motif

An empty interface is all that would survive without the models. Clarity, the engine, turns sources into dated facts and causal links in a memory graph, analyst agents write cited updates and portfolio reads from it, and profiling agents turn conversation into an investor profile and strategy.

Scheduled publication waits on a deterministic check that every citation resolves and an output check that rejects directive buy and sell labels. Proposals move from draft to proposed to approved before an approved strategy can execute on chain, under fee caps, retry limits and expiry, with every change in an audit log.

Answers are graded by a calibrated judge, a separate judge model and a reviewer model, and changes must clear gates of judge score 70, citation grounding 70 percent and reviewer sign off. Dated results are listed, including a private markets desk that does not yet pass, though no independent party has tested the regime.

A closed founding partner cohort of ten platforms, onboarded one a month since Clarity launched in May 2026, has no named member and no outcome figure. Backing from Liminal, a founders' studio backed by Temasek, and more than 60 combined years of wealth industry experience are credentials, not deployments.

Market updates, company reads and the memory graph are shared by every portfolio, while client portfolios and profiles are scoped to each organization. Whether one platform's portfolio, profile or conversation data enters the shared graph is not stated, and the same system serves competing wealth apps.

Written in April 2025 for the website and a consumer app, the privacy policy covers wallet addresses and social accounts rather than the platform a wealth business embeds. The deploying platform is left to set consent and retention rules, and no data processing agreement, subprocessor list or retention period for investor profiles is published.

Encryption, access controls and incident response are described in general terms, alongside API keys, roles and audit logs. Attestations, certifications and a trust center are all unpublished for a platform that sits inside a regulated firm's client app.

Profiling is built to support MiFID II suitability and appropriateness, GDPR consent and EU AI Act transparency, with inputs, calculations and outputs logged and versioned. No license of its own is stated, leaving the advice relationship with the deploying platform.

Inferring risk tolerance and objectives from conversation rather than a form ties profile accuracy to how articulate and financially literate a person is. A persona evaluation track is described, but the firm publishes no results or accuracy analysis across client groups.

Every answer cites its sources, so a disputed statement can be traced, and the dated quality figures, failures included, are falsifiable. Guarantees, indemnities and a correction route for a client who acted on a faulty proposal are unpublished.

All ten analyst agent profiles default to GPT 5.6 Luna through OpenRouter, evaluation graders run on DeepSeek V4 Flash, and a platform can pin other models per agent. Perigon, Tiingo and CoinMarketCap are named data sources, while the hosting provider and any retention or training terms with the model providers are not.

Developer documentation covers an API reference, an SDK, authentication and rate limits, and holdings snapshots are expected from a portfolio system such as Chelmer or a custodian. Production integration with a named custodian or core platform is not demonstrated, and the snapshot API takes no transaction ledger.

The admin panel, app, API and embed run on the firm's own hosted domains with no provider, region or private deployment option named. Storage outside a user's country is allowed by the privacy policy, so a buyer cannot tell where client portfolios and investor profiles rest.

Usage cost is published: an AI cost ledger records every model call as the basis of what a customer pays, with dated measurements of about $0.017 per chat answer and $0.12 per portfolio read. Platform fees stay unpublished, and founding partners get locked rates.

Financial institutions, neobanks and wealth managers are among six named buyer types, and integration guides cover non US listed holdings, private funds and non standard crypto. Functions run from market insight through profiling to proposals and on chain execution, with automated rebalancing not yet available.