NeoXam Agents Puts Auditability First for Asset Managers

Paris-based NeoXam launched an agentic AI platform for asset managers on Wednesday, with audit trails and human-approval controls at the centre of its architecture. The offering, called NeoXam Agents, attaches a fleet of specialised AI agents to the firm’s existing back- and middle-office products and opens an early-adopter programme in the third quarter. The launch lands ahead of the EU AI Act’s high-risk-system rules, which take effect on 2 August 2026.

Most of NeoXam’s clients have already run the AI pilots and seen the demos, said Clément Miglietti, Chief Product and Technology Officer at NeoXam. What they cannot yet do, in his telling, is put an agent in front of a regulator and account for every step it took. NeoXam built that part first. The audit trail, the catalogue of approved agents, and the safeguards for human approval of sensitive decisions are core to the design.

The Three-Part Launch

NeoXam announced the platform from Paris on 24 June 2026, with three distinct pieces. The NeoXam Agents launch announcement names the platform itself, where clients build, run and supervise AI agents, alongside a family of specialised agents in preview and an early-adopter programme for a select group of investment firms. The specialised agents are scheduled for general availability in September 2026, with the programme opening in the third quarter and pointing to production deployments in early 2027.

The platform attaches directly to NeoXam’s existing software. That footprint spans the back and middle offices of an investment firm, covering reference and market data, portfolio and accounting records, position reconciliations, compliance monitoring and regulatory reporting. The agents operate inside the workflows where the operational data already lives. The architecture can run in the cloud, in a hybrid setup, on a client’s premises or in a fully sovereign environment, leaving each firm in control of where its data resides and which regulations apply. Each option ships with the same catalogue, audit trail, and approval safeguards.

Three governance features are built into the platform. The catalogue is a single registry of approved agents. The audit trail is a complete, reviewable record of every action an agent takes. The approval safeguards force human sign-off on decisions the firm flags as sensitive.

NeoXam’s customer base already runs through these same workflows. The firm’s about page lists systems that manage more than €25 trillion in assets for 150+ customers across 21 offices, with 750+ staff. The customer mix includes asset managers, pension funds, insurers and asset servicers. The 2024 acquisition of EZOPS, an AI reconciliation specialist, brought the technology base that now sits inside NeoXam Aro. NeoXam Agents is the next move from that base.

Five Agents, One Job Each

The platform does not ship a general assistant. Instead, NeoXam is releasing five specialised agents, each tied to one of its existing products and each focused on a narrow operational task. They fall into three groups: agents that answer questions, agents that take action, and agents that help configure the software.

The design choice is deliberate. A general assistant would be easier to deploy but harder to govern, and governance is what the customers keep asking about, Miglietti said. Each specialised agent has a narrower surface to monitor, a tighter audit trail, and a clearer hand-off to a human when the decision sits outside its mandate. The architecture lets a firm run a fleet of agents in parallel without losing visibility into what any one of them did. That visibility is the point of the catalogue.

The agents attach to the products where the data already lives. The reconciliation agent runs inside Aro, the same reconciliation system NeoXam has sold to asset managers for years. The compliance agent runs inside PMS, the portfolio management and compliance system. Configuration work happens inside DataHub, the enterprise data management platform; reporting sits in Impress; knowledge agents span the suite.

What the early outcomes look like varies by workload. Some tasks, like configuration, produce clean handoffs to the agent; others, like compliance triage, route the harder cases back to specialists. Every action is logged, every decision can be reviewed, and the agent fleet can be certified to a regulator.

The workload split defines the operating model: routine tasks go to the agent, judgement calls stay with people.

Auditability Comes First, Features Second

The platform’s centre of gravity is regulator-facing controls, and Miglietti made the framing plain. Asset managers can buy an agent, but running a fleet of them inside a regulated firm is harder than the demos suggest. The gating problem, in Miglietti’s framing, is the production problem. NeoXam built that piece first.

Most of our clients don’t have an AI problem. They have a production problem. They’ve run the pilots and seen the demos. What they can’t do yet is sit an agent in front of a regulator and account for every step it took, so we built that part first.

What auditability looks like inside the product is concrete. Every action an agent takes is logged in a complete, reviewable record, with a compliance officer able to inspect the actions of every agent in the fleet and certify the deployment to a regulator. Human approval is required for any decision the firm marks sensitive, and the platform’s catalogue holds only those agents the firm has approved.

The buyers in this market are not software engineers. They are chief operating officers, heads of compliance, and chief risk officers at firms whose systems manage more than €25 trillion in client assets. Those buyers will not put an agent in front of a regulator until the audit story is complete, the press materials argue. NeoXam built the platform around that gate, with the regulator-facing review screen as the part customers ask about first, Miglietti said.

The governance layer is the design lead time the customers cannot compress themselves. Most asset managers have spent the past two years running pilots and writing procurement memos about model selection. NeoXam’s bet is that buying the governance layer with the platform shortens that clock.

