Book a demo

ControlThe harness

AI you can account for

Say yes to AI across the business without losing track of it, or getting locked into one provider. Decide which models people use and what they may spend, stop sensitive data from leaving, and answer any question from the board or the auditor with a record of every request.

Providers and policy

Choose your AI, and change it without disruption

Bring your own accounts with Anthropic, OpenAI, Google or xAI, or run models on your own hardware. Policy decides which models each application and group may use. Switch providers in one place, and nothing else has to change.

  • Policy is published per application, and each request keeps the version it ran under
  • Each request's routing explains why its model was chosen
  • Keys stay on your server, never in people's tools

Budgets

AI spend that never surprises you

Give everyone a monthly allowance, groups their own, and applications a ceiling. The gateway reserves against the allowance before every request, so the month's total cannot run away, and people can ask for more with a reason.

  • One allowance follows a person across chat, tasks, flows and other AI tools
  • See who is nearly used up before they hit the limit
  • Requests for more budget, with their reason, in one list

Inspection

Sensitive data stays in, by default

Every request is checked before it reaches a provider for card numbers, keys and tokens, connection strings, South African ID and tax numbers, and the words you ask to keep private. Watch first to see how many people a rule would stop, then redact or block with confidence.

  • Each detector can be off, record, redact or block
  • Counts of requests and matches before you enforce anything
  • Applies to chat, tasks, flows, WhatsApp and connected tools alike

AI requests

Every request on record, never its content

One row per request: who, from where, which model, the outcome, the tokens and the cost, and how long it took. Prove what AI is used for, and what it costs, from a log that never holds what people wrote.

  • Refused requests show the reason, such as a budget or a detector
  • Open a row to see why its model was chosen over the others
  • Stream it, with the audit log, to your own SIEM

MCP servers and tools

Outside tools, reviewed before anyone uses them

Connect the tool servers your people's AI may use, and approve each tool by what the model is told about it. If a tool changes after approval, it waits for a new review, with the old and new side by side. Every call is governed and logged.

  • One endpoint for every approved tool
  • Each person's own credential, or a shared one stored safely
  • Requests, allowed and refused counts for every tool

AI register and audit

Ready when the auditor asks

A register of every AI system in use, with its owner, purpose, risk tier and confirmations, laid out for the EU AI Act and ISO/IEC 42001. Beside it, an audit trail of every administrative action, including the refused ones, chained so that no entry can be quietly changed.

  • Risk tiers from minimal to unacceptable, with reviews that fall due
  • Changing an entry breaks every check after it
  • Admin changes ask the admin to confirm it is them

Identity and devices

One sign-in, and no way around the rules

People sign in with your identity provider, and each company can require MFA. Enrolled laptops send their AI traffic through your gateway, so the same budgets, inspection and records apply even to AI tools people install themselves.

  • Microsoft Entra ID, Okta, Google or any OpenID Connect provider
  • Devices with a certificate, last seen and status
  • Gateway keys for your own applications

Evaluations

Choose models on evidence, not on headlines

See, for each kind of work, how many answers people rated well, what it cost per request and how much was cached. When a new model arrives, compare it on your own work and see whether it does better, as well or worse.

  • Spend, tokens and ratings by kind of work
  • Runs that compare candidate models, graded side by side
  • Only where your policy allows evaluation

Also in Control

Everything else an organisation needs to run AI responsibly.

  • Event streamingThe audit log, the request ledger and content reads, delivered to your SIEM.
  • HealthHow each provider and service is doing, in one page.
  • Workspaces and peopleWorkspaces, groups, apps, scheduled tasks and approvals across the organisation.
  • Coverage, honestlyWhat is and isn't governed, shown plainly rather than as one flattering percentage.
  • TenantsFor partners: each customer kept apart, run from one console.
  • Step-up for changesPublishing policy or changing keys asks the admin to confirm it is them.
Single sign-on, isolated code, inspection and a full record
Built for data that cannot leavePlatform security →
Back to · View
Everyone sees the numbers they need→ All perspectives
Who uses which, and why→