EU-native Identity Governance & Administration · built in BelgiumEU-native IGA · built in BelgiumCustomer-first · grounded in the running product

AI agents

Ask first.
Every time an agent tries.

Before an AI agent acts on a business system, the system asks RapidValue. The answer comes from decisions your own people already made—and every answer leaves a record.

Animated explainer: an AI agent asks before it acts, is allowed, refused and stopped, and every answer is recorded.

 

Ask first. What an AI agent may do, decided every time it tries.

The decision service and the MCP server are in early access. Owners, review and the kill switch are live today.

Read the whole story
  1. Ask first. What an AI agent may do, decided every time it tries.
  2. Her decision. Maya owns the finance system and decides once: the invoice agent may draft payments, but releasing money needs an approved time window. Then she goes on holiday.
  3. Ask · check. The agent wants to draft a payment. Its gateway asks RapidValue first, and RapidValue checks the facts as they are right now.
  4. Allow · record. Maya’s decision answers for her: yes. The answer names the grant behind it and is sealed in the record, so any later edit would show.
  5. Refused · window closed. Releasing money needs an approved window. This one closed at 14:00, so at 14:00:05 the agent is refused, with the reason.
  6. Stop · kill switch. Sam in security hits the kill switch. The agent’s keys are revoked and it is switched off in Entra, AWS or Google Cloud, where the platform allows.
  7. Stop · the next call. Its next call is refused on the spot, without waiting for the next sync.
  8. Zoom out. Zoom out: every agent, every gateway, one question and one record. Your gateway enforces; RapidValue decides from your people’s choices.
  9. Observe first. You do not switch it on blind. In observe mode RapidValue answers but blocks nothing, so you see what would have been refused before you enforce.
  10. Your assistant asks too · MCP. Jonas asks his own AI assistant which findings need his attention. It asks RapidValue’s MCP server, sees only what Jonas may see, and every question is recorded.
  11. Agents act. Your owners decide. RapidValue keeps the record.

The principles behind it

Six rules we hold ourselves to.

  1. 01

    Owners decide—once.

    The person who owns a system decides ahead of time what an agent may do there. That decision answers for them while they are away.

  2. 02

    Over the loop, not in it.

    People are asked only for what they have not decided yet, such as a short window to release a payment. Everything else answers from standing decisions.

  3. 03

    Ask first, every time.

    Before an agent acts, the system asks. The answer comes from the facts as they are now—not from a key handed out months ago.

  4. 04

    A revoke counts at the next call.

    Stopping an agent does not wait for the next sync. The next request is refused, and the agent is switched off where the platform allows.

  5. 05

    You enforce. We decide and record.

    Your gateway stays in charge of blocking. RapidValue answers from your people’s choices and does not have to sit in your data path.

  6. 06

    Every answer leaves a record.

    Yes or no, each answer is recorded with the decision behind it and sealed every hour, so any later edit would show.

Where it stands

Live, early access, next—said plainly.

Live

  • Agents discovered in Entra Agent ID, Amazon Bedrock, Google Vertex AI and GitHub Apps
  • A named owner and an autonomy tier for every agent
  • Grants with an end date, and approved time windows for sensitive actions
  • A kill switch on Entra, AWS and Google Cloud that reports what it could not stop

Early access

  • Runtime decisions for agent tool calls, starting in observe mode
  • A decision record sealed every hour
  • RapidValue’s MCP server for your own AI assistants

Next

  • RapidValue runs the gateway itself
  • Each yes is good for exactly one action
  • The agent never sees the system’s password

Go deeper

The capabilities behind the story.

AI-agent governanceDiscovery, sponsors, autonomy tiers and the kill switch per platform.Non-human identitiesService accounts, app registrations and the credentials they hold.Trust & architectureWhere RapidValue runs and what crosses the wire.RapidValue MCP serverLet your own AI assistant ask RapidValue—read-only, with your permissions.

Bring one agent

Show us one agent and one system. We will show you its owner, its reach and its record.

Book a 30-min call