Glossary
A reference for terms used throughout AegisGate documentation. Organized by category for easier browsing.
AI Security
| Term | Definition |
|---|
| A2A | Agent-to-Agent protocol. AegisGate secures A2A communication with mTLS, HMAC integrity, and capability enforcement. See pkg/a2a/. |
| ACP | Agent Capability Policy. Controls which actions an AI agent is permitted to perform. See pkg/acp/. |
| Adversarial Pattern | A prompt or content pattern designed to bypass security controls (prompt injection, jailbreak, encoding evasion). AegisGate detects 176 such patterns. |
| ATLAS | MITRE ATLAS (Adversarial Threat Landscape for AI Systems). A framework cataloging attacks against AI systems. AegisGate maps detections to ATLAS techniques. |
| CheckFunc | The Go type for an automated compliance control check. Each CheckFunc evaluates a specific control and returns compliant/partial/non-compliant status. |
| Fail-Closed | A security principle where the system blocks traffic if it cannot scan or verify it. AegisGate is fail-closed by default — if scanning fails, the request is blocked, not allowed through. |
| MCP | Model Context Protocol. A protocol for AI agents to invoke external tools. AegisGate secures MCP server interactions with capability enforcement and content scanning. See pkg/mcpserver/. |
| Prompt Injection | An attack where malicious instructions are embedded in content to manipulate an AI model’s behavior. AegisGate detects and blocks prompt injection patterns. |
| PII | Personally Identifiable Information. AegisGate detects and redacts PII (SSN, credit card, email, phone, DOB, health plan ID) in AI traffic. |
| Trust Framework | AegisGate’s system for cryptographic agent identity, per-session trust scoring, and signed attestations. See pkg/attestation/. |
Compliance
| Term | Definition |
|---|
| Automated Control | A compliance control that AegisGate can verify programmatically using one of 4 automation methods. 1,457 of 2,043 controls are automated (71.3%). |
| Audit Trail Evidence | One of 4 automation methods. Examines audit logs for required evidence artifacts (access logged, data retention enforced, incident response tracked). |
| CCPA/CPRA | California Consumer Privacy Act / California Privacy Rights Act. Developer-tier framework with 26 controls. |
| CMMC L2 | Cybersecurity Maturity Model Certification Level 2. DoD framework for defense contractors. Enterprise-tier, 110 practices. |
| Config State Verification | One of 4 automation methods. Reads running configuration and validates it against framework requirements (TLS enabled, audit logging on). |
| Cross-Framework Mapping | One of 4 automation methods. Maps evidence from one framework to satisfy another’s requirements (HIPAA access logging → SOC 2 CC6.1). |
| Detection Engine State | One of 4 automation methods. Checks that detection rules and scanners are active and covering required patterns. |
| EU AI Act | European Union AI Act. First comprehensive AI regulation. Professional-tier, 120 controls across 8 risk categories. |
| FedRAMP | Federal Risk and Authorization Management Program. U.S. government cloud security authorization. Enterprise-tier, 170 controls. |
| Framework | A compliance framework (HIPAA, SOC 2, EU AI Act, etc.). AegisGate supports 31 frameworks with full Go implementations. |
| HITRUST CSF | HITRUST Common Security Framework. Enterprise-tier framework for healthcare and regulated industries. |
| Manual Control | A compliance control that requires human processes (organizational policies, legal agreements, physical security, HR training). 586 of 2,043 controls are manual (28.7%). |
| NIST AI RMF | NIST AI Risk Management Framework 1.0. Community-tier framework with 50 controls across 4 functions: GOVERN, MAP, MEASURE, MANAGE. |
| SOC 2 | Service Organization Control 2. Developer-tier framework for service organizations. 64 controls across Trust Services Criteria. |
| TISAX | Trusted Information Security Assessment Exchange. European automotive industry security assessment. Enterprise-tier. |
Deployment
| Term | Definition |
|---|
| Air-Gapped | A deployment with no internet access. AegisGate supports fully offline operation with the air-gapped deploy profile. |
| Config Precedence | The order in which configuration values are resolved: CLI flags > env vars > config file > deploy profile > defaults. |
| Deploy Profile | A predefined configuration preset (v4.2.0+). 5 profiles: quickstart, small-team, production, high-security, air-gapped. Each populates all config fields with sensible defaults. |
| Guided Setup | The v4.2.0 initiative comprising deploy profiles, setup wizard, config validation, and maintenance windows. Makes deployment accessible without DevOps expertise. |
| Maintenance Window | A planned downtime period (v4.2.0+) where the platform returns 503 to proxy requests while keeping health/version endpoints accessible. Uses atomic.Bool for zero-overhead state checks. |
| Setup Wizard | Interactive or non-interactive environment detection tool (v4.2.0+). Auto-detects Docker, K8s, systemd, bare metal; recommends a profile; generates a validated config. |
| Standalone Mode | Running AegisGate with --embedded-mcp flag, which starts the MCP server in-process without external MCP infrastructure. |
Architecture
| Term | Definition |
|---|
| 6 Pillars | AegisGate’s six protection domains: HTTP API, MCP, A2A, ACP (Agent Capability Policy), Response, and Trust. |
| Detection Pattern | A regex or ML-based pattern that identifies a specific threat type (secret, XSS, PII, prompt injection, etc.). 176 patterns are wired into every request/response. |
| Graceful Degradation | When ML detection is unavailable (no ONNX Runtime), AegisGate falls back to regex-only detection (83.1% coverage) instead of failing. |
| ML Threat Detection | CNN-BiLSTM neural network (1.58M params) for adversarial pattern detection with 100/100 evasion resistance. Professional+ tier. |
| ONNX Runtime | Open Neural Network Exchange runtime. Optional dependency for ML threat detection. Without it, AegisGate uses regex-only mode. |
| Open-Core | AegisGate’s licensing model: Community edition is Apache 2.0 open source; Developer/Professional/Enterprise features are proprietary. |
| RBAC | Role-Based Access Control. Developer+ tier feature for managing user permissions. |
| SIEM | Security Information and Event Management. AegisGate forwards audit events to 11 SIEM platforms (Splunk, Elasticsearch, QRadar, etc.). Professional+ tier. |
| SOAR | Security Orchestration, Automation, and Response. AegisGate integrates with 4 SOAR platforms (PagerDuty, Jira, ServiceNow). |
| Trust Score | A per-session trust rating computed by the Trust Framework based on agent identity, behavior, and posture. |
See also: Configuration Reference, CLI Reference, Compliance Frameworks.