Glossary
Terms used throughout this document, alphabetized. Cross-references point to the section where the term is treated in depth.
| Term | Definition |
|---|---|
| A2A | Agent-to-Agent protocol. Emerging standard for agent interoperability. Core implemented in v2.0.64–2.0.105 (registry, agent-card discovery, JSON-RPC, X-A2A-Key + JWT auth); advanced features deferred (§42.11, §43.1). |
| ACL | Access Control List. Per-resource allow/deny rules; used in MCP proxy and Gateway routes (§14, §17). |
| ACS | Assertion Consumer Service. The SAML 2.0 endpoint that receives the IdP’s signed assertion (§16.4). |
| ALPN | Application-Layer Protocol Negotiation. The TLS extension that picks the protocol (HTTP/1.1 / HTTP/2 / HTTP/3) per connection (§30). |
| AST | Abstract Syntax Tree. Code-structure representation used by the Knowledge Extractor (§10). |
| BAA | Business Associate Agreement. HIPAA contract between a covered entity and a vendor that handles PHI (§20). |
| bcryptjs | Pure-JavaScript bcrypt implementation. Replaced the native bcrypt in v2.0.42 for Bun 1.3 / Node 24 ABI compatibility (§42.3). |
| Bun | Modern JavaScript/TypeScript runtime + bundler. AgentWatch is pinned to Bun 1.3.14 (§2.1, §30.6). |
| CASB | Cloud Access Security Broker. Egress-control product class (Zscaler, Netskope, Forcepoint); AgentWatch can route egress through any CASB (§39 topology 2). |
| CDP | CRL Distribution Point. URL where a client can fetch the certificate-revocation list; the AgentWatch Connect agent serves an empty CRL from its local control API so CDP URLs work cross-platform (§29.3). |
| CIDR | Classless Inter-Domain Routing. Block notation for IP ranges, e.g., 10.0.0.0/8. Used in IP allowlisting (§5.3, §17). |
| ClickHouse | Columnar analytics database for real-time observability rollups. Optional in AgentWatch (§22, §39). |
| CRL | Certificate Revocation List. List of revoked X.509 certificates served from a CDP. |
| CWE | Common Weakness Enumeration. Standardized vulnerability-class identifier used in security scans (§11). |
| DLP | Data Loss Prevention. Inline scanning + blocking of sensitive data (PII / PHI / PCI / secrets) before it reaches an upstream provider (§18). |
| DR | Disaster Recovery. Cross-region failover and restore (§31, §40.3). |
| FedRAMP | Federal Risk and Authorization Management Program. U.S. government cloud-security baseline; AgentWatch is FedRAMP-compatible via air-gapped deployment (§20). |
| FOCUS | FinOps Open Cost and Usage Specification. Open standard for cloud-cost data export; not natively shipped in v2.0.105 (§43.1). |
| GDPR | General Data Protection Regulation. EU privacy law; AgentWatch has DSR-workflow + tenant-deletion APIs (§20). |
| HA | High Availability. Multi-instance + Postgres replication + LB-fronted topology (§40.3). |
| HAProxy | Open-source TCP/HTTP load balancer; the reference LB for production AgentWatch deployments (§37, §39). |
| HIPAA | U.S. Health Insurance Portability and Accountability Act. Healthcare-data privacy framework; PHI detection inline + BAA-eligible deployment (§20). |
| HITRUST | Healthcare-industry security framework that maps multiple frameworks (HIPAA, NIST, ISO 27001) to one certification. Supported via the Compliance Framework Selector (§20). |
| HSM | Hardware Security Module. Tamper-resistant device for storing cryptographic keys. Out of AgentWatch scope (§43.2). |
| HTTP/3 | HTTP-over-QUIC (RFC 9114). On by default in the endpoint agent’s local TLS listener (§30). |
| IaC | Infrastructure as Code. Terraform / CloudFormation / Pulumi. AgentWatch ships sample Terraform modules; native provider on roadmap (§47.6). |
