Tamper-evident, compliance-grade logs of every agent action in LangGraph agents — without rewriting your code, without adding a proxy to the request path.
The problem
LangGraph is great at what it does, but production-grade audit logging is bring-your-own. Most teams discover this when the first production incident hits, the first OpenAI bill arrives, or the first auditor asks for evidence.
Specifically:
- Multi-node graphs are nearly impossible to debug from stdout
- Interrupt/human-in-the-loop primitives don't include policy or approval routing
- No native cost attribution per node or per graph run
How Prefactor solves it
Prefactor wraps your LangGraph compiled graph and adds audit logging as a runtime layer. Specifically:
- SOC 2 evidence collection
- EU AI Act Article 12 record-keeping
- ISO 42001 documentation lifecycle
- auditor-ready exports
- incident forensics
Install and integrate
pip install prefactor-langchainfrom langgraph.graph import StateGraph
from prefactor import Prefactor
pf = Prefactor(api_key="pf_live_...")
graph = StateGraph(...)
# ... define nodes and edges
compiled = pf.wrap_graph(graph.compile(), graph_id="research-agent-v1")Specific use cases
- Soc 2 evidence collection
- Eu ai act article 12 record-keeping
- Iso 42001 documentation lifecycle
- Auditor-ready exports
- Incident forensics