Core engine built and tested. Product in development.

Backed authority for AI action.

Let AI agents act. Stay in control.

Avallo checks each consequential action an AI agent proposes against limits set by accountable people, at the moment it would take effect. If the action is outside those limits, it does not go ahead.

Discuss an Avallo pilot What Avallo does

The challenge

AI agents are moving from insight to action

AI agents are beginning to commit funds, change records, approve requests and send messages on an organisation's behalf. Those actions must stay within policy, delegated authority and risk appetite, even when an agent is manipulated, compromised, overreaching or simply wrong.

Today, most organisations cope by:

  • Having a person review every action
  • Keeping AI to low-risk tasks
  • Monitoring after the fact, once the action has taken effect

That is slow, costly and hard to scale.

For each action, the real question is whether this specific action is authorised right now, which makes this a delegated-authority problem.

The solution

Check every consequential action before it takes effect

Avallo sits between AI agents and the systems they act on. Accountable people set limits in advance for each workflow or role: which actions, with whom, up to what amount, and when a person must approve first. When an agent proposes an action, Avallo checks that one action against those limits at the moment it would take effect. The action goes ahead only with a single-use authorisation for it, and Avallo keeps a record of every decision that can be checked independently.

In other words, the agent can ask for anything. It can only do what accountable people have already approved.

One check, three outcomesYour people set the limits in advance: which actions, with whom and up to what amount. An AI agent proposes an action, for example paying a supplier invoice. Avallo checks it at the moment it would take effect. Within the limits, the action goes ahead once, with a single-use authorisation. Above the approval threshold, a named person approves first. Outside the limits, it is refused and nothing happens. Every decision is recorded and can be checked independently. Your peopleset the limits in advance:which actions, with whom, how muchAI agente.g. pay a supplier invoiceproposes an actionAvallochecks it at the momentit would take effectWithin the limitsGoes ahead once, with asingle-use authorisationAbove the approvalthresholdA named person approves firstOutside the limitsRefused. Nothing happens.Every decision is recorded and can be checked independently.One check, three outcomesYour people set the limits in advance: which actions, with whom and up to what amount. An AI agent proposes an action, for example paying a supplier invoice. Avallo checks it at the moment it would take effect. Within the limits, the action goes ahead once, with a single-use authorisation. Above the approval threshold, a named person approves first. Outside the limits, it is refused and nothing happens. Every decision is recorded and can be checked independently. Your peopleset the limits inadvance: whichactions, with whom,how muchAI agentproposes an action,e.g. pay a supplierinvoiceAvallochecks each action at the momentit would take effectWithin the limitsGoes ahead once, with asingle-use authorisationAbove the approval thresholdA named person approves firstOutside the limitsRefused. Nothing happens.Every decision is recorded andcan be checked independently.

Example: a payment

A bank's own purchasing agent receives an invoice with changed bank details and proposes paying the new account. Avallo refuses the payment, because the payee can only come from the approved supplier list. No money moves, and the refusal is recorded.

  • Routine work needs no person. Actions inside approved limits go ahead. People see only the exceptions.
  • Each action needs its own authorisation. Where a system acts only on Avallo's single-use authorisation, a manipulated or compromised agent cannot go beyond its limits.
  • A record of every decision. Audit, or an independent party, can check who authorised what.

Use cases

Any action that commits the organisation

Avallo is designed to extend to any consequential action an AI agent takes, and the examples start with customer service. Each kind of action is specified and built before Avallo governs it. Today, the core engine is built and tested for account-to-account payments.

  • Customer service

    Refunds and fee waivers

    An agent refunds a customer or waives a fee. Above its limit, a named person approves first.

    Not yet built

  • Customer service

    Contact and bank details

    An agent updates a customer's phone, email or bank details, a common step in account takeover. Only changes within approved conditions go ahead.

    Not yet built

  • Customer service

    Hardship arrangements

    An agent agrees a hardship or repayment arrangement, only within terms people have approved.

    Not yet built

  • Customer service

    Credit limits

    An agent raises a card or credit limit, up to an approved amount and no further.

    Not yet built

  • Payments and agentic commerce

    Supplier payments

    A purchasing agent pays a supplier invoice, only to an approved supplier and within approved amounts.

    Core built and tested

  • Operations

    Records, approvals and IT changes

    An agent changes a record, approves a request or changes a system, within the authority people have delegated.

    Not yet built

Product in development

Where Avallo stands today

Avallo's core engine is built and tested; the product is in development. Productisation will be through a planned majority Australian-owned joint venture.

Built: the enforcement core. It checks each proposed action against its limits, issues a single-use authorisation, enforces it before the action takes effect and records each decision. It is built to production standards and tested, for account-to-account payments.

To be productised, integrated and hosted in Australia by TrustWorks360, a Melbourne company operating since 2019.

Still to build

  • The distributed, fault-tolerant layer
  • Connections to payment systems
  • The authority workbench
  • The approval and monitoring screens
  • Hosting

Discuss a pilot

Stand behind every AI action

Enable AI agents to act with authority that is backed, bounded and provable.

Start with one workflow

We are seeking a first pilot with an Australian financial institution: one consequential workflow, hosted in Australia and run in shadow mode alongside your existing controls. Shadow mode records what Avallo would have decided and releases nothing.

  1. DesignChoose the workflow and set its limits from your existing policies and delegations.
  2. Shadow modeAvallo runs alongside your controls and records what it would have decided. It releases nothing.
  3. Prove valueCompare its decisions with your expected outcomes, then agree the next step.

Discuss an Avallo pilot

Tell us about the workflow you have in mind. Phone is optional.

The button activates once the security check above is complete.

We use these details only to respond to your enquiry. We do not sell them or add them to marketing lists. Read the privacy notice.