Reflexivity

Agents interpret a financial question and return an analysis, a knowledge graph infers relationships between companies, themes and events that no feed states, and document intelligence reads filings and transcripts at volume. Licensed data feeds with nothing to do with them are what remain once the models go.

Analysis is the limit of its authority, with no execution or order routing, and the system says explicitly when data is unavailable or uncertain. Research produced by agents is not described as reviewed before it circulates inside a firm or reaches a broker's clients.

Code written and executed by the agents makes an analysis reproducible and inspectable, and a full audit trail records provenance and method. Accuracy, evaluation and model risk framework mapping are unpublished, and the no hallucination claim is a design aim no generative system can guarantee.

Interactive Brokers' Investment Themes tool, launched in July 2025, runs on the knowledge graph, and Interactive Brokers, MUFG and General Catalyst appear among backers. No institutional client is named with a result.

Client data gets no persistent storage, clients are fully isolated, and agents run on dedicated private infrastructure air gapped from public AI models and consumer services. Queries therefore neither train a shared model nor pass to a public model provider.

A fund's queries, screens and scenarios reveal its positioning, yet the privacy policy, dated January 2022, covers the website and the earlier Toggle consumer product rather than the institutional platform. Homepage commitments on handling are not carried into published processing terms.

SOC 2 Type 2 with annual third party audits is stated, with controls built to meet institutional due diligence questionnaires. Report request paths and trust centers are not published, and the audit period is not given.

Research distributed to institutions sits near rules on substantiation and conflicts, and the same analysis surfaced to retail clients through a broker falls under fair and balanced communication standards. Neither rule set, nor any supervisor, is addressed.

Many investors querying similar models on the same licensed data can end up with correlated positioning, a crowding risk the company is named after. Whether outputs differ between users, whether crowding is monitored and how coverage skews toward larger issuers are unpublished.

Every insight traces to its source with a data quality rating, so a doubting user can follow a conclusion back to the data. A faulty step found later in published research triggers no guarantee, accuracy commitment or correction process.

Data suppliers are named source by source and licensed by Reflexivity, while the model half is described rather than named. Model, hosting provider and subprocessor list all remain unpublished.

S&P Global, LSEG and Cboe are among the data sources bundled upstream, and downstream the analysis runs inside Microsoft 365 Copilot and through an API with white label options. Both ends are named and in production.

Custom model deployment is offered through the API route, but no hosting provider, region or residency commitment is named. The 2022 privacy policy lets data sit in the United States, the European Union and other destinations, so an institution with location rules cannot pin down where queries run.

Platform Access for investment teams and API Integration for technology teams both run through sales, with no seat, query or platform fee published. Premium market data comes bundled with no separate data contracts, which shows what a buyer could stop licensing elsewhere but not what it would pay.

Asset managers, hedge funds and bank investment teams make up the buyer set, with coverage of more than 40,000 securities across 47 markets. Its scope is research and analysis, stopping short of execution or operations.