LangChain Integration Playbook

Implement zero-latency governance in LangChain pipelines across industries.

Installation

pip install aisentinel langchain

Basic Agent with Preflight

from langchain.agents import initialize_agent, Tool
from langchain.chat_models import ChatOpenAI
from aisentinel.integrations.langchain import SentinelGuard

llm = ChatOpenAI(model_name="gpt-4o")
sentinel_guard = SentinelGuard.from_env()

tools = [
    Tool(
        name="web_search",
        func=sentinel_guard.wrap_tool(web_search_tool),
        description="Search the web under AISentinel governance",
    ),
]

agent = initialize_agent(tools, llm, agent="conversational-react-description")
response = agent.run("Find executive contact info")

Customer Service Workflow

from aisentinel.rulepacks import load_rulepack

rulepack = load_rulepack("customer_service.yml")
guard = SentinelGuard(rulepack=rulepack)

• Blocks data exfiltration and escalates to human review when severity >= high
• Streams decisions to the audit console in <40 ms

Research Automation

research_agent = initialize_agent(
    tools=[sentinel_guard.wrap_tool(pdf_reader_tool)],
    llm=llm,
)

• Enforce citation policies with rulepack research_compliance.yml

Performance Benchmarking

from aisentinel.testing import benchmark

results = benchmark(agent, cases="./benchmarks/langchain.jsonl")
print(results.latency_p95)

Expect <60 ms preflight overhead with local cache enabled.

Multi-Tenant Configuration

sentinel_guard = SentinelGuard.from_pool([
    {"tenant_id": "healthcare", "rulepack": "rulepacks/hipaa.yml"},
    {"tenant_id": "finance", "rulepack": "rulepacks/finra.yml"},
])

agent = initialize_agent(tools=[sentinel_guard.wrap_tool(web_search_tool, tenant="finance")], llm=llm)

Switch tenants per conversation to isolate policy enforcement.