# Tell your board what AI did.

Your teams and their agents already send company data to AI providers.
The board now asks what left, who approved it, and what it cost.
The Gateway decides every request before it leaves and records each decision with the rule behind it.
So you answer the board with a query against the record, not a round of interviews.

[Join the waitlist](/partners) [See pricing](/pricing)

```record One quarter, as the record summarizes it. Illustrative values.
period      Q3, one organization
requests    1,284,906 decided and recorded
screened    18,330 regulated values redacted before leaving
refused     412 requests, each with the rule that refused it
agents      9 coding agents, every tool call judged
savings     accepted on 7 routes, reconciled to the invoice
policy      v42 through v51, every version replayable
```

## What changes for your board

- **The governance answer stops being a project.**
  Of 950 US executives Grant Thornton surveyed in early 2026,
  78% lacked strong confidence they could pass an independent AI governance audit within 90 days.
  Every request through the Gateway is recorded with its policy version,
  so the audit question becomes a query your own team runs.
  [Grant Thornton, 2026 AI Impact Survey](https://www.grantthornton.com/services/advisory-services/artificial-intelligence/2026-ai-impact-survey)
- **Unapproved AI is what makes a breach expensive.**
  In IBM's 2025 breach study, one in five organizations had a breach tied to shadow AI.
  Where unapproved AI use was high, a breach cost about $670,000 more.
  The Gateway cannot see someone using a chatbot in a browser tab.
  What it can do is make the approved path the easy one, so less work goes around it.
  [IBM Cost of a Data Breach 2025, via Cybersecurity Dive](https://www.cybersecuritydive.com/news/artificial-intelligence-security-shadow-ai-ibm-report/754009/)
- **Teams get access by signing in, not by asking.**
  Map your identity groups to workspaces, and each person creates their own key inside the budget you set.
  A person who leaves the group loses their key with it.
  No one files an access request, and no provider key leaves your administrators' hands.
- **Your edge is the decision, not the plumbing.**
  Several products already forward traffic to many models, and do it well.
  The open problem is deciding whether a request should go out, what it should cost, and what proof exists afterward.
  That is the part your board is asking about.

A board does not want a vendor's assurance.
It wants the same answer twice, from the record, months apart.

## Answers you can check before you buy

Each row says where to check the answer, so your own team can verify it instead of taking our word.

| The board asks | The answer | Where to verify |
| --- | --- | --- |
| Will this slow our teams down? | The Gateway's target for added time is 25 ms at p50 and 60 ms at p95, measured at its own boundary with screening on. | [Architecture review](/architecture-review) |
| What is the payback in plain terms? | A printed monthly fee, plus a share of savings verified against your own provider invoice. In a month with no verified saving, the share is zero. | [Pricing](/pricing) |
| Who is accountable when something goes wrong? | A published split for each way the Gateway is deployed. In the hosted form, we own application availability, patching, and incident notification, and you own your policy, credentials, and acceptable use. | [Security review](/security-review) |
| Are we locked in? | No. Your clients keep speaking their own API to the Gateway, so leaving means changing the base URL back. | [One door](/one-door) |
| Does one vendor now hold everything? | Putting policy, credentials, and the record in one place also makes that place a bigger target. We say so openly, and our controls limit what a compromise can reach. | [Security review](/security-review) |
| Is this proven, or are we first? | One real deployment has measured how the mechanism works, but one deployment cannot give you a typical result. Nobody can quote you an expected saving yet, including us. | [Provable economics](/provable-economics) |

## Send this to the people who will ask

- [Security review](/security-review)
  For your security lead.
  Where data lives, who can decrypt your provider keys, and what we refuse to claim.
- [Architecture review](/architecture-review)
  For your engineering lead.
  The request pipeline, the latency target, streaming failover, and the deployment topologies.
- [Finance review](/finance-review)
  For your finance partner.
  What you pay, how spend reconciles to the provider invoice, and what leaving costs.

## When this is not your problem yet

- **One team, one provider, low volume, no regulated data.** Call the provider's SDK directly and revisit later.
- **Nobody has asked you for evidence yet.** Fear of compliance alone is a poor reason to buy.
  Buy when a board, an auditor, or a customer asks a question your teams cannot answer.
- **All your traffic stays in one cloud, and that cloud's controls meet your policy.** Keep them.
  Come back when a second provider joins the stack.

## Start with one real workload.

We admit customers in cohorts, in waitlist order. Don't miss the next one.
When your slot comes up, you run one real workload on the plan you chose,
and you keep a governance and savings report written for the board that asked.
If we stop earning our fee, leaving is the same URL change, reversed.

[Join the waitlist](/partners)