Governance layer for agents.

Compliance observability for every AI agent in finance — logged, auditable, regulator-ready.

Financial institutions face $14.8M average compliance breach costs. AI agents are scaling faster than governance — XeroML closes the gap. Plug in, evaluate, prove compliance.

INTEGRATION TIME < 30 min
FRAMEWORKS SUPPORTED 20+

REAL-TIME COMPLIANCE TRACKING

Every agent decision. Audited in real time.

XeroML wraps your AI agents with a compliance observability layer — every decision is logged, mapped to the applicable regulation, and scored before it reaches production. No manual audits. No spreadsheets. No lag.

🔍

Live Agent Monitoring

Every lending, fraud, and payment decision tracked in real time — not reviewed weeks later in spreadsheets

AI Audit Agent

An autonomous agent reads decision logs, scores against live regulations, and flags violations before regulators do

📋

Regulator-Ready Outputs

Adverse action notices, audit packs, and CCO reports — auto-generated in seconds, not weeks

USER / REQUEST

REQUEST.RECEIVED()

KYC submission, loan approval, or fraud check enters your AI workflow.

AGENT

AGENT.EVALUATE()

Agent analyzes context, reasons over policy, and proposes a decision with evidence.

DECISION.PROPOSE()

CONTEXT.SUMMARY()

COMPLIANCE SDK INTERCEPT

RULE.VERIFY()

SDK intercepts the decision in real time and checks ECOA, SR 11-7, Fair Lending, and BSA/AML rules before execution.

APPROVE / BLOCK

COMPLIANCE.FLAGS

AUDIT.TRAIL

REASONING.LOG

SYSTEM / RECORD

DECISION.EXECUTE()

Approved decisions run, blocked actions halt, and every outcome is sealed in immutable logs.

Agent decision flow

XeroML intercept at step 3

Real-time compliance evaluation

END-TO-END EVAL & TRACING

See every step. Score every output.

Full Workflow Traces

Trace ML pipelines, LLM calls, retrieval steps, and agent decisions in a single view

LLM Judge Evaluation

Calibrated judges score every output for accuracy, grounding, compliance, and coherence

Real-Time Monitoring

Latency, token cost, eval scores, and drift detection across every agent run

Xeroml Trace Explorer

The fastest way to evaluate, debug, and improve your financial AI workflows.

Explore Trace Explorer
  1. Connect your agent

    Drop in our SDK with 3 lines of code. Supports LangChain, LlamaIndex, CrewAI, custom frameworks, and any OpenTelemetry-compatible pipeline.

  2. Configure LLM judges

    Calibrated judges score factual grounding, numerical accuracy, bias/fair lending, and reasoning coherence. Failing scores block deployment.

  3. Set compliance rules

    Map regulatory requirements to automated checks — fair lending, advisory, PII masking, data protection. Flag decisions lacking explainability.

  4. Deploy to production

    One Docker Compose command deploys Xeroml inside your VPC. No external dependencies, no data egress — production-ready in under 30 minutes.

  5. Monitor & improve

    Live dashboards for traces, eval scores, and compliance checks. Low scores surface prompt fixes automatically.

Trace Explorer UI
Connect a new agent
xeroml==2.1.0 pip install xeroml OpenTelemetry compatible
Python SDK LangChain CrewAI
# xeroml==2.1.0 · 3 lines to full observability from xeroml import trace import os trace.init( project="loan-underwriting", api_key=os.environ["XEROML_KEY"] ) @trace.agent("credit-risk-assessor") def assess_risk(applicant: dict) -> dict: return llm.run(risk_prompt, applicant)
Connected · loan-underwriting 847 traces collected

COMPLIANCE INTELLIGENCE

Score every decision. Prove compliance.

Regulatory Scoring

Automated checks mapped to RBI, SEBI, ECOA, and Basel III — every agent output scored before it reaches production

Explainability Reports

Generate ECOA-compliant adverse action notices and audit-ready reasoning trails for every AI decision

Drift Alerts

Statistical drift detection (PSI, KS tests) triggers alerts when agent behavior deviates from approved baselines

Compliance intelligence dashboard
Compliance scorecard
loan-underwriting
RBI SEBI ECOA
Fair lending — explainability ECOA
0.96 PASS
Advisory suitability SEBI IA
0.91 PASS
KYC verification RBI
0.74 FAIL
PII masking DPDP
0.99 PASS
Data protection — consent DPDP
1.00 PASS
7d compliance
96% +14%

Compliance Intelligence

Scores every agent decision against your regulatory frameworks, surfaces explainability gaps, and generates audit-ready reports.

Explore Compliance Engine
  1. Regulatory scoring

    Every agent output is checked against your compliance rules — fair lending, advisory suitability, KYC, PII masking. Failures block production.

  2. Explainability reports

    Auto-generate audit-ready reasoning trails for every flagged decision. One click to export for regulators.

  3. Drift detection

    Continuous monitoring compares agent behavior against approved baselines. Get alerted the moment output distributions shift.

CONTINUOUS IMPROVEMENT

Your agents get better with every run

Regulatory Judges

ECOA, TILA, SR 11-7, and fair lending judges score every agent output before it reaches production

Backtest & Validate

Champion-challenger testing against historical decisions — verify SR 11-7 thresholds before promoting to production

Audit Trail

Every improvement logged with timestamps, score deltas, and approvals — exportable for SR 11-7 model governance audits

The Regulatory Improvement Loop

LLM judges mapped to ECOA, TILA, and SR 11-7 score every output. Violations are diagnosed, fixes backtested against historical decisions, and every improvement is audit-logged.

See How It Works
  1. Regulatory judge scores

    Every agent output scored by judges mapped to ECOA, TILA, SR 11-7, and fair lending frameworks. Failures block deployment.

  2. Violation diagnosis

    When a regulatory judge fails, Xeroml identifies the specific violation, traces the root cause, and generates a regulation-compliant prompt fix.

  3. Backtest & validate

    Apply the fix, run champion-challenger testing against historical decisions, compare before/after across all judges, then promote.

  4. Improvement audit trail

    Every fix logged with timestamps, score deltas, and approvals. Export the full trail for SR 11-7 audits.

Improvement loop UI
Regulatory judge scores — loan-underwriting
loan-underwriting
ECOA TILA SR 11-7 HMDA
ECOAECOA adverse action completeness 0.94
PASS
TILATILA APR calculation accuracy 0.68
FAIL — threshold 0.95
SR 11-7SR 11-7 model risk documentation 0.97
PASS
HMDAFair lending bias detection 0.91
PASS
RBIRBI KYC verification completeness 0.96
PASS

REQUEST EARLY ACCESS

Stop wondering if your AI agents will pass the next regulatory exam.

XeroML gives compliance teams real-time proof that every AI decision meets ECOA, TILA, SR 11-7, and Fair Lending requirements. Deployed in your VPC in under 30 minutes. No data leaves your infrastructure.

Book a Demo