AI Warden platform
One control point for LLMs, MCP tools, enterprise agents, DLP and AI governance.
AI Warden combines an OpenAI-compatible LLM Gateway, MCP Gateway and Fleet, ICAP-supported DLP enforcement, no-code Agent Builder, governance controls and AI FinOps into a single platform. Security teams get policy and evidence. AI platform teams get a practical gateway developers can adopt.
Platform architecture
A control plane for policy. A data plane for enforcement.
The portal owns governance, policy authoring, approvals, inventory and analytics. The gateway sits in the live request path for LLM and MCP traffic, enforcing identity, budgets, scanners, ICAP decisions and audit logging before anything leaves or returns.
Designed for regulated AI adoption.
Most AI gateway products start with provider routing and observability. AI Warden includes those primitives, but puts enterprise security and governance at the centre: request and response enforcement, ICAP-supported DLP, MCP tool governance, approvals, product controls and evidence.
- Portal control plane: policy, RBAC, products, controls, approvals, inventory, analytics and request-log review.
- Gateway data plane: OpenAI-compatible LLM proxying, MCP proxying, scanner enforcement, enterprise DLP inspection and audit writes.
- Evidence stores: encrypted state plus a customer-controlled log store for high-volume LLM and MCP request evidence.
- Identity boundary: user, service principal, hosted agent and delegated-user attribution stay visible across the request path.
employee / app / agent | v AI Warden Gateway identity -> budget -> policy -> ICAP -> scanner -> route | | v v LLM providers MCP tools | | +------------- response enforcement --+ | v Portal control plane: policies · approvals · products · controls · evidence
Product components
Start with the outcome. Drill into the component.
Security, FinOps, governance and enablement are connected outcomes. These product components are the technical surfaces that implement them.
01 LLM Gateway
Drop-in OpenAI-compatible gateway for secure model access.
Give developers a familiar endpoint while platform teams control provider routing, model aliases, budgets, compression, prompt and response policy, DLP inspection, request logging and cost attribution.
- OpenAI-compatible API and model aliases.
- Token compression, caching, budgets and showback.
- Prompt/response scanning, ICAP and audit evidence.
02 MCP Gateway & Fleet
Govern the tools agents can call, not just the model.
Register MCP servers, scan capabilities, approve tool exposure, enforce per-server policy, inspect JSON-RPC requests and responses, and retain evidence for every tool call.
- Register, scan, approve and publish MCP servers.
- Per-user, per-agent and per-server policy controls.
- Request-log evidence for local blocks and upstream calls.
03 ICAP / DLP Integration
Bring existing enterprise DLP into the AI path.
Use ICAP-supported DLP platforms to inspect prompts, uploads, tool arguments, tool results and model responses. Allow, modify or block decisions can trigger AI Warden session and tool controls.
- Request and response inspection coverage.
- Configurable fail-closed handling when inspection is unavailable.
- Behaviour-based actions for repeated DLP events.
04 Agent Builder
No-code agents built by business teams, governed by the platform.
Departments can create focused agents with instructions, knowledge, approved tools and launch workflows. AI Warden adds approval rules, budgets, DLP, tool containment and request-log evidence.
- Guided builder from business problem to agent.
- Approved tools, MCP capabilities and workflow handoffs.
- Focused chat sessions with the named agent identity.
05 Governance Control Plane
Inventory, products, controls, approvals and evidence.
Map AI assets to business products, apply controls from standards or regulators, bind controls to live policies, set review periods and retain evidence snapshots for assurance review.
- Product ownership and labels such as gdpr=true.
- Controls mapped to policies, predicates and review cadence.
- Four-eyes approvals for material enforcement changes.
06 Deployment and operations
Self-hosted, managed or customer-controlled operation.
Run AI Warden where the customer needs it: inside regulated infrastructure or as a managed service with clear boundaries between platform administration and system operations.
- Customer-owned policy, users, routes, agents and reports.
- Managed install, upgrades, monitoring and runtime support.
- OpenAPI-described administration and auditable changes.
Two buyer paths
Security and platform teams can adopt the same control point for different reasons.
Reduce AI data leakage and tool risk.
Inspect the approved path, bring DLP into LLM and MCP traffic, configure fail-closed handling for inspection gaps, contain risky sessions, and preserve reviewable evidence.
- Policy control matrix and runtime evidence.
- ICAP request/response enforcement.
- SOC alerts, session containment and tool disabling.
Give developers and departments a usable AI platform.
Keep familiar SDKs, route providers centrally, observe cost and latency, publish approved MCP tools, and let departments build agents without bypassing governance.
- OpenAI-compatible quickstart.
- MCP adoption journey for developers and security.
- Agent Builder and enterprise chat for shadow-AI replacement.
Platform proof
OpenAI-compatible for adoption. ICAP and MCP-aware for enterprise control.
AI Warden is built to be familiar enough for developers, controlled enough for cybersecurity, and evidence-rich enough for governance review: one gateway path for LLM requests, MCP tool calls, DLP outcomes, cost controls and approval history.