🛡️ langchain-suprawall

Official SupraWall security and compliance integration for LangChain Python.

SupraWall provides a deterministic security layer for AI agents, enabling enterprise-grade governance, human-in-the-loop approvals, and forensic audit logs.

Built for teams that need to ship autonomous agents while staying compliant with the EU AI Act (Articles 9, 12, & 14) and SOC2/ISO27001 requirements.

🚦 Quick Start

Install via pip:

pip install langchain-suprawall

With LangChain Agents (create_react_agent)

from langchain_suprawall import SupraWallCallbackHandler
from langchain.agents import create_react_agent, AgentExecutor
from langchain_openai import ChatOpenAI
from langchain import hub

llm = ChatOpenAI(model="gpt-4")
tools = [...] # Your tools
prompt = hub.pull("hwchase17/react")

# Initialize the security callback
suprawall = SupraWallCallbackHandler(
    api_key="your_suprawall_api_key",
    agent_id="payment-agent-01"
)

agent = create_react_agent(llm, tools, prompt)
agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
    callbacks=[suprawall] # Attach here
)

With LangGraph

from langgraph.prebuilt import create_react_agent
from langchain_suprawall import SupraWallCallbackHandler

suprawall = SupraWallCallbackHandler()

# Pass callbacks in the config during invocation
graph = create_react_agent(model, tools)
graph.invoke(
    {"messages": [("user", "delete my account")]},
    config={"callbacks": [suprawall]}
)

🏛️ EU AI Act Compliance

SupraWall is the first integration designed specifically to meet the high-risk AI requirements of the EU AI Act:

• Article 9 (Risk Management): Identifies and mitigates risks through deterministic policy enforcement.
• Article 12 (Logging): Automatically generates tamper-proof, time-stamped execution logs.
• Article 14 (Human Oversight): Built-in hooks for "Human-in-the-Loop" (HITL) triggers on destructive actions.

🔒 Common Security Policies

Define policies in your SupraWall dashboard and they are automatically enforced via the callback:

1. Financial Guardrails (Refund Limits)

Policy: "If tool process_refund is called with an amount > $500, block execution." SupraWall intercepts the tool start, evaluates the arguments, and raises a PolicyViolation exception if the check fails.

2. Destructive Action Approval

Policy: "Require human approval for delete_customer_record." The callback will pause execution and wait for a signed approval from the SupraWall dashboard before letting the agent continue.

3. Prompt Injection Shield

SupraWall's runtime analysis detects adversarial intent (jailbreaks, DAN-style prompts) before they reach your sensitive tools.

📖 Documentation & Support

• Full Documentation: docs.supra-wall.com
• Dashboard: app.supra-wall.com
• Discord: Join our community

© 2026 WiserAutomation. Licensed under the Apache License 2.0.