🛡️ AgentBridge SDK
Compliance Black Box for AI Agents in Indian Fintech
AgentBridge provides transparent compliance monitoring for AI agents operating in Indian financial services, automatically enforcing:
- RBI FREE-AI Framework (2025) - All 7 Sutras
- PMLA 2002 - Anti-Money Laundering
- KYC Master Direction - Customer verification requirements
🚀 Quick Start
Installation
pip install agentbridge
Basic Usage
from agentbridge import monitor
# Your existing AI agent
class LoanApprovalAgent:
def approve_loan(self, amount, kyc_verified, reasoning):
# Your approval logic
return {"status": "approved", "amount": amount}
# Wrap with AgentBridge (2 lines!)
agent = LoanApprovalAgent()
monitored_agent = monitor(agent, api_key="ab_your_key_here")
# Use exactly like before
result = monitored_agent.approve_loan(
amount=45000,
kyc_verified=True,
reasoning="Verified income ₹80k/month, CIBIL 750+"
)
That's it! Your agent now has:
- ✅ Real-time compliance enforcement
- ✅ Behavioral drift detection
- ✅ Structuring pattern monitoring
- ✅ Automated audit logging
- ✅ AI-powered risk analysis
📊 What You Get
1. Behavioral Drift Detection
Automatically flags when your agent's decision-making patterns change:
{
"behavioral_drift": {
"status": "drift_detected",
"finding_count": 2,
"findings": [
{
"signal": "approval_rate",
"previous": "45%",
"recent": "78%",
"delta": "+33%",
"severity": "high"
}
]
}
}
2. Structuring Detection
Catches money laundering patterns like:
- Transactions just below ₹50,000 threshold
- Repeated identical amounts
- Velocity bursts (rapid-fire approvals)
- Cross-customer smurfing
{
"structuring_detected": {
"findings": [
{
"pattern": "threshold_structuring",
"severity": "high",
"description": "8 approvals between ₹40,000–₹49,999",
"str_trigger": true // STR filing required
}
]
}
}
3. AI-Powered Analysis
Groq LLM analyzes patterns and generates compliance reports:
{
"ai_analysis": {
"drift_narrative": "Agent showing increased leniency...",
"structuring_narrative": "Threshold avoidance pattern detected...",
"recommended_action": "escalate",
"compliance_officer_notes": "Review last 50 decisions..."
}
}
🔧 Advanced Configuration
Enable Verbose Logging
monitored_agent = monitor(
agent,
api_key="ab_xxx",
verbose=True # Print detailed compliance checks
)
Output:
[AgentBridge] Monitoring agent: LoanApprovalAgent
[AgentBridge] Session ID: sess_a3f9b21c
[AgentBridge] Checking compliance for approve_loan...
Action: approve
Amount: ₹45,000
KYC: True
Verdict: APPROVE
Risk: LOW
[AgentBridge] ✓ Compliance passed (120ms)
Custom Session Tracking
monitored_agent = monitor(
agent,
api_key="ab_xxx",
session_id="loan_batch_2024_04", # Track related decisions
agent_id="production_agent_v2" # Identify agent version
)
Disable Strict Mode (Allow execution on errors)
monitored_agent = monitor(
agent,
api_key="ab_xxx",
strict_mode=False # Don't block on gateway errors
)
🎯 Use Cases
1. Loan Approval Agents
@monitor(api_key="ab_xxx")
class LoanAgent:
def decide(self, application):
# Your ML model
score = self.model.predict(application)
if score > 0.7:
return self.approve(
amount=application.amount,
kyc_verified=application.kyc_done,
reasoning=f"ML score: {score}, verified income"
)
2. Transaction Monitoring
monitored_monitor = monitor(TransactionMonitor(), api_key="ab_xxx")
for txn in transactions:
result = monitored_monitor.classify_transaction(
amount=txn.amount,
confidence=0.92,
reasoning="Pattern matches known laundering typology"
)
if result['verdict'] == 'reject':
# Automatically blocked by compliance rules
flag_for_investigation(txn)
3. KYC Verification
kyc_agent = monitor(KYCAgent(), api_key="ab_xxx")
verification = kyc_agent.verify_customer(
pan="ABCDE1234F",
aadhaar_verified=True,
reasoning="Documents verified via DigiLocker"
)
🔐 Security & Privacy
- End-to-end encryption for all gateway communication
- Zero data retention option available
- On-premise deployment for regulated entities
- Role-based access control for audit logs
📈 Dashboard & Reporting
Access your compliance dashboard at https://agentbridge.in/dashboard
- Real-time decision monitoring
- Behavioral drift alerts
- Structuring pattern reports
- AI-generated compliance summaries
- Downloadable audit trails (PDF/CSV)
🛠️ API Reference
monitor(agent, api_key, **options)
Wrap an agent with compliance monitoring.
Parameters:
agent(Any): Your AI agent instanceapi_key(str): AgentBridge API keybase_url(str): Gateway URL (default: production)session_id(str): Optional session identifieragent_id(str): Optional agent identifierenable_behavioral_analysis(bool): Enable drift detection (default: True)enable_structuring_detection(bool): Enable pattern detection (default: True)strict_mode(bool): Raise errors on violations (default: True)verbose(bool): Print detailed logs (default: False)
Returns: Wrapped agent with compliance enforcement
Raises:
ComplianceError: When decision violates compliance rulesAuthenticationError: Invalid API keyAgentBridgeError: Gateway communication errors
🧪 Testing
# Install dev dependencies
pip install agentbridge[dev]
# Run tests
pytest tests/
# Type checking
mypy agentbridge/
📝 License
MIT License - see LICENSE file
🤝 Support
- Email: support@nova.team
- Docs: https://docs.agentbridge.in
- Issues: https://github.com/nova-tech/agentbridge/issues
🎉 About
Built by NOVA • Designed for Indian fintech compliance
Get your API key: https://agentbridge.in/signup
Read the docs: https://docs.agentbridge.in
Metadata
Release files for agentbridge-build-sdk 0.6.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agentbridge_build_sdk-0.6.0-py3-none-any.whl | Python 3 | none | any | Details for agentbridge_build_sdk-0.6.0-py3-none-any.whl |
Release files / agentbridge_build_sdk-0.6.0-py3-none-any.whl
| Download URL | agentbridge_build_sdk-0.6.0-py3-none-any.whl |
|---|---|
| Size | 10.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ccdf287a75437c057ab468419dc0020e2a48076c54e2dee7041a78a59aaf7023
|
|
BLAKE2b-256 checksum How to use checksums |
fff0449d365d5282ab0f035936095e3fc1ce6ff19a91af62f55b946f74b0487c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.10.11
|
Security for agentbridge-build-sdk 0.6.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.