Roadmap

LangGraph

Make any existing LangGraph graph replayable and auditable with one import. LangGraph already checkpoints state at every step — point that checkpointer at Kraterion and every run records itself.

Install

pip install kraterion-langgraph

Swap the checkpointer

Replace your existing checkpointer with WalrusCheckpointSaver. Nothing else about your graph changes.

from langgraph.graph import StateGraph
from kraterion.langgraph import WalrusCheckpointSaver

builder = StateGraph(State)
# ... add your nodes and edges as usual ...

graph = builder.compile(
    checkpointer=WalrusCheckpointSaver(project="support"),
)

Run and replay

Run the graph as you always have. Each run prints a receipt you can replay against the same inputs and the same retrieved data.

result = graph.invoke({"question": "What is our refund policy?"})
# run recorded · receipt 3f4d…ae

Add memory

Turn on persistent memory and the graph gets memory.remember and memory.recall as tools. The model decides when to use them, and every call is scoped to the agent and recorded with the run.

graph = builder.compile(
    checkpointer=WalrusCheckpointSaver(project="support", memory=True),
)