Skip to main content
Open-Source Runtime AI Governance

Agents that act. Governed before they do.

comply54 is an open-source runtime AI governance platform that ensures AI agents operate within organizational policies, security guardrails, and regulatory requirements — before taking action.

Not a linter. Not a post-hoc audit. Enforcement at the moment the agent decides to act.

GitHub starsPyPI versionnpm versionApache 2.0

The enforcement layer between your AI agents and every consequence of acting without boundaries — org policy, security, and regulatory compliance in one runtime decision.

21 policy packs10+ jurisdictions100% offline enforcementOpen source · Apache 2.0

What comply54 enforces at runtime

Organizational Policies

Tool restrictions, rate limits, output constraints, and data handling rules defined by your team — enforced before any agent action executes.

Tool allowlistsOutput redactionRate limitsEscalation gates

Security Guardrails

Prompt injection detection, exfiltration prevention, covert channel blocking, and OWASP LLM Top 10 controls — active at every agent decision point.

Prompt injectionExfiltrationOWASP LLM Top 10Identity abuse

Regulatory Requirements

NDPA 2023, CBN NIP, NIMC Act 2026, KDPA, POPIA, and 15+ African regulations — with the exact law, section, and penalty when a rule fires.

NDPA 2023CBN NIPKDPAPOPIA + 15 more

AI frameworks give agents capability.
None of them enforce governance.

Every major AI framework ships with tools, memory, and reasoning. None of them ship with runtime enforcement — the layer that ensures every agent action has been checked against your organizational policies, security guardrails, and regulatory requirements before it runs.

That gap is yours to close. When an agent acts outside the boundary — exposing sensitive data, breaching a transaction cap, or following an injected instruction — it is your organization that answers for it, not the framework's authors.

comply54 closes that gap. One integration. Every agent action evaluated against org policy, security, and regulatory requirements before it runs.

LangChainNo runtime governance enforcement
LangGraphNo runtime governance enforcement
CrewAINo runtime governance enforcement
AutoGenNo runtime governance enforcement
comply54Org policies + Security guardrails + Regulatory requirements

What happens when an agent gets it wrong

These are not hypothetical. Each scenario below maps to a real regulatory obligation that an AI agent can violate without any human intent.

NDPA 2023NDPC

Your agent returns a customer's BVN in a chat response. That's a Schedule 1 sensitive data breach.

ConsequenceUp to ₦10M fine + 2% of annual gross revenue. Criminal liability for officers.
NIMC Act 2026NIMC

Your onboarding agent stores NIN data after the verification completes. The NIMC Act 2026 explicitly prohibits this.

Consequence₦20 million corporate fine. Minimum 5 years imprisonment for the responsible individual.
CBN NIP FrameworkCBN

Your payment agent approves a ₦12M transfer for a Tier 2 customer. That exceeds the ₦10M single-transaction NIP cap.

ConsequencePer-transaction fines. Potential CBN operating licence suspension.
MLPPA 2022 / NFIUNFIU

Your agent processes a ₦5M+ transaction without flagging it for Currency Transaction Reporting within 24 hours.

ConsequenceCriminal prosecution under MLPPA 2022 §10–11. AML licence revocation.

How comply54 works

It sits between your agent and its tools. Every action is checked. Blocked calls never execute. Your team gets a paper trail without writing a single compliance rule.

01

Agent decides to act

Your AI agent calls a tool — transfer funds, look up a NIN, respond to a customer, export data.

02

comply54 intercepts

Before the tool executes, comply54 evaluates the action against every applicable regulation for your jurisdiction and sector.

03

You get a structured decision

Allow, block, or escalate — with the exact law, section, penalty, and rule that triggered it. Not a vague warning. The actual citation.

04

A signed receipt proves it happened

Every decision produces a cryptographically signed JWT receipt — tamper-evident, offline-verifiable proof that the check ran, what input was evaluated, and what the decision was. Store it in your audit log. Show it to a regulator.

One import. The law enforced.

When a violation fires, comply54 returns the exact regulation, the specific section, the rule that triggered, and a unique audit ID — not a vague error message. Your engineering team ships in hours. Your compliance team can read the output.

Blocked before it runs
from comply54 import NigeriaFintechCompliance

compliance = NigeriaFintechCompliance()

result = compliance.check(
action="transfer_funds",
params={"amount": 15_000_000, "currency": "NGN"},
context={"kyc_tier": 2},
)

# Blocked before execution — no need to handle it downstream
if result.blocked:
v = result.primary_violation
print(v.messages[0])
# → CBN NIP §4.2: Single-transaction cap ₦10M exceeded (₦15M attempted)
print(v.rule_triggered) # nip_cap
print(v.citations[0].document)
# → CBN NIP (NIBSS Instant Payment) Framework §4.2 — Per-Transaction Cap

Works with the frameworks you already use

LangGraph, CrewAI, AutoGen, and LangChain adapters ship out of the box. Drop comply54 into an existing agent in under an hour — no rewrite, no new infrastructure, no OPA binary to install.

LangGraphCrewAIAutoGenLangChain
LangGraph — add one node
from comply54.langgraph import Comply54Guard, comply54_route
from comply54 import NigeriaFintechCompliance

# One node. Every tool call your agent makes is
# evaluated against NDPA + CBN + BVN/NIN + NFIU
# before it executes. Blocked calls never reach the tool.
graph.add_node("compliance", Comply54Guard(
compliance=NigeriaFintechCompliance()
))
graph.add_conditional_edges(
"compliance", comply54_route,
{"tools": "tools", "agent": "agent"}
)
NEW in v0.4.0Signed Compliance Receipts

Every decision, cryptographically signed.

