Real-Time Policy & Data Protection

Govern your data
before it reaches your AI.

Auvis intercepts ingestion pipelines to enforce granular access controls, sanitize sensitive records, and guarantee compliance audit trails before data reaches vector databases or AI context windows.

Get startedView on GitHub
19.7k req/s
< 1ms overhead
0% error rate (400k reqs)
GDPR (Art. 30)
HIPAA (Safe Harbor)
PCI-DSS v4.0
SOC 2 Type II
CCPA / CPRA
SOX Compliance
ISO/IEC 27001
NIST SP 800-171
FedRAMP Moderate
CMMC 2.0
GDPR (Art. 30)
HIPAA (Safe Harbor)
PCI-DSS v4.0
SOC 2 Type II
CCPA / CPRA
SOX Compliance
ISO/IEC 27001
NIST SP 800-171
FedRAMP Moderate
CMMC 2.0
THE GAP

WhencompaniesconnectinternaldatatoAImodels,everysourceaccesscontroldisappears.HRrecords,salaryfiles,andcustomerPIIgetflattenedintovectordatabaseswithzeropermissions.Auvissitsdirectlyinsideyourpipelinestoenforcegovernanceinflightmasking,redacting,andauditingeveryrecordbeforeyourAIeverseesit.

One software.Every pipeline governed on the record.

Everything here ships from the first seat. Nothing is an upgrade.

01 / 05

Policy Enforcement, rule by rule in flight

Mask, redact, tokenize, or drop any field in real time. Dynamic conditions evaluate row-level rules with deterministic priority resolution.

  • Mask, redact, tokenize, or drop any field
  • Row-level filtering with dynamic conditions

Pipeline Policies

mask-pii.yaml
Hot-reload
  • email → mask***@example.com
  • salary → removerole != 'hr'
  • phone → mask***-***-0198
  • ssn → dropGDPR enforced
Eval <5msExplicit > Wildcard

THE PIPELINE ARCHITECTURE

Set the standard, connect your agents, keep the record. One platform.

Auvis operates as a transparent proxy and streaming engine inside your network, enforcing governance policies in four continuous, deterministic steps.

01

Intercept

Every record, before it lands anywhere.

Auvis sits inside your ingestion pipelines; Kafka streams, Confluence syncs, database CDC intercepting records in flight before they reach vector databases or LLM context windows.

01 / 04
02

Evaluate

Matched against policy in under 5ms.

Each record is evaluated against your defined rules; role, resource, and field using a compiled decision tree with zero added latency your team will notice.

02 / 04
03

Transform & Write

Mask, redact, and fail-closed sync.

Sensitive fields are rewritten based on policy rules, and governed records are written synchronously to Postgres, pgvector, Weaviate, or Pinecone with fast-fail circuit breakers.

03 / 04
04

Audit

Proof for every decision, automatically.

What was masked, removed, allowed, and denied is logged asynchronously as JSONL ready for Splunk, Datadog, and compliance audits without touching the hot path.

04 / 04

Stop waiting on the security review.

19.7k req/s, sub-millisecond proxy overhead, zero application code changes. Auvis governs the pipeline you already have.

Read the DocsView on GitHub
Open source (Apache 2.0)•Zero SDK Lock-in•Docker Ready

Frequently Asked Questions

Everything you need to know about pipeline governance, latency overhead, and destination fail-closed security.

Policy evaluation runs in under 5ms per record using a compiled decision tree, and the gateway's internal processing overhead is sub-millisecond (+0.70ms). In 400,000-request load tests across dedicated VPC instances, Auvis delivers 19.7k req/s (19,682 req/sec) with 0% error rate. Secondary destination sink writes are also decoupled with <0.1ms circuit breaker fast-fails, so slow or offline vector databases never delay client HTTP responses.