All engineering notes
Web3, Payments, design record

Agent-to-Agent Deferred Settlement

When software agents transact with each other continuously, paying on-chain per call is both expensive and revealing: every counterparty learns your spending pattern. This design applies a familiar consumer-finance idea to that problem. Agents accrue obligations under a credit line, settle periodically rather than per transaction, and keep the credit relationship inside a private subnet — so a competitor cannot read your terms while the network can still verify that settlement happened.

PeriodicSettlement, not per transaction
PrivateCredit terms, subnet-scoped
AgentsCounterparties, software, not humans
Design + prototypeStage, honestly labelled

Repository shape

Design repository with a prototype implementation

How this is presented

Documented as a design record. This is an early-stage concept and is presented as one — the reasoning is the artefact, not a product.

Accrue privately, settle publicly

Follow a stream of agent-to-agent calls and watch where obligations accumulate before anything reaches the public chain.

Allowance policyService

Bounds what the buying agent may commit on its principal's behalf.

Client surfaceServiceOn-chainWorker / job
Accruing an obligation

What happens on each of many calls.

Highlights

  • Moves agent commerce off per-call payment, which is prohibitively expensive at machine transaction rates.
  • Keeps the credit relationship private, so counterparties cannot infer commercial terms from public chain data.
  • Settlement remains publicly verifiable even though the terms behind it are not.
  • Allowance policy bounds what an agent may commit on its principal's behalf.
  • Subscription and deferred-payment models expressed in machine-readable terms agents can evaluate.

Private credit, public settlement

Obligations accrue inside a permissioned subnet where only the parties see the terms. Periodically, net positions settle and the settlement is anchored publicly — giving the network verification without disclosure.

Agent layer

Autonomous parties transacting under a principal's allowance policy.

  • Each call is authorised against an allowance rather than paid for individually.
  • Terms are machine-readable so agents can evaluate them without a human.

Credit subnet

Holds the credit relationship, accrued obligations and terms.

  • Visible only to the parties involved.
  • Accrues rather than settles, which is what makes high call volume viable.

Settlement

Periodic netting and public anchoring.

  • Net positions settle on a schedule instead of per transaction.
  • The public record proves settlement occurred without revealing terms.

Design surfaces

An early-stage design repository: the documentation carries as much of the work as the prototype does, which is stated plainly rather than dressed up. 8 units across 4 groups.

Prototype

1
app

Design record

1
docs

Commerce model

4
deferred settlementsubscriptionscredit linesallowance policy

Privacy

2
permissioned subnetperiodic anchoring

Stack by layer

Commerce model

Deferred settlement, Subscriptions, Credit lines, Allowance policy

Privacy

Permissioned subnet, Member-scoped visibility, Periodic anchoring

Agent interface

Machine-readable terms, Per-call authorisation, Programmatic settlement

Delivered work

Design and prototype

  • Deferred-settlement and subscription models for agent-to-agent commerce.
  • Private credit relationship scoped to a permissioned subnet.
  • Allowance policy bounding agent spending authority.
  • Prototype implementation of the settlement path.

Key modules

Deferred settlement

Makes high-frequency agent commerce economically viable.

  • Accrue continuously, settle periodically.
  • Removes the per-call transaction cost.

Private credit terms

Stops counterparties reading your commercial position.

  • Terms live in the subnet; only settlement is public.

Allowance policy

Bounds autonomous spending.

  • A principal delegates a budget, not unlimited authority.

Presentation approach

How this project is presented

  • Presented as a design record, since that is what it is.
  • The economic argument explained rather than benchmarked.
  • No claim of production readiness.

Documentation and notes

Settlement model

Accrual, netting and the anchoring schedule.

Privacy rationale

Why credit terms are subnet-scoped and settlement is not.
  • Network and organisation names are intentionally omitted.
  • This is an early-stage design with a prototype, deliberately not presented as a shipped product.
All 21 engineering notes