
from agentweave.integrations.langgraph import langgraph_node

runtime = ...  # AgentWeaveRuntime

graph = StateGraph(MyState)
graph.add_node('route', langgraph_node(runtime, query_key='query'))
graph.add_edge(START, 'route')
graph.add_edge('route', 'work')
graph.add_edge('work', END)
app = graph.compile()

The node returns:

• agentweave_selected_tools;
• agentweave_permitted_tools;
• agentweave_routing_confidence;
• agentweave_routing_abstained;
• agentweave_routing_provenance.

A context_factory can turn graph state into a RunContext for tenant, role, permission, scope, environment, and approval-aware routing.

Public API boundary

The integration does not reach into RequirementAnalyzer, AgentMatcher, GlobalTeamOptimizer, registry internals, or observability internals. Those components may evolve without forcing LangGraph applications to change.

See examples/langgraph_agentweave.py for a local routing-only example.

Evidence boundary

This is an ecosystem integration feature, not a benchmark result. It does not imply LangGraph/LangChain endorsement or modify frozen AgentWeave research evidence.