Payments & Fintech
PCI-DSS aware. PAN/CVV redaction, high-value review, OFAC screening.
- PCI · Block PAN storage
- PCI · Block CVV/CVC in arguments
- High-value transfer — human approval
- AML · Sanctions screening
- Refunds require audit reason
Live demo · ~ 90 seconds end-to-end
Watch AEGIS scan a real payments-agent repo, extract the workflow, install a Payments policy pack, and block a $47,000 transfer that carries a credit card number — all on this page.
One CLI command. Tree-sitter parses every Python / JS / TS file
and finds every LLM call, every @tool, every
MCP config — even the ones imported dynamically.
$ agentguard scan .
Scanned 8,402 source files under ~/acme-payments
Repo: acme-payments@4.7.2 · Branch: main
CANDIDATES:
KIND CONF FRAMEWORK ENTRY PATH
import high anthropic yes src/agents/billing.py
import high langgraph WF src/orchestrator/router.py
import high crewai WF src/crews/refunds.py
http medium openai-http svc/legacy_proxy.py
WORKFLOW GRAPH (langgraph)
nodes: 14 (1 entry · 11 agent · 1 router · 1 terminal)
edges: 19 (16 control · 3 handoff)
tools: 22 — stripe_charge, send_email, db_query, refund_card, …
risk_signals: 3 sensitive-relays
db_query → send_email
stripe_charge → http_post
refund_card → file_write
✓ ready to instrument — run `agentguard inject` next The scanner doesn't just count files — it builds the actual control + handoff graph that your LangGraph / CrewAI / AutoGen orchestrator runs. That's what every policy gets attached to.
No DSL. No JSON Schema. Type what you want — AEGIS generates a policy bundle that passes the AJV self-test before it's saved.
"Refunds over $5,000 should always go to a human. Block any tool call that includes a raw credit card number. And if the counterparty country is on the OFAC list, block immediately and page on-call."
"id": "refund-high-value-pending",
"risk_level": "HIGH",
"decision": "pending",
"applies_to": ["refund_card"],
"when": { "amount": { ">": 5000 } }
,
"id": "pci-no-pan-in-args",
"risk_level": "CRITICAL",
"decision": "block",
"applies_to": ["*"],
"match": { "pattern": "\\b([0-9][ -]?)19\\b" }
,
"id": "ofac-sanctions-block",
"risk_level": "CRITICAL",
"decision": "block",
"applies_to": ["stripe_charge", "http_post"],
"match": { "country": ["CU", "IR", "KP", "SY"] },
"on_block": { "page": "oncall-secops" }
Curated bundles for regulated industries. Each pack ships with 5 named policies a compliance lead can defend in an audit.
PCI-DSS aware. PAN/CVV redaction, high-value review, OFAC screening.
HIPAA-aware. PHI redaction, minimum-necessary, treatment-change approval.
BSA / AML / SOX. KYC enforcement, cross-border review, sanctions block.
GDPR + CCPA + multi-tenant isolation. Erasure, scope, prod-write guards.
The billing agent tries to charge a card. The card number leaks into the tool argument. AEGIS blocks it in under 5 ms and writes the decision into the audit log.
The bottom row, orchestrator → db_query · DROP TABLE users,
fires the sql-injection default policy. The pending
finance-bot → transfer · amount: $47,200 fires the Payments
pack's high-value-transfer-review. Both pages your on-call
via the configured destination.
Free tier: scan, install one pack, watch 1,000 tool calls a day. No card.