Copilots and agents don't get model access with an API key. The EmpowerID LLM Gateway checks who is asking, under whose authority, for what intent, within what budget — before the provider is called — and signs the evidence.
OpenAI- and Anthropic-compatible routes · Multi-provider control · AuthZEN-compatible policy
Governed request trace showing an allowed path before provider consumption.
Teams can connect a copilot or agent to a model in hours. Production governance is harder. Who is making the request? Is an agent acting for a person — and is that delegation still active? Does the prompt indicate sensitive intent? Is the model approved for this subject? Has the budget been reached? Which policy allowed the call?
Generic API keys, per-app guardrails, and after-the-fact dashboards answer fragments. The EmpowerID LLM Gateway brings them into one runtime authorization decision at the model boundary.
Evaluate model access for people, applications, workloads, and AI agents using Identity Fabric sessions or personal access tokens. For agents, policy considers the identity chain and active delegated authority—not possession of a shared model key.
Only allowlisted, high-confidence labels cross the policy export boundary. The classifier supplies context; the PDP makes the decision. Safety guards configured as mandatory fail closed when classification is unavailable.
Evaluate current AI budget state before the upstream model request. Policy can deny a call, constrain requested tokens, or select a lower-cost allowed model before provider charges are incurred.
Applications authenticate to the Identity Fabric. Organization-managed provider credentials are resolved at the gateway and injected only on the upstream provider connection.
For completed allowed calls, create signed, hash-linked receipts that bind policy context to measured usage. Denied calls generate decision and operational telemetry—not signed receipts.
Multi-provider routing, circuit breaking, and failover — with alternatives constrained by model policy, so an availability event never becomes an authorization exception.
Every governed request runs the same deterministic sequence — even when the policy itself is context-sensitive.
Authorization is not a wrapper around inference. It is the decision that determines whether inference should occur.
Authenticate the subject
Resolve the person, application, workload, or agent — and its identity chain.
Classify the request
Derive selected intent, safety, and data context from the prompt.
Ask for an authorization decision
Send subject, model, action, classification, delegation status, estimated cost, and context to the policy decision point.
Enforce constraints before inference
Allow or deny, clamp maximum tokens, or route to another permitted model.
Call the provider securely
Resolve the organization-managed credential and relay the compatible request.
Settle measured usage
Capture available provider usage and update the governed spend path.
Create evidence
Emit a signed receipt for the completed allowed call and send it to analytics.
Most AI gateways begin with connectivity: one endpoint, multiple providers, traffic logs, retries, rate limits. Useful — and incomplete. Production agents create an authorization problem. The EmpowerID LLM Gateway connects model traffic to the Identity Fabric's shared truth about identity, delegated authority, policy, and evidence.
| Conventional gateway question | Identity Fabric question |
|---|---|
| Is this API key valid? | Which governed subject is acting, and for whom? |
| Is the route allowed? | May this subject use this model for this intent now? |
| Has the request rate been exceeded? | Is current spend within the policy boundary for this user, agent, or application? |
| Did the request succeed? | Which decision authorized it, what constraints applied, and what usage followed? |
AI agents cross two different boundaries. Both enforcement points use the Identity Fabric's shared policy and identity context — without pretending that generating language and executing a consequential tool are the same kind of event.
LLM Gateway · model boundary
MCP Gateway · tool boundary
The LLM Gateway exposes provider-compatible routes for common OpenAI and Anthropic request patterns. Applications authenticate through the Identity Fabric and point their SDK at the governed endpoint—the request body can remain unchanged while authorization, credentials, provider selection, budget policy, and evidence move into a shared control layer.
View integration documentation →Governance is defined by failure behavior as much as by the happy path. The gateway avoids silently bypassing policy when the decision service is unavailable; budget enforcement can fail closed when authoritative spend state cannot be established for a limited subject; circuit breakers and failover support resilience without turning an outage into an authorization exception. Configuration and last-known-good behavior are selected to match the organization's risk and availability requirements.
Give internal copilots controlled access to approved providers and models without distributing provider keys to each application or browser client.
Require an active identity chain and delegated authority before an agent invokes a model, then attribute usage to the governed subject.
Use high-confidence prompt labels as policy inputs to distinguish a permitted general business question from a restricted request about an individual or protected data.
Evaluate current budget state before provider consumption, constrain requested tokens, or route to a lower-cost allowed model under policy.
Bind completed allowed model use to signed receipts and analytics so investigators can reconstruct the decision, constraints, effective model, and usage path.
"Authorization Before Inference" — the full architecture, lifecycle, and evaluation checklist.
Download →Feature availability—including prompt classification depth, budget enforcement modes, model switching paths, provider routes, receipt coverage, circuit breaker behavior, and edition packaging—varies by release and deployment. Confirm scope with EmpowerID before customer-specific commitments.
Watch a governed subject blocked by policy before the provider is called—sensitive intent flagged, a budget stop enforced pre-request, and the signed receipt trail for the calls that were allowed.
Online
Powered by EmpowerID AI
