DuoKey

One compromised agent should not empty the fleet.

When AI agents can pay, transfer, or approve spend, a shared key or a secrets file becomes a business outage waiting to happen. Agent Vault gives each agent its own wallet and a clean kill switch, so you can revoke one agent without rotating credentials for every other agent in production.

What breaks when agents hold money

Autonomous payments need the same controls you expect from treasury, ownership, limits and a way to stop one bad actor without freezing the business.

01

A single secret empties the agent

If the full signing key lives in one vault, TEE, or .env file, stealing it drains the agent, often with no alert and no clean recovery path.

02

Revoking one agent hurts the fleet

Shared service credentials mean shutting down a compromised agent often forces a fleet-wide key rotation and operational downtime.

03

Autonomy without spend boundaries

An agent that can sign anything is one prompt injection or software bug away from moving money it should never touch.

04

No independent agent identity

Without a named identity separate from the key, you cannot prove which agent acted, or revoke authority without touching the wallet itself.

Contain compromise without stopping the programme

Contain compromise without stopping the programme

Agent Vault splits each agent's signing authority so no single party, the agent, your operators, or DuoKey, can move funds alone. Business outcome first: a stolen agent process cannot empty the wallet; a compromised operator laptop cannot either.

  • No single point of failure: the full private key is never assembled in memory, on disk, or on the network

  • Blast-radius control: one stolen shard cannot produce a valid signature on its own

  • Policy outside the agent: spend limits, payee allow-lists and escalation rules are enforced where your cloud shard lives, not inside the agent's own process

Revoke one agent. Leave the rest running.

Every agent gets its own named identity. If one is compromised, decommissioned, or simply misbehaving, an operator pulls its authority in a single call. That agent is blocked immediately; every other wallet keeps working, no fleet-wide rotation, no emergency change window.

Above a spend threshold you define, signing does not complete on policy alone, it escalates to a human co-signature before money moves.

Revoke one agent. Leave the rest running.

What changes for the business

Controls finance, security and platform teams need before agents are allowed to spend

  • Named identity per agent

    Each agent signs as itself, not from a credential copied across your fleet, so audits and incidents have a clear owner.

  • No single point of compromise

    A stolen agent, operator device, or cloud component alone cannot produce a valid signature.

  • Instant, targeted revocation

    Pull one agent's authority and stop it immediately without touching any other wallet.

  • Policy enforced outside the agent

    Spend limits, allow-lists and escalation rules stay outside the agent process, so a compromised agent cannot sign its way around them.

Next step

Still weighing the control model?

Bring the constraint. We’ll map the shortest practical path.

Let agents spend, without handing them the vault

Agent Vault is rolling out by invite. Talk to us about per-agent wallets, spend policy and revocation before your next automation goes live.