Blocking a violation is half the story. The other half is proving it happened — to your board, your auditors, and the CBN or NDPC if they come asking.

comply54's signed receipt system attaches a tamper-evident Ed25519 JWT to every compliance decision. Offline-verifiable, tied to the exact input evaluated, and signed with keys that never leave your infrastructure.

🔏

Ed25519 signed JWT

Every ComplianceResult carries a receipt_token — a compact JWT signed with your Ed25519 private key. Tamper-evident by design.

🔌

100% offline verification

Auditors and regulators verify receipts with only your public key. No network, no comply54 instance, no third-party service.

🔒

Bring your own key

comply54 never generates or stores your signing key. Your keys live in your secret manager (AWS KMS, HashiCorp Vault, GCP KMS).

🧾

Input digest — exact proof

A SHA-256 digest of the exact action and parameters is embedded in every receipt. Recompute it to prove the receipt covers the specific transaction under inspection.

Signed + verifiable — one param
from comply54 import NigeriaFintechCompliance
from comply54.receipts import verify_receipt

compliance = NigeriaFintechCompliance(signing_key=private_key_pem)

result = compliance.check(
action="transfer_funds",
params={"amount": 8_500_000, "currency": "NGN"},
context={"kyc_tier": 2, "sanctions_screened": True},
)

# Every result carries a signed, tamper-evident receipt
print(result.receipt_token)
# eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJjb21wbHk1N...

# Verify offline — no network, no comply54 instance needed
payload = verify_receipt(result.receipt_token, public_key_pem)
print(payload.decision) # "allow"
print(payload.input_digest) # sha256:3f4a8c... — proves the exact input
print(payload.rule_triggered) # None (allowed — no rule fired)
print(payload.comply54_version) # "0.4.0"
Interactive Demo

See what happens when an agent breaks African law

Agent Disaster Lab simulates real AI agent failure scenarios — each mapped to the exact African regulation violated. The same comply54 packs that run in production, run here.

NDPA 2023 / NIMC Act 2026

Agent leaks customer BVN in a chat response

₦10M fine + 2% of annual gross revenue
CBN NIP Framework §4.2

Payment agent approves ₦12M transfer for Tier 2 customer

Per-transaction fines + CBN licence suspension
OWASP LLM01:2025 / NDPA 2023 §25

Prompt injection redirects funds to attacker account

Unauthorized transfer + Schedule 1 data breach

9 failure scenarios · real compliance decisions · live in your browser

Open Agent Disaster Lab ↗

Built for every regulated sector in Africa

Pre-assembled compliance bundles for the sectors with the highest regulatory exposure.

🏦

Nigerian Fintech

Payment agents, lending platforms, digital banks, and wallet providers operating under CBN licence. Every agent action touching funds or identity is a regulatory event.

NDPA 2023CBN NIPBVN/NINNFIU-AMLNIMC Act 2026
🏥

Healthcare AI

Clinical decision support, EHR agents, and telemedicine platforms. Patient consent, clinical data confidentiality, and cross-border health data rules are all enforced.

NHA 2014FMOH Guidelines (Draft 2024)NDPA 2023BVN/NIN
🛡️

Insurance Platforms

AI-assisted claims adjudication, underwriting engines, and fraud detection. NAICOM auto-denial caps, anti-discrimination rules, and AML thresholds enforced automatically.

NIIRA 2025NAICOM GuidelinesNDPA 2023NFIU-AML
🌍

Pan-African AI Platforms

Nigerian, Kenyan, and South African packs are production-ready. Ghana, Rwanda, Egypt, Tanzania, Uganda, Ethiopia, and Mauritius packs are included in preview — coverage depth varies by jurisdiction.

KDPA ✓POPIA ✓Ghana DPARwanda DPAEgypt PDPL+5 preview
21
Policy packs
10+
African jurisdictions
4
Framework adapters
100%
Offline — no network
0
OPA binary needed
Working Paper · July 2026

The Enforcement Gap: Why Africa's AI Regulations Need Runtime Enforcement

Africa does not have an AI regulation gap. It has an accountability gap. A 26-page paper on why AI agents need runtime enforcement infrastructure — and what it looks like.

Now Open · Cohort 1

comply54 Open Source Fellowship

A structured open-source programme for engineers and researchers who want to contribute to the first African AI compliance library — and get recognised for it. Pick a track, pick an issue, ship real code.

🔌
AI Integrations
Adapters for AutoGen, Semantic Kernel, Haystack
📜
AI Governance Research
New African jurisdiction packs
🛠️
Developer Tooling
OTEL exporter, async engine, CLI
Developer Experience
TypeScript sector classes, ComplianceResult UX
View open issues and apply →
Open fellowship issues10
Contribution tracks4
Effort levels (days, not months)S–L
Impact — African AI is under-governed
Open to engineers at all levels. No prior comply54 experience needed — good first issues are clearly marked.

Your agents should know the law
before they act.

comply54 is open source, installs in minutes, and works with every major AI agent framework. Nigerian and African compliance is a requirement, not a feature.

Python (PyPI)
$pip install comply54
TypeScript / Node (npm)
$npm install @comply54/core