| IdP | Identity Provider. The system that authenticates users (Okta, Entra ID, Keycloak, Google Workspace, Auth0). |
| ISO 27001 | International information-security management standard. AgentWatch is ISO 27001-compatible via the Golden Security Configuration baseline (§20). |
| ISV | Independent Software Vendor. AgentWatch’s resellable SaaS topology supports ISVs offering AI gateway services downstream (§32). |
| JWT | JSON Web Token. The session-token format AgentWatch uses for stateless authentication (§16.1, §35). |
| Kafka | Apache distributed event-streaming platform. AgentWatch event pipeline (§22). |
| LanceDB | Embedded vector database; powers semantic cache + session embeddings (§7, §34). |
| LDAP / LDAPS | Lightweight Directory Access Protocol (and TLS-over-LDAP). Active Directory integration uses LDAPS (§16.5). |
| LLM-as-a-judge | Pattern of using an LLM to evaluate another LLM’s output (for jailbreak detection, quality scoring, etc.). Not natively in AgentWatch (§43.1); Datadog AI Guard implements it (§45.4). |
| LRU | Least-Recently-Used (cache eviction policy). Used by the Presidio NER sidecar (§18). |
| MCP | Model Context Protocol. Open standard for tool/resource provisioning to AI agents (§13, §14). |
| MITRE ATLAS | Adversarial Threat Landscape for Artificial-Intelligence Systems. MITRE’s adversarial-ML knowledge base; out-of-the-box testing not natively shipped (§43.1). |
| MSP | Managed Service Provider. AgentWatch’s multi-tenant SaaS topology supports MSPs offering AI gateway services (§32). |
| mTLS | Mutual TLS. Both client and server present certificates; AgentWatch uses it for fleet agent enrollment (§28, §35). |
| NDJSON | Newline-Delimited JSON. The streaming format for the AgentWatch Connect egress tunnel + flat-file audit archive (§22, §29). |
| NER | Named Entity Recognition. Identifying entities (people, places, organizations) in unstructured text. Presidio + spaCy power AgentWatch’s NER (§18). |
| NHI | Non-Human Identity. Workload / service / agent identity (as opposed to a human user). Governance not natively shipped (§43.1). |
| NIST | U.S. National Institute of Standards and Technology. NIST AI Risk Management Framework support via the Compliance Framework Selector (§20). |
| OAuth 2.0 / 2.1 | Authorization framework for delegated access. AgentWatch supports authorization-code + PKCE, refresh-token, ROP, and (via API keys) M2M (§16.3). |
| OIDC | OpenID Connect. Identity layer on top of OAuth 2.0; AgentWatch’s primary SSO protocol (§16.3). |
| OTel / OTLP | OpenTelemetry / OpenTelemetry Protocol. Vendor-neutral observability standard; AgentWatch’s default trace/metric/log export (§21). |
| OWASP LLM Top 10 | OWASP’s top-10 vulnerabilities for LLM applications. Out-of-the-box testing not shipped (§43.1). |
| PCI-DSS | Payment Card Industry Data Security Standard. Card-data redaction + tokenization integrations (§20). |
| PHI | Protected Health Information. HIPAA-regulated data; inline DLP category (§18). |
| PII | Personally Identifiable Information. SSN, driver’s license, passport, email, phone, etc. Inline DLP category (§18). |
| PgBouncer | Lightweight connection pooler for PostgreSQL; recommended for AgentWatch deployments beyond 1K users (§40.3). |
| PKCE | Proof Key for Code Exchange (RFC 7636). OAuth 2.0 extension protecting public clients from auth-code interception (§16.3). |
| POC | Proof of Concept. Demo-mode is designed to make POCs trivial — seeded data without real LLM keys (§15.2). |
| Presidio | Microsoft’s open-source PII detection + anonymization framework. The detection engine behind AgentWatch’s DLP (§18). |
| Prisma | The ORM AgentWatch uses for PostgreSQL + SQLite (§2.1). |
| QUIC | Transport protocol underneath HTTP/3. UDP-based with built-in TLS 1.3 (§30). |
