Secure, Compliant, and Observable AI Gateway for Enterprise
Edition: v5 | Server version: 2.0.72 | Client agent version: 1.1.0 | Last revised: August 5, 2026
Changes since Enterprise Guide v4: introduces the Observe Platform — a themeable, unified observability workspace embedded natively in the admin console (AI, Services, Fleet, Optimize, Business, Security, Insider Risk, Delivery, Logs); AgentFleet, a deployable fleet agent with a Kubernetes-native
sidecar/DaemonSet mode, Helm chart, multi-arch image, and one-line VM installer; AgentKeel, embedded per-agent response caching and prompt/token compression with direct-vs-optimized telemetry; and Insider Risk (AI Agent) — governance of the non-human actor surface with Agent Risk Indicators, custom indicator authoring, AI incident case management, and inline adaptive enforcement in the gateway. A per-user Demo Mode populates every new surface with realistic data for evaluation. See new Sections 4.11–4.15 (Core Capabilities) and 7.19–7.23 (Feature Deep Dive). None of these features requires a schema migration — upgrading is a plain image swap (down → pull → up -d).
Changes since Enterprise Guide v3: introduces the on-disk encrypted request body archive (full payloads moved out of SQLite); 90-day automated retention sweep for archived bodies; superadmin diagnostics endpoints (
/admin/diagnostics/memory,/admin/diagnostics/heap-snapshot); native-memory hardening (LanceDB handle pooling, Python bridge response matching, rate-limit map eviction); foreign-key based model routing with automatic catalog reconciliation on container boot. See Section 7.16 for the storage and reliability changes; see the Operations chapter for memory-pressure diagnosis and the upgrade playbook.
Executive Summary
Section titled “Executive Summary”AgentWatch is an enterprise-grade AI gateway that sits between your organization’s AI agents/applications and Large Language Model (LLM) providers like OpenAI, Anthropic, Google, and others. It provides comprehensive security, compliance, cost management, and observability capabilities that enterprises need when deploying AI at scale.
The v5 release advances AgentWatch from a governed gateway into a full AI operations platform. The new Observe Platform brings model, cost, latency, prompt-lineage, service-trace, fleet-health, spend, security, and delivery telemetry into a single themeable workspace inside the admin console — no external observability tooling to license or wire up. AgentFleet extends attribution and control from browser and API traffic to your production estate, with a fleet agent that deploys as a Kubernetes sidecar or per-node DaemonSet, a Linux VM installer, and a Helm chart. AgentKeel lets each deployed agent transparently cache and compress LLM traffic — serving hits without ever touching the provider — and reports the savings back to an Optimize surface. And Insider Risk (AI Agent) governs the actor that traditional human-DLP and endpoint tools cannot see — the AI agent itself — scoring every tool call, API invocation, and model output against a library of Agent Risk Indicators, with inline adaptive enforcement that can step-up or block a risky agent before its request ever reaches a provider. Every new surface can be populated with realistic Demo Mode data for evaluation, and the entire wave installs as a zero-migration image swap.
Table of Contents
Section titled “Table of Contents”- The Enterprise AI Challenge
- Without AgentWatch: Current State Analysis
- With AgentWatch: Transformed Enterprise AI
- Core Capabilities
- Use Cases by Industry
- Architecture Overview
- Feature Deep Dive
- 7.2 Budget Management
- 7.4 Audit & Compliance Reports
- 7.5 Gateway Security (WAF)
- 7.6 Single Sign-On (SSO)
- 7.7 JSON Schema Validation
- 7.8 Gateway Plugin System
- 7.12 Tenant Routing Rules
- 7.13 AI Assistant Settings — System LLM
- 7.14 Gateway Routes
- 7.15 Semantic Caching (LanceDB)
- 7.19 Observe Platform — Unified Observability Workspace
- 7.20 AgentFleet — Deployable Fleet Agent
- 7.21 AgentKeel — Embedded Optimization
- 7.22 Insider Risk (AI Agent Governance)
- 7.23 Demo Mode & Zero-Migration Upgradeability
- AgentWatch Connect Agent
- MCP Proxy Access Control
- ROI Analysis
- Getting Started
- UI Navigation Guide
- 12.1 Analytics — Observability
- 12.2 Analytics — Usage Insights
- 12.3 Analytics — System Health
- 12.4 Management — Team Management
- 12.5 Management — Employee Management
- 12.6 Management — Budget Management
- 12.7 Management — Company Settings
- 12.8 Management — API Key Management
- 12.9 Management — Model Catalog
- 12.10 Management — AI Insights
- 12.11 Management — MCP Proxy
- 12.12 Management — Audit Export
- 12.13 Settings — General
- 12.14 Settings — Security
- 12.15 Settings — Features
- 12.16 Settings — System
- 12.17 Settings — AI Assistant
- 12.18 Settings — Gateway Routes
- 12.19 Settings — Gateway Analytics
- 12.20 Settings — Tenant Routing Rules
- 12.21 Settings — Schema Validation
- 12.22 Settings — Plugins
- 12.23 Settings — WAF Protection
- 12.24 Settings — OAuth2 / SSO
- 12.25 Settings — API Docs
- 12.26 Platform Admin — Tenants
- 12.27 Platform Admin — Plans
- 12.28 Platform Admin — Audit Log
- 12.29 User Dashboard — Overview
- 12.30 User Dashboard — My API Keys
- 12.31 User Dashboard — My Usage
- 12.32 User Dashboard — My Budget
- 12.33 User Dashboard — My Settings
1. The Enterprise AI Challenge
Section titled “1. The Enterprise AI Challenge”As enterprises rapidly adopt AI and LLM-powered agents, they face critical challenges:
| Challenge | Business Impact |
|---|---|
| Data Leakage Risk | Sensitive PII, PHI, financial data sent to external AI providers |
| Compliance Violations | GDPR, HIPAA, SOX, PCI-DSS requirements not met |
| Cost Overruns | Uncontrolled API usage leading to unexpected bills |
| Lack of Visibility | No insight into what AI agents are doing |
| Security Gaps | Prompt injection, data exfiltration risks |
| Audit Failures | Unable to demonstrate AI governance to regulators |