Security-First Architecture

Fundry uses a process-in-place architecture: your documents are never stored on our infrastructure. We connect to your systems via read-only OAuth, extract structured data, and delete raw files within 24 hours. Every control needed for SOC 2 certification is already built and operational.

Built on certified infrastructure

SupabaseDatabase & AuthSOC 2 Type II
VercelApplication HostingSOC 2 Type II
AnthropicLLM (Claude)SOC 2 Type II
OpenAIEmbeddingsSOC 2 Type II
StripePaymentsPCI DSS Level 1
ResendTransactional EmailSOC 2

These are our infrastructure providers' certifications. Fundry inherits these controls by building on certified platforms.

Data Protection

  • Process-in-place — raw documents never stored on our infrastructure
  • Raw files deleted within 24 hours of processing
  • PII stripped before any LLM call (OWASP LLM08:2025)
  • AES-256-GCM encryption at rest; TLS 1.3 in transit
  • Encrypted PII vault with numbered placeholder substitution

AI Governance

  • 4-tier model routing — RESTRICTED data never sent to open-source APIs
  • Agent autonomy levels 0-3 (Level 3 = human-only actions)
  • 5-layer quality stack: RAG grounding, confidence scoring, evaluator review, citation trails, human approval
  • Privacy router enforces data classification before every LLM call
  • Confidence scoring on every output — green (>85%), yellow (70-85%), red (<70%)

How we measure accuracy

DDQ accuracy is graded by a 5-layer quality stack: RAG-grounded retrieval, per-answer confidence scoring, evaluator agent review (Claude Opus), citation verification, and mandatory human approval before any output reaches LPs. Live baselines are captured inagent_eval_runsand posted publicly once the firm-KB target band is met.

Access & Compliance

  • Row-level security on every table with firm-scoped isolation
  • MFA support with Supabase Auth
  • 4-role RBAC: Owner, Admin, Member, Viewer
  • GDPR data minimization + right to erasure; CCPA compliant
  • SEC audit trail for all agent outputs and user actions

SOC 2 Readiness

All SOC 2 controls are implemented and operational. We haven't yet undergone the formal audit — our infrastructure providers carry the certifications today, and we plan to complete our independent SOC 2 Type II audit within 12 months of launch.

Role-based access control (Owner, Admin, Member, Viewer)
Multi-factor authentication support
Audit logging for all agent outputs and user actions
AES-256-GCM encryption at rest for all data
TLS 1.3 encryption in transit for all connections
PII detection and stripping pipeline (OWASP LLM08:2025)
Secrets scanning and rotation procedures
Incident response procedures with defined SLAs
Data retention and deletion policies
Row-level security enforcing multi-tenant isolation

Document Processing Flow

How your data moves through Fundry — and where it gets deleted

Customer System

Google Drive, Gmail, SharePoint

Read-Only OAuth

Fundry never writes to your systems

PII Detection & Stripping

Regex + NER scan, placeholder substitution, vault storage

LLM Processing (Tier 1 Only)

PII-free text sent to Anthropic Claude for extraction

Structured Data Extraction

Q&A pairs, classifications, parsed fields stored encrypted

Raw Document Deletion

Within 24 hours — no raw content retained

Incident Response

Defined response times by severity. Breach notification within 72 hours per GDPR.

SeverityDescriptionResponse Time
SEV 1
Data breach, PII exposure1 hour
SEV 2
Auth bypass, incorrect output4 hours
SEV 3
Feature outage, degraded accuracy24 hours
SEV 4
UI bug, cosmetic issue72 hours

Questions about security?

We're happy to discuss our security architecture, share our DPA, or schedule a call with our engineering team.

This page is provided for informational purposes and does not constitute legal advice. Consult legal counsel for compliance determinations specific to your organization.