| RAG | Retrieval-Augmented Generation. Pattern where the LLM is fed retrieved context (documents, knowledge base) before generation. AgentWatch hosts a RAG infrastructure (§9, §45). |
| RBAC | Role-Based Access Control. Granular per-resource permission model (§17). |
| RPO | Recovery Point Objective. Maximum acceptable data loss in a DR event. AgentWatch achieves RPO < 5 min via Postgres WAL streaming (§31, §40.3). |
| RTO | Recovery Time Objective. Maximum acceptable downtime in a DR event. AgentWatch achieves RTO < 30 min on Tier-2 deployments (§31, §40.3). |
| SAML | Security Assertion Markup Language v2.0. XML-based SSO protocol; native support in v2.0.40+ (§16.4). |
| SARIF | Static Analysis Results Interchange Format (OASIS standard 2.1.0). Security-scan output format produced by AgentWatch’s SAST/SCA pipeline (§11). |
| SAST | Static Application Security Testing. Code-pattern vulnerability scanning. Powered by Semgrep in AgentWatch (§11). |
| SCA | Software Composition Analysis. Dependency-vulnerability + license scanning. Powered by Trivy in AgentWatch (§11). |
| SCIM | System for Cross-domain Identity Management (RFC 7644). Standard for user/group provisioning from an IdP. Native support, per-org endpoints (§16.7). |
| SDS | Sensitive Data Scanner. Datadog’s PII-scrubbing feature; comparable to AgentWatch DLP but post-trace rather than inline (§45). |
| Semgrep | Open-source static-analysis engine used as the SAST scanner (§11). |
| SIEM | Security Information and Event Management. ELK / Splunk / QRadar / Sentinel; AgentWatch exports via Kafka topics + OTLP + webhooks (§22, §23). |
| SLA / SLO | Service Level Agreement / Service Level Objective. AgentWatch is software, not a hosted SLA; HA is the customer’s deployment contract (§43.1). |
| SOC 2 | AICPA Service Organization Control 2 (Type II). Audit framework for service-organization security; supported via the Compliance Framework Selector (§20). |
| SOX | Sarbanes-Oxley Act. U.S. financial-controls regulation; immutable audit log + RBAC + segregation of duties (§20). |
| SPIFFE / SPIRE | Secure Production Identity Framework / Reference implementation. Hardware-attestation-based workload identity; not natively shipped (§43.1). |
| SSE | Server-Sent Events. Streaming HTTP response format; used for LLM streaming completions + MCP transport (§3.3, §14). |
| SSO | Single Sign-On. AgentWatch supports OIDC, SAML, and LDAP (§16). |
| TEE | Trusted Execution Environment. Hardware-isolated compute (Intel TDX, NVIDIA CC, AMD SEV). Out of AgentWatch scope (§43.2). |
| TLS | Transport Layer Security. Standard encryption-in-transit; AgentWatch uses 1.2 + 1.3 (§35, §44.3). |
| TPM | Trusted Platform Module. Hardware key storage chip. Out of AgentWatch scope (§43.2). |
| Trivy | Open-source vulnerability scanner used as the SCA + secrets + IaC misconfiguration scanner (§11). |
| TTFT | Time To First Token. Latency from request submission to the first streamed token; tracked per request (§3.3, §21). |
| Trojan Source / Unicode attacks | Invisible-text and bidi-override attacks detected by LLM Guard (§19). |
| vLLM / TGI / Triton / Ollama | Open-source LLM inference servers. AgentWatch routes to them as Custom or Ollama providers (§4.1). |
| WAF | Web Application Firewall. AgentWatch ships an AI-aware WAF (token stuffing, prompt flooding, abnormal-session blocking) on top of the OWASP Top 10 protections (§24). |
| WebSocket | Bidirectional persistent HTTP connection; used for MCP transport + endpoint agent control channel (§14, §29, §30). |
| WAL | Write-Ahead Log. Postgres mechanism for crash recovery + streaming replication; the basis for AgentWatch’s < 5 min RPO (§40.3). |