01 / Platform
Five layers.
One ledger.
Neuryn is not a dashboard bolted onto your warehouse. It sits in the ingest path, which is the only place where a check is still cheap.
02 / Architecture
What happens to a row.
Follow a single record from the moment it arrives to the moment an auditor asks about it eighteen months later.
- 01
Ingest
Native connectors for every common source, plus a generic REST, webhook and SFTP path. Schemas are inferred on the first pass and versioned from then on. No migration, no agent to install inside your database.
- Connectors
- 38
- Inference
- First pass
- Agent
- None required
- 02
Validate
Rules compile to bytecode and run on the ingest node itself, before the row is written anywhere. That is what makes row-level checking affordable at four million rows a minute instead of sampling one in a thousand.
- Throughput
- 4.2M / min
- p99 latency
- 11 ms
- Coverage
- Every row
- 03
Quarantine
Failing rows go to a separate store with the rule and rule version that rejected them. Nothing is deleted. Loosening a rule later becomes a query rather than a leap of faith.
- Retention
- 90 days default
- Typical volume
- 0.1 to 0.8%
- Replay
- One click
- 04
Encrypt
AES-256 at rest with keys you hold. Policy is declared once and pushed to every provider you run on. Usage passes through a broker so there is one ordered access log, in one clock, in one format.
- Cipher
- AES-256-GCM
- Key custody
- Customer managed
- Clouds
- AWS, GCP, Azure
- 05
Audit
Every validation, every decrypt, every policy change lands in an append-only ledger. Seven year retention on Enterprise. Exportable in a format an auditor will accept without a translation layer.
- Ledger
- Append only
- Retention
- Up to 7 years
- Export
- CSV, Parquet, JSON
03 / Sources
Every source, no agent.
Read-only credentials, scoped to a prefix. If a connector needs more than that to work, we consider it a bug.
38 native connectors
- Snowflake
- BigQuery
- Redshift
- PostgreSQL
- MySQL
- SQL Server
- Kafka
- Kinesis
- Pub/Sub
- Amazon S3
- Google Cloud Storage
- Azure Blob
- Salesforce
- HubSpot
- Stripe
- REST
- Webhook
- SFTP
- +20 more →
04 / One row
What happens to a single record.
The five layers above describe the system. This describes one row, from the moment your connector reads it to the moment someone can prove what was done to it.
- 01
Read
The connector reads the row with the read-only credential you issued. Nothing is written back to the source, ever.
- 02
Infer
On the first pass the schema is inferred and versioned. Every later row is compared against that version, not against a guess.
- 03
Validate
Compiled rules run on the ingest node, before the row is written anywhere. Pass and it continues untouched.
- 04
Quarantine
Fail and it goes to a separate store with the rule and rule version that rejected it. Nothing is deleted.
- 05
Replay
Loosen the rule later and the quarantined rows come back through the same path. A query, not a leap of faith.
- 06
Prove
Every one of those steps is already in the append-only ledger, in one clock and one format, exportable.
No step above requires an agent inside your database. See what credentials we ask for.
05 / Compliance
Controls that run as code.
Declared in a repository, enforced continuously, evidenced automatically. The audit becomes an export rather than a project.
- SOC 2 Type II
- Continuous evidence collection against the full trust services criteria.
- GDPR
- Lawful basis, retention windows and erasure requests tracked per field.
- HIPAA
- BAA available. PHI fields tagged at ingest and excluded from exports by default.
- FedRAMP
- Moderate baseline on the self-hosted deployment. Available on Enterprise.
06 / Get started
Connect one source. See what it catches.
The free tier validates a million rows a month. That is usually enough to find the thing you did not know was broken.