Compress Against Learned Grammar

Pick a preset stream type, click "Load & Auto-Learn", then compress any sample record. The grammar is built from a training corpus; new records compress at 10–20× ratio.

Preset Record Generators — click to load a stream + sample record
IoT Sensor Stream
Environmental sensor — temp, humidity, CO2, battery, pressure
50-record training corpus · ~15× ratio
Malware Signatures
AV telemetry — sig hash, family, MITRE tactic, confidence
30-record training corpus · ~18× ratio
Network Flow Events
NetFlow/IPFIX — src/dst IP, ports, protocol, bytes, packets
40-record training corpus · ~14× ratio
Medical Vitals Stream
ICU telemetry — HR, SpO2, BP, respiratory rate, temp, EWS
24-record training corpus · ~12× ratio
Web Server Access Log
NGINX/Apache combined log — method, status, path, UA, bytes, ms
35-record training corpus · ~11× ratio
Device Configuration
Edge node registry — OS, firmware, CPU, RAM, rack location
20-record training corpus · ~17× ratio
Compress Record
← Back to Learn
How it works
  1. Click a preset above — loads training corpus + sample record
  2. Click Load & Auto-Learn — builds the grammar from the training corpus
  3. Click Compress — compresses the sample record against that grammar
  4. Edit the record textarea and compress again to see different results

The grammar captures statistical patterns (value ranges, field co-occurrences, cardinality) and encodes each record as a walk through the learned state space.

Walk State