motif

The removal test leaves an empty interface. The product is a temporal knowledge graph that models how markets, assets and financial relationships connect and how those connections change over time, and a set of agents that reason over it to explain a market movement in terms of one client's holdings, build an investor profile through conversation, or produce and justify a portfolio recommendation. There is no workflow layer, no data terminal and no non model component that would survive on its own. The founding argument is itself a claim about model architecture rather than about software.

The most explicitly reasoned autonomy position in this index, and it is unusual because it argues for more machine autonomy rather than less. The company publishes a taxonomy of three approaches and names which it is building: not copilots that speed up research, and pointedly not the pattern where the machine explains options, the customer clicks invest and the platform thereby avoids full advisory responsibility, which it identifies as a limited position.

It builds instead for artificial intelligence that recommends, executes, monitors and documents decisions with the institution providing supervision, licences, insurance coverage and accountability. That places high autonomy deliberately inside the regulated perimeter rather than beside it, and the reasoning is stated: licensing, examinations, supervision culture and clean track records take years to build, institutions already have them, and the machine should operate under that umbrella. Supporting mechanisms follow, with trades routed through custodians under proper authentication, settlement creating audit trails, and recommendations carrying reasoning a client can interrogate.

The failure mode is named with unusual precision by the chief executive, who describes existing financial services artificial intelligence as wrappers around language models connected to data feeds that retrieve information without understanding it, producing hallucination, inconsistency and no comprehension, and states the system was rebuilt from the ground up in response.

The answer offered is architectural rather than procedural: a temporal knowledge graph holding structured relationships and their changes over time, output described as structured and sourced, and reasoning exposed so a client can ask why a recommendation was made and what would happen under different assumptions. Interrogable reasoning is a real control.

What is entirely absent is measurement, and with a system launched weeks before assessment there is no accuracy result, no evaluation and no independent benchmark against the failure mode it was built to solve.

Founded in December 2024 with its core system launched in May 2026, so the evidence base is thin by age. One figure is stated and it is a meaningful one for a company at this stage: multiple financial institutions held contracts ahead of launch, collectively representing more than 1.5 million end users and billions in assets under management. That is committed demand rather than live deployment, and no institution is named, no user is yet served through it, and no outcome is measured.

Backing comes from a venture creation group founded by a major sovereign investor, and the founding team describes a combined 60 years inside the institutions it now sells to, as advisers, engineers and product leads.

No data boundary statement was located. The architecture implies a shared substrate, since a knowledge graph tracking market relationships over time is most valuable when it is common to all users, while portfolios and investor profiles are necessarily per institution, and nothing describes where the line falls between them.

The question matters because the same system will serve competing wealth platforms, and whether client behaviour or profile data observed at one contributes to anything served at another is unaddressed.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The profiling agent creates a distinctive exposure, since replacing a static questionnaire with adaptive conversation means the system holds a running record of how a person discusses money, risk and their own circumstances, described by the company as a living investor profile, which is a richer and more revealing artifact than the form it replaces. Nothing states how long that profile persists, whether the end client knows it is being maintained, or how it is separated between the institutions deploying the agents.

No attestation, certification, trust centre or enumerated framework was located. That is unsurprising for a company eighteen months old with a system launched weeks ago, and it is also the gap that will decide whether the institutions holding pre launch contracts can move into production, since a bank or wealth manager embedding a third party agent into its own customer application will require an independently assessed control set before that agent speaks to its clients.

The posture is articulated rather than assumed, which is what this axis rewards for a technology supplier. The company holds no advisory licence and does not claim one, and it has thought through where the licence sits instead, designing so that the deploying institution supplies supervision, permissions, insurance and accountability while the technology operates beneath them.

Framing that as the institution's durable advantage rather than as a compliance inconvenience is a more considered position than most vendors take. What is absent is specificity: no regulator, statute or conduct rule is named, and for a system generating suitability profiles and investment recommendations across multiple jurisdictions the applicable regimes differ materially and none is identified.

Two things sit opposite each other and neither is measured. The inclusion argument is explicit and credible, that institutions can now serve customers who previously sat below the threshold at which human advice was economic, which is the clearest statement of that case in this lane and describes real expansion rather than reallocation of existing service.

Against it, the profiling agent infers risk tolerance, capacity and objectives from conversation rather than from a completed form, and conversational inference varies with how articulate, confident and financially literate a person is, so the client who expresses themselves least clearly may be profiled least accurately and steered accordingly. No fairness testing, no analysis of profiling accuracy across client populations and no account of how recommendations vary by portfolio size was located.

The allocation is the most explicitly reasoned in this index and it is deliberately weighted toward the institution rather than away from it. The company describes the institution providing supervision, licences, insurance coverage and accountability, and naming insurance specifically is unusual, since it acknowledges that something will eventually go wrong and identifies who carries the cost.

It also rejects by name the design where a platform lets the customer click invest in order to avoid full advisory responsibility, which is the arrangement most consumer facing systems adopt precisely to shed liability. The end client gains genuine explanation, able to ask why a recommendation was made and what would change under different assumptions. What is missing is anything binding the vendor itself, and no correction process is described for a recommendation later found to rest on a faulty inference.

The knowledge graph is presented as the company's own work, rebuilt rather than assembled, which is a disclosure about approach rather than about dependencies. Nothing else is named. No model provider appears for the agents that reason over the graph and converse with clients, no market, news or reference data supplier is identified even though the system's whole function is connecting those feeds, and no hosting arrangement or subprocessor list was located. For a product an institution will embed directly into its customer experience, the fourth party question arrives early and nothing published answers it.

The delivery model is designed for integration rather than adoption, with a modular interface and toolkit that an institution drops into an existing product, selecting individual agents, matching tone and language to its own brand and going live in days rather than months, so the end customer never leaves the institution's application. That is the right shape for this buyer.

The wider architecture the company describes, connecting recommendations to execution, routing trades through custodians under authentication and generating settlement audit trails, is stated as what it is building rather than what it has shipped, and no custodian, portfolio system or market data provider is named anywhere.

No hosting provider, region selection, residency commitment or private deployment option was located. The company is incorporated and headquartered in Switzerland, which carries its own data protection expectations, and it sells to institutions in multiple jurisdictions with differing requirements, so the question of where client portfolios and investor profiles rest arises immediately on any cross border deployment. Nothing published addresses it.

No pricing, packaging or basis of charge is published. The adoption model is described, with institutions selecting only the agents they need and the company recommending deploying one module, proving value and expanding from there, which tells a buyer they are not obliged to take the whole system at once. That is a commitment shape rather than a price, and nothing indicates whether charge falls per module, per end user, per query or on assets served.

Six buyer types are addressed explicitly, spanning established financial institutions, neobanks, fintech platforms, wealth managers, digital asset platforms and investment applications, which is a wider set than most vendors in this lane name and reflects a product designed to sit inside somebody else's application rather than to be one. Function coverage runs from market insight through suitability profiling to portfolio recommendation.

The limits are stage rather than intent: no geographic footprint is evidenced, no institution type is demonstrated in production, and the execution and rebalancing layer that would complete the proposition is described as being built.