SentriAI Feature

Wire Governance Into Agents Five Ways

A single decision contract, reached however your stack is built - as an API gateway, MCP middleware, a LangChain hook, a sidecar proxy, or an in-process SDK. Call decide to get a verdict, or dry-run to preview one.

Talk to usFree to start - no card required.
SentriAI · Runtime Enforcement
ONE CONTRACT
/enforce/decide
+ /dry-run
MODES
5 ways in
same shape
LATENCY
fast path
deterministic
API gatewayfront the agent’s calls
MCP middlewaregovern tool traffic
Framework hookLangChain / SDK
Sidecar proxyno code change
In-process SDKlowest latency

Illustrative product view

One decision, many deployment shapes

The hardest part of governing AI actions is not the policy - it is getting in the path of the action without rewriting the app. Runtime enforcement exposes one decision endpoint that five integration shapes all call with the same request, so you pick the one that fits your architecture and get identical verdicts everywhere.

Where it sits

  • API gateway - front an agent’s outbound calls with a verdict check
  • MCP middleware - govern Model-Context-Protocol tool calls at the boundary
  • Framework hook - drop into a LangChain or SDK agent loop
  • Sidecar proxy - enforce with no application code change
  • In-process SDK - lowest-latency checks inside your service

Frequently asked questions

What is runtime enforcement?

It is the customer-facing decision endpoint that agents call to get an authorization verdict before acting, reachable through five integration modes that all share one request shape.

What are the five ways to integrate?

An API gateway, MCP middleware, a framework hook such as LangChain, a sidecar proxy, and an in-process SDK - each calls the same decide or dry-run contract.

What is the difference between decide and dry-run?

Decide returns and records a verdict to act on; dry-run previews the verdict without committing to enforcement, so you can validate policy before turning control on.

Does it govern MCP tool calls?

Yes. MCP middleware is one of the five modes, putting a verdict in front of the fastest-growing agent action surface - tool calls.

Stay in the loop

One practical finance briefing a week - new guides, checklists, and benchmarks.

 

Put a decision - with proof - in front of every action

See how Fintra decides allow / step-up / hold / block per action and writes each verdict to a tamper-evident ledger.

Talk to us