Skip to main content
Trust Center

How We Handle Your Data

No marketing claims. No false certifications. Here's exactly what happens to your documents — and what you can control.

Data Lifecycle

1

Upload

Document sent via HTTPS (TLS 1.3) to our backend

2

Process

Text extracted and sent to LLM API for analysis

3

Store

Results stored in encrypted PostgreSQL database

4

Expire / Delete

Auto-purged per retention policy, or manual deletion

Encryption Posture

In TransitTLS 1.3All connections: frontend, backend, database, LLM API
At RestAES-256Via Railway managed PostgreSQL (inherited from provider)
OAuth TokensFernetSymmetric encryption for stored provider tokens
Passwordsbcrypt (cost 12)Industry-standard password hashing

LLM Data Handling

Your document text is sent to third-party LLM providers (Google Gemini, OpenAI) for analysis via their API.

  • Providers do not train on API data per enterprise terms of service
  • Data is not stored by providers beyond processing window (typically 30 days for abuse monitoring)
  • Enterprise API agreements with both Google and OpenAI

Retention Policy

Default: indefinite (data kept until you delete it). You can configure automatic purging at 30, 90, or 365 days. When data is purged: review results are nullified, uploaded files are deleted, metadata tombstone is kept for audit trail.

Your Rights (GDPR)

Data Portability

Download all your data as a ZIP file: profile, reviews, credit transactions, audit log.

Log in to manage

Right to Erasure

Permanently delete all your data: reviews are purged, knowledge base removed, account anonymized.

Log in to manage

Compliance Posture

PAK4L Controls

  • JWT authentication with refresh tokens
  • bcrypt password hashing (cost 12)
  • Atomic credit deduction (no race conditions)
  • Tier-based access control (RBAC)
  • Append-only audit log (tamper-evident)
  • Configurable retention policies
  • GDPR data export & erasure endpoints
  • Rate limiting on auth endpoints

Inherited from Providers

  • SOC 2 Type IIRailway, Google Cloud, OpenAI, Vercel
  • ISO 27001Google Cloud
  • GDPRRailway (EU), Google Cloud, OpenAI
  • PCI DSS Level 1Stripe (payments only)
  • AES-256 Disk EncryptionRailway managed PostgreSQL

Deployment Modes

Available

SaaS

Multi-tenant. EU-hosted database. Global CDN. Managed by PAK4L.

Roadmap

Dedicated

Single-tenant. Your choice of cloud region. Full data isolation.

Roadmap

On-Premise

Self-hosted in your infrastructure. Air-gapped capable. Full control.

Data Residency

ComponentProviderRegion
Database (PostgreSQL)RailwayEU (Frankfurt)
Frontend CDNVercelGlobal (Edge)
LLM API CallsGoogle / OpenAIUS
PaymentsStripeUS / EU

AI Agent Reliability Benchmark

We continuously evaluate our multi-agent system against an annotated corpus of 150+ documents across legal, compliance, financial, and technical domains. All metrics are self-reported with full methodology disclosure.

69.2%

Issue Detection Rate

vs. annotated ground truth

32.1%

Cross-Agent Agreement

severity consensus ≥2 agents

78.3%

Severity Accuracy

±1 level tolerance

55.6%

Consistency Score

same doc → same results

8.5x

Thoroughness Multiplier

issues found vs. baseline

Methodology

Test Dataset

  • 153 documents, 2,847 pages total
  • Contracts, compliance reports, financial analyses, technical specs
  • 1,204 expert-annotated issues across all severity levels
  • 4 languages: Italian, English, Spanish, German

Evaluation Process

  • Double-blind: annotators and agents see documents independently
  • Each document reviewed 3x to measure consistency
  • Cross-agent debate validated against expert consensus
  • Benchmark re-run monthly after each model or agent update

Transparency Note

These metrics are self-reported from our internal benchmark suite. We are actively pursuing independent third-party audit and SOC 2 Type II certification. Benchmark results may vary based on document complexity, language, and domain. PAK4L is an AI-assisted tool — final decisions should always involve human review.

Agent Debate Transparency

Every PAK4L review includes a full Boardroom transcript showing how agents reach their conclusions. This means you can always verify:

1

Which agent flagged each issue

Full attribution with agent role and expertise area

2

Why they disagree

Cross-agent disputes and severity debates are logged

3

How consensus was reached

Coordinator synthesis with evidence-based reasoning

Questions about security?

Contact our team for detailed security documentation, custom deployment options, or compliance questionnaire support.