langchain-comply54 0.3.0

LangChain / LangGraph integration for comply54 — African AI agent compliance enforcement

langchain-comply54

LangChain and LangGraph integration for comply54 — runtime compliance enforcement for AI agents operating under African data protection and financial-sector regulations.

Overview

langchain-comply54 wraps the comply54 policy engine as a first-class LangGraph node and LangChain tool. It intercepts agent tool calls before execution, evaluates them against African regulatory frameworks (NDPA 2023, CBN, NFIU, Ghana DPA, Kenya DPA, and 15+ more), and either blocks the call or lets it through — all in-process, with no OPA binary or external service required.

Agent → Comply54Guard → [deny] → ToolMessage error back to agent
                      → [allow] → Tools execute normally

Installation

pip install langchain-comply54

Quick Start — LangGraph Guard Node

The recommended pattern is Comply54Guard as a pre-execution node in a LangGraph ReAct graph:

from langchain_comply54 import Comply54Guard, comply54_route
from comply54.sectors import NigeriaFintechCompliance
from langgraph.graph import StateGraph, MessagesState, END
from langgraph.prebuilt import ToolNode

# 1. Build your tools and agent as normal
tools = [transfer_funds_tool, lookup_account_tool]
tool_node = ToolNode(tools)
agent_node = build_agent(tools)  # your LLM + tools

# 2. Create the guard
guard = Comply54Guard(
    NigeriaFintechCompliance(),
    context={"kyc_tier": 3, "customer_verified": True},
)

# 3. Wire it into the graph between agent and tools
graph = StateGraph(MessagesState)
graph.add_node("agent", agent_node)
graph.add_node("comply54_guard", guard)
graph.add_node("tools", tool_node)

graph.set_entry_point("agent")
graph.add_conditional_edges("agent", should_continue,
    {"comply54_guard": "comply54_guard", END: END})
graph.add_conditional_edges("comply54_guard", comply54_route,
    {"tools": "tools", "agent": "agent"})
graph.add_edge("tools", "agent")

app = graph.compile()

When the agent attempts a tool call that violates compliance policy, Comply54Guard injects a ToolMessage with a structured denial that the agent can explain to the user:

{
  "blocked": true,
  "decision": "deny",
  "reason": "CBN NIP Framework: single transfer of ₦15,000,000 exceeds the ₦10,000,000 per-transaction cap",
  "regulation": "CBN Transaction Limits & Controls",
  "audit_id": "a3f2c1d4-...",
  "jurisdictions": ["NG"]
}

Compliance Decision Outcomes

Every evaluation returns one of four outcomes:

Decision	Meaning	Guard behaviour
allow	All checks passed	Tool executes normally
audit	Action logged for regulatory trail	Tool executes; state gets compliance_result
escalate	Elevated-risk action — human review recommended	Tool executes by default; blocked if block_on_escalate=True
deny	Hard regulatory violation	Tool blocked; agent receives structured ToolMessage error

Available Sector Packs

Import	Jurisdictions	Key Regulations
NigeriaFintechCompliance	NG	NDPA 2023, CBN NIP (₦10M cap), NFIU AML/CFT, BVN/NIN
NigeriaHealthCompliance	NG	NDPA 2023, NHA, NHIA — patient data & health records
NigeriaInsuranceCompliance	NG	NDPA 2023, NIIRA 2025 (replaces Insurance Act 2003)
KenyaFintechCompliance	KE	Kenya Data Protection Act 2019, CBK guidelines
PanAfricanCompliance	NG, KE, GH, RW, EG, MU, ZA, TZ, UG	9-jurisdiction pack

from comply54.sectors import (
    NigeriaFintechCompliance,
    NigeriaHealthCompliance,
    NigeriaInsuranceCompliance,
    KenyaFintechCompliance,
    PanAfricanCompliance,
)

Patterns

Block on escalate

guard = Comply54Guard(
    NigeriaFintechCompliance(),
    block_on_escalate=True,  # treat escalate as deny
)

Per-request context from agent state

Merge dynamic context (e.g. session KYC tier, consent flags) at evaluation time:

# Set compliance_context in your agent state before the guard runs
state["compliance_context"] = {
    "kyc_tier": user.kyc_tier,
    "customer_verified": user.is_verified,
    "consent_documented": user.has_data_transfer_consent,
}

LangChain StructuredTool (self-check pattern)

For agents that should self-check before acting rather than being intercepted automatically:

from langchain_comply54 import comply54_tool
from comply54.sectors import NigeriaFintechCompliance

# The agent can call this tool to ask "is this action allowed?"
compliance_check = comply54_tool(NigeriaFintechCompliance())
tools = [transfer_funds_tool, compliance_check]

Legacy state-dict node

For LangGraph graphs that pass structured state rather than messages:

from langchain_comply54 import compliance_node
from comply54.sectors import NigeriaFintechCompliance

node = compliance_node(NigeriaFintechCompliance())
# state must contain: action, params, output, context
# node writes back: compliance_result, compliance_blocked, compliance_message

Compliance Certificate

Every evaluation produces an auditor-exportable compliance certificate:

result = guard._compliance.check(action="transfer_funds", params={"amount": 5_000_000})
cert = result.to_certificate(
    sector_pack="NigeriaFintechCompliance",
    jurisdictions=["NG"],
    regulations=["NDPA 2023", "CBN FPR"],
)
print(cert.to_json())

Requirements

• Python 3.9+
• comply54[langgraph] >= 0.3.0 (installed automatically)
• No OPA binary, no Docker, no external services

Links

• comply54 documentation
• comply54 on PyPI
• Source repository
• African regulatory policy reference

License

Apache 2.0