A Bet Against Model Lock-In

The platform’s second design choice is model portability. Clients can choose, and later switch, the underlying AI model without rewriting the agents they have approved. The choice is also a hedge against vendor lock-in for a regulated firm signing a long contract. Miglietti framed the move as deliberate, not opportunistic.

We are deliberately not betting our clients’ firms on one model provider. If a better model comes along next quarter, they should be able to use it without a migration project. Locking a regulated institution into a single AI vendor for a decade is a risk, not a feature.

The platform abstracts the model behind the catalogue. A firm can route one task to one model and another task to another, then change the routing as the model market shifts. The same swap inside NeoXam Agents is a configuration change in the catalogue, with the agents themselves unchanged.

The model market moves fast, and a model state of the art in early 2026 can be eclipsed within a quarter, creating a version of lock-in that does not show up on the procurement sheet for a regulated institution signing a ten-year software contract. What an end-to-end agentic finance redesign looks like is part of the same shift, with portability cited as one of three adoption hurdles in EY’s analysis of finance operations. NeoXam’s argument is that portability belongs inside the audit story, alongside the catalogue and the human-approval gates. The early-adopter programme will test that argument under live regulatory conditions.

The Production Gate Meets the EU AI Act

The launch lands before the EU AI Act‘s high-risk-system rules take effect on 2 August 2026. Credit-scoring and compliance-triage agents, the categories NeoXam’s platform covers, fall squarely inside the high-risk classification. From August, providers and deployers of those systems face detailed obligations on data quality, transparency, human oversight and documentation. The first high-risk system the platform covers, the compliance-triage agent inside PMS, will be subject to those obligations from day one of deployment. The reporting and configuration agents sit further from the high-risk line, but the audit posture the platform ships is uniform across them.

The platform’s design aligns with those obligations in four places: a complete audit trail for every agent action, a catalogue of only the agents the firm has approved, human-approval gates for sensitive decisions, and the freedom to swap models without rewriting the agents. Agentic AI project cancellation cost analysis from EY cites a Gartner poll forecasting that more than 40 percent of agentic AI projects will be cancelled by the end of 2027. The forecast cites escalating costs, unclear business value, and inadequate risk controls as the three drivers. The features NeoXam has led with are the features that address those failure modes.

The two tracks, regulatory deadline and platform architecture, meet at the same gate. The EU AI Act high-risk rules take effect on 2 August 2026, with the platform’s specialised agents reaching general availability in September 2026. The early-adopter programme runs in parallel for a year, ending with production deployments in early 2027.

The opening move NeoXam is making with the platform is to put the audit posture in the architecture itself, ahead of the productivity features.

NeoXam Structure Set the Template

The agent platform sits on top of a recently deployed precedent. NeoXam Structure, the firm’s AI-powered document processing capability, extracts figures from PDFs, broker statements and trade confirmations and routes validated data into internal systems. A-Team Insight reported that processing time fell from 15 minutes to about 1 minute per document. The NeoXam Structure product page lists 95%+ accuracy on PE transactions, broker confirmations and reference data. The accuracy is measured at the field level, with confidence scoring on each extracted value.

Structure is in production at customer sites today, with the same governance posture the agent platform now extends. A regulated institution can route AI actions through the same reviewable record Structure already uses for extracted fields, with the same flags for human review. The early-adopter programme will measure how the trust placed in Structure carries over to agents that take operational action.

Frequently Asked Questions

What is NeoXam Agents?

NeoXam Agents is an agentic AI platform launched by Paris-based financial software firm NeoXam on 24 June 2026. The platform lets investment firms build, run and supervise AI agents inside the back- and middle-office workflows where the firm’s existing products already operate, with built-in audit trails, a catalogue of approved agents, and human-approval controls for sensitive decisions.

When will the specialised agents be generally available?

NeoXam said the family of specialised agents is in preview at launch and will become generally available in September 2026. The agents cover reconciliation in Aro, compliance triage in PMS, configuration in DataHub, reporting in Impress, and a knowledge function across the product suite.

How does the early-adopter programme work?

The programme runs for one year starting in the third quarter of 2026, with test deployments through the fourth quarter and a path to production in early 2027. Participation is open to a select group of investment firms; NeoXam has not disclosed the size of the cohort.

What performance numbers has NeoXam reported?

NeoXam’s launch materials cite up to a 75% reduction in manual reconciliation work from the Aro agent, a 90% reduction in time required to address compliance alerts from the PMS agent, and around 90% time savings on routine configuration tasks from the DataHub agent. NeoXam Structure, the document-processing capability that preceded the launch, reports processing time dropping from about 15 minutes to about 1 minute per document.

How does the platform relate to the EU AI Act?

The platform’s design aligns with the EU AI Act’s high-risk-system rules, which take effect on 2 August 2026. Compliance-triage and credit-related agents fall inside the high-risk classification, and the platform’s audit trail, catalogue, and human-approval controls are the features that map to the Act’s documentation and oversight obligations.