Most compliance automation assumes the systems have to be connected before any work can start.
Sphinx was built around a different assumption: the way into a case is the same way an analyst gets in. A login.
Browser agents are how we deliver on that. They sign into the tools a team already runs and work cases inside them. Same screens, same clicks, same case notes.
Why a browser
A compliance stack is rarely one system. It is several, tied together: a third-party screening provider, a transaction monitoring tool, a case manager, and a layer of tools built in-house to hold it all together. Each was added when it was needed, and the connections between them are complicated and messy.
Some of those systems have an API. Many do not. So integration is partial by default, and the data that moves between systems arrives incomplete, reshaped, or late. Compliance is rarely engineering's top priority, so the work to fix that sits in a backlog. The team keeps working the queue by hand.
Bad data means bad decisions. An agent that only sees what an API exposes is deciding on a fraction of the case. So Sphinx operates the way a team of analysts does: through the browser. The agent sees every system the analyst sees, with the same data in front of it, and needs no build to get there.
A login is the integration
A new analyst does not wait on a platform migration or a middleware project. They get a user account. Browser agents are onboarded the same way: a username and password in screening, monitoring, or case management. Where a provider offers an API, that connection can be used instead. Where it doesn't, the browser is enough.
How a case gets worked
Inside Sphinx, the platforms already connected natively are visible. ComplyAdvantage is one. Sphinx plugs in, picks up alerts as they land, analyzes each hit, and writes its findings back where the team already reviews them.
On every case, the agent gathers evidence from the connected systems and from external research, then applies the institution's procedures. Clear cases get a disposition. Anything that isn't clear is escalated with the work already done: the evidence, the sources, the reasoning, and a drafted RFI.
The analyst stays in their own tool. A Chrome extension sits on top of it with a suggested decision, the reasoning, the supporting data, and the draft ready to send. Which queues the agent handles, and how far it goes before a person steps in, is the institution's decision.
What this looks like in production
Conduit gave Sphinx a username and password for its transaction monitoring tools and cleared a six-month backlog in two days. Morse pointed Sphinx at its screening alerts. Sanctions hits now resolve at roughly 99.9% precision, and the financial crime team gets about 200 hours a month back. Neither team migrated a platform or pulled engineering in to build an integration.
Designed for scrutiny
Every case keeps a record: the screens visited, the data pulled, the sources checked, the procedure applied, and the recommendation made. The analyst sees that record in the extension before they act, and it stays on the case. When an examiner asks how a decision was made six months ago, that file is the answer. Browser agents inherit the Interpretable Agentic Framework.
.png)
