Opensee

The AI line is real and growing, and the engine underneath does not depend on it. The company brands itself AI first and has a working agentic product. Its own description names the ingredients in a revealing order, pairing high performance computing with AI. Its market position was earned as a category leader in risk data aggregation, a feat of architecture rather than inference.

The value proposition is real time access to every data point at full granularity, across very large volumes and long histories. A horizontally scalable store and deterministic calculation deliver that, not a model. Without Agensee, the generative assistant and the data quality assistants, the platform still ingests, aggregates and recomputes value at risk.

It serves the same users through its interface and native Python integration, which is what it did successfully for years before the agentic layer existed. The data arrives through ingestion pipelines rather than through extraction, so the software keeps everything it operates on when the models are removed.

Oversight properties are asserted with real specificity, and the enforcement mechanism is still not described. On the credit side the company states full regulatory audit trails. The semantic layer is presented as producing information that is auditable and verifiable, and automated explainability is offered alongside the natural language interface. Data versioning is a documented capability.

It lets a user reconstruct what the platform said at a past point in time, a concrete form of accountability for an analytics product. What is missing is the control path around the agentic layer. Agensee is described as automatically building data models, calculators and dashboards. Nothing states who approves a model it generates before risk numbers are produced from it, or what happens when it is wrong. Nor does anything say whether a human signs off before an automatically built calculator feeds a regulatory report.

The strongest model claim is also the least checkable one. The semantic layer is described as producing information that is auditable, verifiable and provably correct. Provably correct is an absolute with no proof method attached. Automated explainability is offered without a description of what is explained or how. The agentic capability automatically builds data models and calculators, yet there is no accuracy measurement, error rate or human review rate.

Nor is there a versioning policy for the models it generates, or any drift monitoring. That matters because an automatically constructed calculator feeding a regulatory risk number is exactly the artifact a supervisor would ask to validate.

A named tier one bank is quoted, and the independent evaluations behind it are dense. Credit Agricole CIB describes the platform as a change in the speed, efficiency and governance of risk management for its market activities. Further quoted material credits access to granular and historical information and taking trade analytics further. Taula Capital, a large hedge fund launch, is named as a deployment. The independent recognition is specific and dated rather than generic.

Opensee is a category leader in the Chartis RiskTech Quadrant for market risk risk data aggregation. It climbed twenty places in the Chartis RiskTech100, with a special award for risk data aggregation and reporting, and is included in the Chartis RiskTechAI50 for use of AI in risk data. At the Risk Markets Technology Awards 2026 it was named Risk Data Repository and Data Management Product of the Year, and it has two further 2025 awards. No figure is attached to any named institution. The bank's endorsement is qualitative, and the millisecond value at risk recomputation claim is self reported and demonstrated in the company's own webinar.

Nothing states whether customer data trains the vendor's models. One architectural property runs in the customer's favor, though it is not a commitment. The platform is described as delivering analytics without moving the customer's data. Deployment inside the customer's own cloud account means the dataset does not leave infrastructure the institution controls. That constrains where data sits.

It is not a statement that the data is excluded from model training, and no such statement exists. The platform holds granular positions and exposures for competing institutions, so the separation question is a live one and goes unanswered.

No privacy posture is published. The data involved is overwhelmingly institutional rather than personal, which lowers the exposure relative to a consumer facing vendor, but nothing addresses retention, access controls over historical positions, or the handling of any personal data reaching the platform through counterparty, client money or retail credit portfolio records, and the credit risk product explicitly covers retail portfolios.

Opensee claims no security certification. There is no ISO 27001 and no SOC report of either type, and no penetration testing statement or trust portal. The absence matters for a supplier whose platform holds full granularity trade, position and exposure data for an investment bank's market activities, among the most sensitive data a financial software supplier can hold. The deployment architecture eases part of the concern. Running inside the customer's own cloud account means the institution's own controls apply to the environment, though that is not a certification.

This is a software supplier with no license, authorization or supervisory relationship of its own. The regulatory language throughout describes what the platform helps a bank achieve, namely auditability, regulatory reporting and compliance with capital and liquidity standards. That is a product claim rather than a statement of standing. No sandbox participation, registration or enrollment in any supervised program is published.

There is no fairness or AI governance position of any kind. Nor is there a reference to any named framework or to the EU AI Act, despite a French base and a European bank buyer base. The direct fairness exposure is lower than for a consumer lending vendor, because the outputs are portfolio and market risk metrics rather than decisions about individuals. It is not zero.

The credit risk product surfaces probability of default, loss given default and exposure at default across retail and small business portfolios. How those aggregates are constructed and presented shapes where a risk team directs attention.

No recourse position is published. Nothing states what happens when an automatically generated calculator is wrong or a data quality assistant fails to flag a bad feed. Nor does anything cover the generative assistant answering a risk question incorrectly, or who carries the consequence if an erroneous figure reaches a regulatory submission. The vendor is purely institutional.

Every affected party is a sophisticated firm operating under a negotiated contract, so recourse collapses into commercial terms, and the gap means less than it would on a product that touches individuals.

None of the AI capabilities names a model, family, version or provider. Agensee and the generative data assistant are described by what they do, as are the automated explainability and the data quality assistants, and none by what they run on. There is no statement of whether the models are built in house or licensed, no version policy and no per capability breakdown. Amazon Web Services is named repeatedly, but as deployment and procurement infrastructure rather than as a model provider. Naming the cloud is not naming the model. A model provider with Opensee as a customer may have published the relationship even though the vendor has not.

Designed as an overlay rather than a replacement, which the company states plainly: a cross risk analytics layer that sits on top of existing infrastructure, aligning identifiers and reference data and preserving history across systems that were never built to be queried together. That positioning requires ingesting from market risk, credit risk, liquidity, collateral and trade systems simultaneously, which is real integration breadth.

Native Python integration lets a customer build capital and other models against the same data the end users see, and deployment is available through a customer's existing cloud account. No specific risk engine, core banking platform or market data provider is named as a certified integration, so the depth is described in the abstract rather than enumerated.

The position is well disclosed, and it is architectural rather than a policy statement. The platform runs on any cloud, on premise or hybrid cluster. The company states that a customer accesses analytics without moving their data, a substantive answer to where processing happens rather than a promise about it. Procurement and deployment are available through the customer's own cloud account.

That keeps the workload inside infrastructure the institution already governs and already has a residency position for. No specific regions are enumerated, and no residency commitment is stated in contractual terms. There is no explicit treatment of the EU requirements a European bank buyer would be assessed against.

Pricing is not published. There are no plan tiers, unit basis or data volume bands, and no seat cost or indicative contract size. Every route into the product is a demo request. The platform is available for procurement and deployment through a customer's existing Amazon Web Services account. Cloud marketplace listings routinely carry contract pricing and structured commercial terms that vendors usually keep off their own sites. That listing may carry pricing, and it is not reflected here.

Coverage is broad across function and buyer type, and scale is unproven. The platform addresses market risk, credit risk, liquidity and asset liability management. It also covers trade analytics, collateral management, profit and loss and performance, market data storage and climate risk. It is sold to banks, asset managers and hedge funds, with separately presented buy side solutions.

Named institutions span the range, from Credit Agricole CIB on the sell side to Taula Capital on the buy side. No customer count, assets under service figure or geographic footprint is published. The company is a Series A business with modest disclosed funding, so the installed base cannot be characterized from public material.