Integration · Framework
Kyvvu for LangGraph
The LangChain handler works on graphs as they are. The LangGraph handler adds each step's node name and records interrupt() as an approval gate.
Set up
A few lines, then every step is checked.
- Node names
- Every step carries the graph node it ran in, so you can see where in the graph a verdict landed.
- Human in the loop
- An interrupt() is recorded as a gate, so gate policies see your approval steps.
- Flat traces
- Tool calls inside nodes appear as ordinary steps; there are no node-transition pseudo-steps.
from kyvvu_langchain import KyvvuLangGraphHandlerhandler = KyvvuLangGraphHandler(kv)result = app.invoke(input_data, config={"callbacks": [handler]})Two handlers
Same engine, more graph detail.
The LangChain handler works on graphs as they are. The LangGraph handler is opt-in, for when you want to know where in the graph a step ran.
| KyvvuLangChainHandler | KyvvuLangGraphHandler | ||
|---|---|---|---|
| Tool and model calls checked before they run | Yes | Yes | |
| The graph node each step ran in | No | Yes, in properties.langgraph.node_name | |
| interrupt() | Not recorded as a gate | Recorded as a step.gate, so gate policies see your approvals | |
| Trace | Flat: no node-transition steps | Flat: no node-transition steps |
Example
A whole graph, offline.
From the docs: a one-node graph with one declared tool, under the bundled baseline. No account and no network.
- Pass it in config
- The handler goes in config={"callbacks": [handler]} on invoke().
- Declare your tools
- The bundled baseline blocks tool calls that aren't on the declared list.
- No account needed
- Offline, the client needs an agent_id and a policy source, here the bundled baseline.
from typing import TypedDictfrom langchain_core.runnables import RunnableConfigfrom langchain_core.tools import toolfrom langgraph.graph import END, START, StateGraphfrom kyvvu import Kyvvufrom kyvvu_baseline import load_baseline_policiesfrom kyvvu_langchain import KyvvuLangChainHandlerclass State(TypedDict): message: str@tooldef echo(message: str) -> str: """Return the input for this offline example.""" return f"Received {message}"def respond(state: State, config: RunnableConfig) -> State: return {"message": echo.invoke(state["message"], config=config)}graph = StateGraph(State)graph.add_node("respond", respond)graph.add_edge(START, "respond")graph.add_edge("respond", END)app = graph.compile()kv = Kyvvu( agent_id="echo-graph", policies=load_baseline_policies(), declared_tools=["echo"], read_only_tools=["echo"], log_location="none",)input_data = {"message": "Hello"}handler = KyvvuLangChainHandler(kv)result = app.invoke(input_data, config={"callbacks": [handler]})print(result["message"])Good to know
- Install langgraph itself separately. Tested with langgraph 1.2.
- Parallel tool calls in one step are recorded side by side. Path policies order steps across steps, not within concurrent calls.
- A resumed run is recorded as a new task.
Next
Try it before you install it.
In your browser
The playground
Build an agent's path, pick a policy and watch the engine decide.
On your machine
Five minutes, no account
kyvvu try runs a demo agent against the bundled baseline.