How does the agent get the transactions from Stripe?
It connects to Stripe with your API credentials, pulls recent transactions for the selected time window, and passes them into the analysis flow as structured data.
What kinds of fraud patterns can it detect?
It can look for transaction spikes on a single card, repeated small amounts, strange merchant or category usage, location or device changes, and other odd patterns that often show up in fraud.
How does it combine rules and LLM checks?
Rule based checks catch clear problems like too many payments in a short period. The LLM step then reviews the remaining data for more subtle patterns such as odd combinations of merchant, time, and location that do not match typical behavior.
What does the Slack summary include?
The Slack message lists each flagged case with key details such as transaction id, amount, user or card reference, suspected pattern type, and a short explanation of why it was flagged.
What is in the JSON output for dashboards?
The JSON includes one object per flagged case with fields like transaction id, user or account id, reason codes, risk score, timestamps, and notes. Your fraud team can feed this into internal dashboards or queue tools for follow up.
.webp)