Summary

  • Sidetrade says SAFE combines its privately operated models and compute with a shared policy framework for order-to-cash agents; subscriptions include tasks and “O2C IQ” usage.
  • Compute-price predictability is not unlimited work or a fixed total operating cost. Production Cash Collection Agent use must also be separated from the planned fourth-quarter beta-customer rollout of two other products.

A finance team can stop worrying about one supplier's token price and still have an unsettled automation bill. The next questions are contractual and operational: what counts as a task, what happens when the allowance runs out, and who finishes a case that the agent must pass to a person?

Sidetrade's September 10 SAFE announcement brings those questions into the order-to-cash software contract. Order-to-cash is the chain of business work that takes an order through billing and payment. SAFE is presented as a common foundation for Aimie IQ, ready-made Aimie Agents and Customer Agents built in Agent Builder Studio. Sidetrade says subscriptions include volumes of tasks and O2C IQ, its token-equivalent measure, while multi-year contracts guarantee compute pricing.

That describes included capacity, not an unlimited entitlement. The reviewed announcement provides no tariff, allowance size, definition of a billable task, retry rule or overage schedule. It also does not establish that implementation, supervision and exception handling are covered by a single fixed total price.

A policy stop can be the correct result

SAFE's architecture separates agent connections from execution. Sidetrade describes registries, reusable skills and event routing on one side, and a runtime layer that assembles context, chooses models and tools, applies policy and keeps an audit trail on the other. Some work is predictive modelling; routing and policy steps use deterministic code; language-heavy tasks reach fine-tuned open-weight language models. These are the vendor's descriptions, not independently inspected execution traces.

The connection between permission and cost is particularly useful. Sidetrade says deterministic escalation prevents agents spending tokens on decisions reserved for humans. A correctly stopped action can therefore be evidence that a boundary worked. It is not necessarily an unsuccessful automation attempt to be pushed through on the next try.

Nor does it finish the business case. Someone may still need to resolve the exception. The execution-layer product page describes autonomous actions under human governance and custom workflows with controls and audit trails. It does not mean a person approves every action in advance, and it does not disclose a particular customer's rules. Buyers need the cost of an appropriately completed task, including necessary review, rather than a count of model calls alone.

The production clock has more than one date

This is not the first announcement of included-volume subscriptions. Sidetrade's earlier first-half report already described agent-task allowances and O2C IQ in the base Aimie IQ subscription. It also distinguished Cash Collection Agent, launched in the first quarter, from the second-quarter firm pre-bookings for Aimie IQ and Agent Builder Studio.

The September announcement says SAFE is in production beneath Cash Collection Agent, naming Securitas, Accor and Sodexo. It schedules production rollout of Aimie IQ and Agent Builder Studio to beta customers for the fourth quarter of 2026. Those supplier-reported stages cannot be collapsed into a claim that every component is already deployed generally. Orders and pre-bookings are not completed collections.

Private infrastructure changes another part of the dependency. Sidetrade says it operates its own teams, GPUs and interconnected private-cloud data centres on both sides of the Atlantic, without sending customer prompts or records to public model providers or hyperscalers. That is a supplier assertion, not proof of customer-owned on-premises hosting, a chosen-country residency guarantee or independently tested isolation.

The same care applies to security language. Sidetrade says agents are checked against OWASP's Agentic Top 10 before registry publication. OWASP describes that resource as a framework for identifying and reducing agentic risks. It is not an OWASP certification of SAFE or evidence that a customer's particular workflow has passed an independent audit.