Integrators
ThunderCat Technology
A faster, trusted path to data-centric security for DoD missions. ThunderCat’s contract vehicles get the platform to government teams sooner.
One gateway for AI traffic.
One policy for the data behind it.
Kong routes and secures every LLM, MCP and agent request. Data Harbor Platform makes sure each answer only includes data the requester is allowed to see.
How Kong AI Gateway and Data Harbor Platform work together to govern AI traffic and the data behind it.
Kong AI Gateway is one control plane for all LLM, MCP and agent-to-agent traffic. Teams set a single endpoint to route and balance across providers like OpenAI, Anthropic, Amazon Bedrock and Google Gemini, with built-in authentication, policy enforcement and live analytics for tokens, latency and errors.
How it works together
An agent’s calls route through Kong to fetch enterprise data, such as objects in Amazon S3. Before any file reaches the agent, attribute-based access control checks the request in real time against the policy embedded in the object. Identity and attributes come straight from your identity provider, and only authorized records come back.
Kong governs tool permissions and response content. Virtru governs the file underneath. Control travels with the data, whatever the tool call pipeline.
Three integration points
Kong clears the agent to call a tool under a given OAuth2 scope. Data Harbor Platform decides what that tool can hand back, object by object, so exposing more tools never quietly exposes more records.
The same OIDC or OAuth2 attributes Kong uses to authenticate and route a request also drive the policy decision at the data. Change a role once in your identity provider and both layers follow, with no second policy to keep in sync.
Kong’s data planes and Data Harbor Platform can both run in your own environment. The gateway, the tool calls and the access decision all sit inside one accreditation boundary, with no vendor-hosted control plane in the loop.
What you get when Kong AI Gateway and Data Harbor Platform run side by side.
Each retrieval through a Kong-exposed tool is checked against the policy in the object, so the agent only receives the records the requester’s current attributes allow.
The OIDC or OAuth2 attributes that route traffic through Kong also drive access decisions at the data, so clearance and role changes apply everywhere without duplicate setup.
Kong AI Gateway runs fully self-hosted and Data Harbor Platform runs self-managed, with no vendor-hosted control plane in the data path.
Every new tool Kong exposes inherits the same central data policy automatically, instead of needing its own access rules.
Protection is enforced at the data, not inside one LLM integration, so it holds as Kong routes across OpenAI, Anthropic, Amazon Bedrock, Google Gemini or any other provider.
Alongside Kong’s analytics for requests, tokens, errors and latency, every grant or denial is recorded. You can show the disclosure was authorized, not just that the call was logged.
Kong’s gateway, AI policies and provider routing stay as they are. Data Harbor Platform’s Storage capability and Key Access Service enforce policy at the data layer behind it.
Running Data Harbor Platform behind Kong AI Gateway needs active deployments of both platforms, plus an identity provider to supply the attributes that drive access decisions.
Konnect-managed or fully self-hosted, fronting LLM, MCP and tool-call traffic through the AI MCP Server entity.
In your VPC or on premises. The Kong integration needs the self-managed deployment, not the Virtru-hosted SaaS.
Amazon S3 or other object storage, fronted by the platform’s Storage capability and Key Access Service.
OAuth2 scopes on the exposed MCP tools, such as read or write access.
Microsoft Entra ID in the reference architecture, or any OIDC or OAuth2 system that supplies the attributes in your policy.
Attributes and rules you write, checked at every data retrieval a tool call triggers.
Between Kong’s data plane, the agent’s tool-call path and the Virtru Key Access Service.
Recommended for regulated environments.
How the two platforms split the work across traffic, compliance and privacy.
Kong secures and governs the traffic: authentication, OAuth2-scoped tool calls through AI Auth Strategies, rate limiting and observability across every LLM, MCP and agent-to-agent connection. Its prompt and response guardrails (AI Prompt Guard, AI Semantic Prompt Guard, AI Sanitizer and AI Semantic Response Guard) screen content for safety and PII leaks.
Data Harbor Platform secures the data itself, with object-level TDF encryption and ABAC that holds whichever tool, model or agent makes the request. Kong never sees protected plaintext, only the traffic metadata and prompt and response content it needs to route and govern.
Compliance
Kong AI Gateway offers a FIPS 140-3 compliant mode on its data planes, using the OpenSSL 3.1.2 FIPS provider set in kong.conf.
Data Harbor Platform is FedRAMP Moderate authorized, built on the OpenTDF open standard, and uses a FIPS 140-3 validated cryptographic module (VirtruCrypto, certificate #4440).
Privacy
The Key Access Service releases keys only when the requester’s attributes meet the policy embedded in the object. Kong authenticates, routes and screens the request. It doesn’t decide what the data response contains.
Integrators
A faster, trusted path to data-centric security for DoD missions. ThunderCat’s contract vehicles get the platform to government teams sooner.
Collaboration
Continuous, attribute-based access control for WickrGov rooms. Virtru checks who can join each room, and keeps checking for as long as they stay.
Collaboration
Channel access that keeps up with your team. Virtru checks members against your identity provider, and revokes access once they no longer qualify.