Framework Guide

Stress-test LangGraph agents before they reach production.

Wrap an existing LangGraph app with Crucible, run seeded scenarios, and export trace-backed reports without rewriting your graph logic.

How this integration works

Keep your existing graph. Crucible adapts state at the boundary.

Run public suites while developing and hidden-eval style rotations before release.

Export signed traces and deployment reports for every candidate graph.

Starter example

from crucible import evaluate, wrap_langgraph_agent

wrapped = wrap_langgraph_agent(my_langgraph_app)
result = evaluate(
    wrapped,
    name="SupportGraph-v4",
    scenario="tool-heavy-support",
    seed=42,
)

print(result.trace_path)
print(result.durability_scores["D8"])