Why Flow exists.

Most AI automation assumes it may read everything. In regulated work, that assumption is the entire problem.

Flow was founded by Sali Sapan (CEO) and Tomer Steinmetz (CTO).

The founding team

01

Sali Sapan

CEO & Co-Founder

• LL.B. and B.A. in Government, Strategy & Political Science, Reichman University; M.Sc. in Data Science & Machine Learning, Bar-Ilan University
• Reserve officer, elite IDF special forces unit
• Attorney at Fischer Behar Chen (FBC), one of Israel's leading law firms
• Entrepreneur with prior ventures in hightech and real estate

02

Tomer Steinmetz

CTO & Co-Founder

• B.Sc. in Computer Science, Reichman University
• AI Engineer at Amazon, building and shipping applied AI systems at scale
• Previously Team Lead at Hypernative, a cybersecurity and Web3 threat-prevention startup
• 3+ years as a backend engineer at Skai and CorrActions, shipping large-scale, Kafka-based data infrastructure

The problem we started from

A law firm, a hospital, a bank, and a utility all have the same complaint about AI tooling: the demo is impressive, and the deployment is impossible. The blocker is rarely capability. It is that the workflow requires handing privileged, protected, or export-controlled material to a system whose data handling cannot be fully accounted for — and the person signing off on that has personal exposure if it goes wrong.

So the work stays manual. Not because automation would not help, but because no one can defend the deployment.

What we built instead

Flow starts from the deployment constraint rather than treating it as a later integration problem:

• Where processing happens is a choice, not a default. Flow runs in a masked secure cloud today. Running entirely inside your own infrastructure with Flow-Edge is on the roadmap and scoped per engagement, because for some organizations it will be the only option.
• Sensitive fields are masked before processing, not filtered out of logs afterward.
• Every write is reversible, because "we think it did the right thing" is not an answer to an auditor. A full append-only audit trail is on the roadmap for regulated deployments; it is not a live feature today.
• High-risk actions wait for a human. Autonomy is bounded by policy you set.

The technical detail is on the security page.

How we write

We publish our research openly — thirty articles on deploying AI under HIPAA, GDPR, FCRA, 21 CFR Part 11, NERC CIP, CCPA, export controls, and the professional-responsibility rules that govern legal work. Much of it explains where automation is a bad idea, including for us. We would rather be useful to someone who does not buy from us than be persuasive to someone who should not have bought.

Where we are

Flow is in pre-release and working with a small number of early design partners in regulated industries. If your workflows are blocked by exactly the constraint above, that is the conversation we want to have.