AI extraction

A language model reads every incoming message and returns payment fields as structured JSON for comparison against the regex result.

Extraction model

Enable AI extraction

Small, fast models are enough — extraction accuracy is measured per model on the test panel below.

Stored encrypted. Rotate from the merchant security settings, never share it.

Sent as the system prompt ahead of every message the engine parses.

Last used today at 16:04 · 1,284 payments detected via AI this week

Expected JSON

{ "transaction_id": "", "amount": "", "currency": "", "sender": "", "receiver": "", "payment_method": "", "transaction_type": "", "timestamp": "" }

Responses that miss required fields are treated as extraction failures — the message then falls back to regex alone and loses the comparison bonus.

Cost and volume

Calls today
3,492
Avg. latency
1.2 s
Avg. cost / call
₹0.03
Extraction failures
0.4%

Test AI parser

Sends the configured model request — in this demo the response is mocked

No run yet

Run the test to see the model response and parsed fields here.