Skip to main content

Sentinel

Sentinel is Caldorus’s hosted AI gateway for governed model access. It gives teams one stable access layer for model traffic so they can standardize routing, policy enforcement, spend controls, and auditability without rebuilding those controls in every application.

Why teams adopt it​

Direct provider integrations work at first, but they become harder to manage as more teams, applications, and models get involved.

Teams adopt Sentinel when they need to:

  • enforce consistent policy across multiple applications
  • control spend without slowing down delivery teams
  • support more than one provider or model lane
  • expose one stable integration surface to application teams
  • retain durable evidence of access decisions and request behavior

What Sentinel does​

Sentinel gives platform teams a governed access layer for model traffic.

  • provides an OpenAI-compatible lane for broad client compatibility
  • supports native Anthropic and Google lanes where provider-specific behavior matters
  • centralizes routing and endpoint capability checks
  • enforces policy, rate limits, endpoint restrictions, and budget controls
  • records operational telemetry and audit evidence around requests and decisions
  • provides operators one hosted control surface for provider setup, keys, routing, and request visibility

When Sentinel is the right fit​

Sentinel is a strong fit when:

  • multiple applications need model access under one governance standard
  • platform teams want provider flexibility without forcing client rewrites
  • engineering and security teams need traceability on access decisions
  • budget and endpoint controls must be enforced centrally
  • operators need better visibility into model access behavior across teams

What changes with Sentinel​

Without Sentinel, each application tends to carry its own provider logic, credential handling, guardrails, and observability patterns.

With Sentinel:

  • applications call one gateway instead of managing provider logic independently
  • operators manage providers, keys, routing, and policy in one place
  • governance decisions happen before provider execution
  • request visibility and audit evidence live at the model-access layer
  • provider choice can evolve without forcing every client to change with it

Where Sentinel sits​

Sentinel sits between your applications and model providers.

It provides:

  • a hosted gateway for inference traffic
  • an operator control surface for provider setup, keys, routing, policy, and request review
  • one consistent access layer where governance state stays outside application clients

In practice, this means application teams integrate once, while platform teams control how model access is routed, governed, and observed.

What Sentinel does not replace​

Sentinel does not replace your application’s business logic, product decisions, or provider account governance.

Instead, it adds a consistent control layer for:

  • model access behavior
  • operator controls
  • policy enforcement
  • request visibility
  • audit evidence

Start with these pages​

  • Quickstart for the fastest first request through the hosted gateway
  • Architecture to understand request flow and trust boundaries
  • SDKs for client integration guidance
  • Providers for provider setup and lane selection