Caudena

Tracing, demixing, fingerprinting, labeling and the rules engine carry the product, and they would still run with the language models removed. The AI drafts on top of them: case summaries, investigation graphs and suspicious activity report narratives, plus Prism MCP, which lets an outside agent drive Caudena's forensic tools and write up what it finds.

AI here drafts rather than decides. Summaries, investigation graphs and suspicious activity report narratives are generated for an officer or investigator to use, and Prism MCP lets an investigator's agent run risk checks, exposure analysis and tracing on its own before returning a narrative.

Caudena says investigators remain responsible for reviewing the agents' conclusions, but what a person must approve before a narrative is filed, and any confidence measure on an agent's conclusion, are not described. Monitoring rules are set by the compliance team without code.

An October 2026 benchmark of Caudena's investigation agents publishes its method and results. Eleven model configurations made 69 scored attempts between September 25 and 28 on reference cases covering Bitcoin to Ethereum tracing, onward Bitcoin payments and TRON network analysis. Each attempt was scored on graph recall, graph precision and whether the report carried the required findings, within 25 minutes and $3 of model usage.

Gemini 3.8 Flash at medium effort scored 100 in all 15 of its tests, averaging 3 minutes 40 seconds at an estimated $0.32 an investigation. Caudena states the limits itself: a top score means the automated checks passed, not that every claim in a report was supported, and nothing compares the agents with human investigators. Agent briefings cite the transaction hashes behind them. Accuracy figures for demixing, fingerprinting and risk scoring are not given, and the testing is not tied to a supervisory model risk framework such as SR 11-7.

Customers go unnamed. Caudena's own estimates for the compliance workspace are that a team needs four officers instead of five, saving two to three hours a day per officer and lifting return on investment by 35 percent, with no exchange or bank attached to the figures. It says it works with leading law enforcement agencies, and its investigation data is pitched as admissible in court.

Whether client case data, or the questions agents send through Prism MCP, are kept apart between customers or used to improve Caudena's models is not stated. Prism MCP connects Caudena's tools to the language model an investigator chooses, such as Claude, so that model sees whatever forensic output the investigator requests.

A data processing agreement, subprocessor list or terms for the case data and client information compliance teams load are not published. Caudena's privacy policy covers its services in general terms: personal data is used for service delivery, legal obligations and consented marketing, shared with service providers for IT, data storage, analytics and marketing, and kept only as long as needed, with transfers outside the European Economic Area acknowledged.

Caudena publishes no security attestation, certification or trust center, and its privacy policy describes security measures only in general terms. Access runs through its hosted applications, with case management built for several users working on the same investigation.

Caudena names no regulator, license or supervisory recognition of its own. Its compliance tools are framed around anti money laundering and suspicious activity reporting duties for financial institutions and for regulated virtual and crypto asset service providers, and its investigation data is presented as admissible in court.

Risk labels on addresses and entities, and the narratives drafted about them, come with no stated rate at which a label is wrong and nothing on how labeling errors fall across chains or regions. The October 2026 agent benchmark checks whether reports carry the required findings on reference cases, which is a quality test rather than a fairness one. Narratives produced through Prism MCP cite the transaction hashes behind each conclusion, so a reader can check what supports it.

Caudena publishes no guarantee or accuracy commitment for its platform, and its terms of service disclaim warranties on its materials and limit liability. What a user can check is the evidence: briefings cite the transaction hashes behind them, so a conclusion can be traced back to the chain and challenged.

The model behind Caudena's summaries, narratives and agents in production is not named, nor is a hosting provider. Its agents run on BeamWeaver, its own open source agent framework, which works with models from six providers, OpenAI and Anthropic among them, and its October 2026 benchmark compared 11 configurations, including Gemini 3.8 Flash and Claude Opus 5.5, without saying which one customers get.

Its intelligence rests on its own off chain open source intelligence and a proprietary dataset of more than 400 terabytes, and Prism MCP hands analysis to whichever language model an investigator connects, such as Claude.

Integration with a bank's transaction monitoring, case management or core systems is not named. Delivery runs through Caudena's own applications, Prism for investigations and a compliance workspace with case management and real time multi user collaboration, and through Prism MCP, which connects its tools to any agent that supports the Model Context Protocol with a single address.

The hosting provider, region and residency of client data are not stated, and no private or on premises option is described. Caudena runs as hosted applications, Prism and its compliance workspace, with a protocol endpoint that outside agents can call.

Caudena publishes no price, tier or basis of charge, and demos, training and courses are arranged through a request form. A calculator on the compliance product estimates savings from staffing rather than the cost of the software.

Each buyer group Caudena names has its own product and page. Law enforcement gets Prism and an investigations course, financial institutions get transaction monitoring and suspicious activity reporting, and centralized exchanges and other virtual and crypto asset service providers get case management and entity risk profiling. Decentralized finance has a page of its own too, with cross chain visibility, risk alerts set to a team's own compliance thresholds and counterparty due diligence. Coverage spans the major chains and privacy coins including Monero and Dash, though no geographic footprint or institution count is stated.