Security & Trust
ArgusAI is designed as a security control plane between your applications and LLM providers. This page describes our architecture, data handling, and operational practices.
Architecture
ArgusAI runs as an OpenAI-compatible proxy (Fastify gateway) with a Next.js dashboard. Detection runs inline on every request before upstream forwarding. Audit events are persisted to MongoDB; taint and quotas use Redis when enabled.
Data handling
Prompt content is analyzed for threats and optionally redacted. Audit logs store detection metadata and truncated content per your retention policy. We do not use customer prompts to train models.
Detection pipeline
L1 signature rules, L2 heuristic/ML scoring, optional L3 LLM reasoning, and PII scanning run on ingress. Egress responses are scanned for PII when policy requires redaction.
Compliance posture
ArgusAI supports configurable policies per project, audit export on Pro plans, and SIEM integration. Contact us for enterprise compliance requirements.