Per-call policy checks
Policy as code evaluates every request before it reaches a model.
AI
AI Gateway
One policy-checked path for every model call
Every model call routes through one gateway where policy is checked on each request, so you govern AI usage in one place instead of per app. Any state-changing action pauses for a human to approve before it runs, and each connector the assistant uses receives scoped, short-lived credentials. Usage is recorded, and the gateway speaks a standard chat-completions interface so your existing clients connect without rewrites.
- One governed path for every model call across the organization
- A recorded reason behind every request, policy decision, and approval
- Narrow, expiring credentials on every connector the assistant touches
The problem
Every team that reaches for an AI model creates its own path to that model, with its own rules and its own credential handling. You have no single place to see what is being called, no policy that runs on every request, and no guarantee that a state-changing action waited for a person before it ran.
Without IntegraCI
- Each app calls models on its own ungoverned path
- No policy evaluating what requests are permitted
- Credentials shared broadly, not scoped to each connector
- No record of what was called, approved, or denied
With IntegraCI
- One endpoint that every model call routes through
- Policy as code evaluates each request before it reaches a model
- State-changing actions pause until a person approves
- Each connector receives narrow, expiring credentials from a dedicated secrets store
What you get
Human approval gate
Any state-changing action pauses for a person to approve before it proceeds.
Scoped short-lived credentials
Each connector the assistant touches gets narrow, expiring credentials from a dedicated secrets store.
Standard chat interface
The gateway speaks a standard chat-completions interface so existing clients connect unchanged.
How it works
- 1
Route the call
Your client sends model requests through the single gateway endpoint.
- 2
Check and approve
Policy runs on the call and state-changing actions wait for human approval.
- 3
Scope and record
Connectors receive short-lived credentials and the usage is logged for review.
How it stays governed
The same gates everyone passes, applied here.
Gated by policy
Every model request passes through policy as code before it reaches a model. The rule set applies to every client on the same path, so there is no ungoverned route to a model and no team that bypasses the check.
Recorded, tamper-evident
Each request, policy decision, and usage record writes once to a tamper-evident audit trail. You can show exactly what was called, what policy ruled, and who approved any state-changing action.
A human in the loop
Any action the assistant proposes that would change state pauses at the gateway until a person reviews and approves it. The call does not proceed until sign-off is recorded, so no state-changing action runs without a human in the loop.
Works with your stack
Connect the tools you already run.
Any client speaking the standard chat-completions interface connects to the gateway without changes, and scoped credentials are issued per connector through a dedicated secrets store.
- Aqua Security
- DefectDojo
- Elastic
- Google Cloud
- Greenbone
- HashiCorp
- IBM QRadar
- Isovalent / Cilium
- Mend
- Microsoft Azure
- Open Policy Agent / CNCF
- OpenBao
- OWASP ZAP
- PlexTrac
- ProjectDiscovery
- Prowler
- ScanCode
- Snyk
- +39 more
Who it’s for
Where teams reach for it.
Standardize model access across teams
When different teams reach for different models through different apps, you route every call through one endpoint. Policy runs the same way on each request so you govern the whole organization from one place, not per app.
Govern agentic workflows in regulated environments
When an AI assistant takes actions on your systems, every state-changing step pauses for a person to approve before it runs. The full chain of calls, decisions, and approvals lands in a tamper-evident audit trail.
Add governance without rewriting clients
Teams with existing chat-completions integrations point them at the gateway endpoint and gain per-call policy checks, human approval gates, and usage records without changing the client code.
Questions, answered.
Does this replace our LLM provider?
No. The gateway routes calls to the model providers you already use. Your provider relationship and model choices stay in place; the gateway adds policy checks, approval gates, and usage records on top.
Which clients are compatible?
Any client that speaks the standard chat-completions interface connects without modification. You do not need to rewrite existing integrations to route calls through the gateway.
Can we write our own policy rules?
Yes. Governance runs as policy as code, which your team authors and version-controls. You define what requests are permitted, what triggers a human approval, and what is denied outright.
How are state-changing actions handled?
The gateway holds any state-changing call until a person approves it. The action does not proceed until sign-off is recorded, and the decision writes to the tamper-evident audit trail alongside the original request.
Put AI Gateway on your stack.
Request a demo, or read the docs to see how it fits the tools you already run.