The AI Security Gateway
Autonomous Agents Deserve.

QuiGuard is a production-grade AI Security Gateway that sits between your autonomous AI agents and LLM providers like OpenAI, Anthropic, and Google. It intercepts every request and response, automatically detecting and redacting personally identifiable information (PII) including names, emails, Social Security numbers, credit card details, and custom patterns. Unlike simple API wrappers, QuiGuard uses Microsoft Presidio NER with spaCy language models for context-aware entity recognition, deterministic hash-based placeholders for lossless round-trip restoration, and recursive JSON traversal that secures nested tool calls and function arguments.

QuiGuard employs a multi-layered detection pipeline. First, it uses Microsoft Presidio's Named Entity Recognition (NER) engine powered by spaCy's en_core_web_lg language model (400MB) to identify 19+ PII categories including personal identifiers, financial data, government IDs, healthcare records, and technical secrets. Second, it applies custom regex patterns for organization-specific identifiers like employee IDs, internal document references, and proprietary API key formats. Each detected entity is replaced with a deterministic hash-based placeholder (e.g., <EMAIL_ADDRESS_abc123>) that preserves uniqueness, enabling accurate LLM responses while ensuring no raw PII reaches external providers. The system handles overlapping detections, supports configurable confidence thresholds, and offers multiple action modes: redact, mask, fake, block, or warn.

Yes. QuiGuard is specifically designed to help organizations meet GDPR, HIPAA, SOC 2, and similar regulatory frameworks. For GDPR compliance, QuiGuard prevents personal data from being transferred to third-party LLM providers by redacting PII at the network edge before data leaves your infrastructure. For HIPAA, the healthcare entity detection category identifies medical license numbers, patient references, and protected health information patterns. For SOC 2, the Secure Ledger Dashboard provides real-time audit trails with event-level logging, risk detection tracking, entity-level forensic analysis, and exportable compliance reports. All audit logs include timestamps, event types, blocked entity classifications, sanitized content, and user attribution through Supabase Row-Level Security.

Absolutely. QuiGuard is completely provider-agnostic and works as a transparent reverse proxy with any OpenAI-compatible API. It currently routes through OpenRouter, giving you access to over 100 models including GPT-4, Claude, Gemini, Llama, Mistral, and open-source alternatives. The proxy intercepts traffic at the HTTP level, so your application only needs to change its API endpoint to point to QuiGuard. No SDK changes, no library updates, and no code modifications to your existing AI agent logic. Smart model routing automatically selects the optimal model based on intent classification: reasoning tasks use specialized models while coding and creative requests are routed to fast-response models.

QuiGuard deploys in under 5 minutes using Docker. For production, the recommended stack is Railway for the backend (auto-deploys from Git, no cold starts, up to 4GB RAM) and Vercel for the management dashboard. The Docker image includes all dependencies: FastAPI, Microsoft Presidio, spaCy language models, and Gunicorn for production serving. Configuration is managed through a single policy.yaml file or the web-based No-Code Policy Editor. Environment variables for API keys, Supabase credentials, and provider settings are the only required configuration. Railway auto-detects the Dockerfile and handles SSL, scaling, and health monitoring automatically.

The No-Code Policy Editor is a web-based interface that allows you to customize your security rules without editing YAML files. It provides five configuration sections: Global Settings (confidence threshold slider, action mode selector), PII Detection (toggle 19+ entity types on/off by category), Custom Patterns (add organization-specific regex patterns with confidence scoring), Agent Security (manage blocked domains, prompt injection defenses, and tool restrictions), and Model Routing (configure intent-to-model mappings). Changes take effect immediately for new requests and are stored as per-user overrides in Supabase, with the server computing only the diff from defaults for efficient storage.

QuiGuard implements a complete multi-tenancy system with SHA-256 hashed API keys, subscription-based access control, and per-user rate limiting. Each API key is prefixed with 'qg_sk_' for identification and hashed before storage, meaning raw keys are never persisted and shown only once at creation time. The system enforces plan-based limits: Community (1 API key, basic PII), Starter ($49/mo, 3 keys, extended PII), Pro ($149/mo, 10 keys, full PII with agent security), and Enterprise (unlimited keys, custom configuration). Monthly request counting tracks usage against plan limits with automatic 429 responses when thresholds are exceeded. All data isolation is enforced through Supabase Row-Level Security policies.

QuiGuard detects 19+ PII categories across five domains. Personal identifiers include person names, email addresses, phone numbers, physical locations, and date/time values. Financial data covers credit card numbers, IBAN codes, US bank account numbers, and cryptocurrency wallet addresses. Government and legal IDs encompass US Social Security Numbers, US passports, driver licenses, US ITINs, UK National Insurance numbers, and Singapore NRIC. Healthcare identifiers include medical license numbers and HIPAA-relevant patterns. Technical secrets cover IP addresses, URLs, domain names, and API key formats. Additionally, users can create unlimited custom regex patterns for organization-specific identifiers like employee IDs (EMP-123456), confidential document references (Conf-SAL-2024), and internal API key formats.