What it does
Four things, at once, in real time
Reads the structure
Fetches event names, field names and timestamps via Nango. Counts and relates — without looking at values or document content.
Filters PII
PII Guardian + Data Minimizer in series: per (system, resource) whitelisted fields are kept; personal or identifying fields are pseudonymised with SHA-256.
Survives network faults
Circuit breaker per connector, adaptive polling and a fallback cascade. When a source wobbles you notice nothing — Plan B activates.
Idempotent
Every event gets a fingerprint (SHA-256). The same event is never written twice. The watermark store ensures a re-run never duplicates anything.
How it works technically
The No-Data pipeline in three steps
Nango fetches
Observer Layer talks only to Nango — never directly to the customer's API. Nango handles OAuth, refresh and rate limits per system. Raw values stay in the customer node.
Trust Layer cleans
Data Minimizer lets only whitelisted fields through. PII Guardian replaces values with SHA-256 tokens — same value, same token, but never the identity.
The event table is the destination
The cleaned event gets its fingerprint and an INSERT OR IGNORE into the events table. The Process Graph Engine gets a signal — and takes over.
Where the boundary sits
Three design boundaries
✕No raw values upstream
Field values are minimised and tokenised at the Trust Layer; raw values stay in the customer node and are not passed to higher layers in normal operation.
✕No state between runs
No memory between runs. Each cycle reads from the DB, runs in memory, writes back — a crash loses nothing.
✕No raw data to the cloud
Processing runs on your Node — your server, your VPC, your machine. Raw customer operational data is not sent to an external LLM or to ABN Platform during normal operation; when external reasoning is enabled, it receives only abstracted or tokenised context.