Sidetrade launches SAFE agentic framework with owned GPU and private
Sidetrade (Euronext Growth: ALBFR.PA) has launched SAFE, the Sidetrade Agentic Framework for Enterprise, a full-stack platform for deploying autonomous AI agents across the Order-to-Cash cycle. The company says SAFE runs entirely on infrastructure it owns: private data centres on both sides of the Atlantic, GPU compute purchased rather than rented, and fine-tuned open-weight language models alongside proprietary predictive models for payment behaviour and buyer risk.
The announcement positions SAFE as a direct answer to two concerns Sidetrade says CFOs are raising with agentic AI: uncertain running costs driven by third-party per-million-token pricing, and data sovereignty risk when sensitive receivables data flows through hyperscaler or frontier-lab infrastructure. Sidetrade sells SAFE on subscription, bundling a task volume and token equivalents it calls "O2C IQ," with compute pricing guaranteed on multi-year contracts.
What SAFE comprises
SAFE is organised into two runtime layers. The Agent Connect layer handles event routing, an agent registry, and a library of reusable Order-to-Cash skills. The Agent Harness governs each agent at runtime, assembling context from Sidetrade's data lake, orchestrating reasoning on its own models, enforcing policy, and maintaining memory across runs so agents do not repeat the same reasoning from scratch.
The framework underpins three products: Aimie IQ, a natural-language orchestration interface; Aimie Agents, a library of pre-built agents starting with the Cash Collection Agent already deployed with multinationals including Securitas, Accor and Sodexo; and Agent Builder Studio, through which customers or partners can build their own agents. Aimie IQ and Agent Builder Studio are scheduled for production rollout to beta customers in the fourth quarter of 2026.
Sidetrade's infrastructure carries ISO 27001 and SOC Type II certification, and agents must pass validation against OWASP's Agentic Top 10 before publication to the agent registry. Chief Technology Officer Mark Sheldon noted that Sidetrade applies the same framework internally, describing an "inside-first" development model intended to accelerate platform iteration.
The data layer that supports SAFE is a proprietary O2C data lake aggregating close to $10 trillion in B2B transactions across nearly 45 million buying companies, accumulated over more than a decade. Olivier Novasque, CEO and founder, stated: "Owning the stack only makes AI sovereign. What makes it intelligent is the data, and that is the layer we own that no AI company can easily buy or replicate."
Market context and competitive read
The agentic-AI-for-finance category is forming quickly, with specialist Order-to-Cash vendors competing against broader ERP incumbents extending their AI layers and a cohort of well-funded horizontal agent-platform startups. Sidetrade's vertical stack argument mirrors a broader debate in enterprise AI about whether buyers are better served by composable, best-of-breed agent tooling built on hyperscaler inference or by vertically integrated platforms with proprietary compute and data.
The sovereignty and cost-predictability framing is well-timed. Consumption-based inference pricing has made CFO sign-off on large agentic deployments genuinely difficult; several vendors across the market are responding by shifting toward subscription wrappers, though few outside the largest hyperscalers claim to own the underlying GPU infrastructure.
Sidetrade's certifications and OWASP validation address governance requirements that are becoming standard procurement criteria for finance-sector buyers, particularly under the EU's Digital Operational Resilience Act, which extends cyber-risk obligations to third-party technology providers serving regulated entities. The company operates across 85 countries with 450 employees, giving the SAFE launch global commercial reach from day one.
The SAFE launch is framed as a milestone in Sidetrade's O2C Intelligence 2030 plan, which targets over 50% of revenue from AI-native products and an EBITDA margin of 30% to 35% by 2030. The company's first-half 2026 financial results are scheduled for release on 22 September 2026, which will offer the first public read on how AI-native revenue is tracking against that plan.