About

We started at the boring end.

Most agent platforms are built for the demo. They are designed to look extraordinary in a forty minute meeting, and the questions that come afterwards get answered later, if at all: who approved this, what did it read, what happens when it is wrong, where does the data sit.

For the organisations we work with, those questions are the project.

A local authority cannot deploy an assistant that occasionally invents a bylaw. A bank cannot put a credit memo in front of a regulator and explain that the model was having an off day. A ministry cannot send resident data to a jurisdiction it does not control. None of that is an edge case for a future release. It is the first constraint, and we built the product backwards from it.

Who we work with

Liba Systems puts AI to work in real processes for local authorities and other public bodies, utilities, and companies in document-heavy fields such as energy and real estate, with people approving the decisions that matter. Our current projects are run in Hebrew, with the teams that do the work.

What we bring

• Process mapping with the people who run the process, based on real cases rather than a description of how things should work.
• A platform for governed workflows: a visual builder, knowledge with citations, approval steps, versioning and a complete record of every run.
• Implementation on your existing systems, such as ERP and CRM systems, project boards, spreadsheet models and public data sources, with your systems kept as the system of record.
• Governance design: where people approve, how exceptions are routed, how rules and thresholds are versioned, and what is recorded.
• Testing, training and handover, so that responsibility for running the workflow is clear.

What that means in practice

Every step is saved. What the system read, what it did and who approved it. When someone asks how a decision was made eighteen months ago, you can look it up instead of starting an investigation.

A human can always be in the way. Approval gates go anywhere in a flow, and when the agent is unsure the decision goes to a named reviewer rather than out of the door.

Sovereignty costs nothing extra. Every capability is available in a deployment that never leaves your perimeter, including disconnected networks running open models on your own hardware. Nothing is held back for the cloud tier.

Language came first. The product was built in the languages our users actually work in, including right to left ones, rather than translated into them afterwards. Accessibility follows WCAG 2.1 AA.

How we work

We would rather deliver one agent that survives contact with your auditors than fifty that never leave the pilot.

So we start with one process, usually the worst one: the one with the backlog, the paper and the person who has been doing it by hand for eleven years. Before building, we agree a written scope with you: what the workflow must do, which systems it touches, and how it will be accepted. Our engineers then build it together with your team, on your systems, with your governance in place before anything reaches production.

Then we step back and the capability stays with you. We train your own builders so departments can ship their own agents while central IT keeps oversight. If your team can never work without us, we have done the job badly.

Who leads Liba

Or Dadush, co-founder and CEO, leads product and technology. He holds a degree in computer science with honours from the Hebrew University of Jerusalem, specialising in AI systems. He is a security researcher and entrepreneur who has built products backed by leading investors, including Samsung.

Bear Sasson Mahlev, co-founder and COO, leads operations and implementation with clients. He brings a background in running operations at technology companies.

Liba Systems was founded in 2026 and is based in Tel Aviv. Its registered company name is Liba AI Ltd. We work with organisations from local authorities to enterprises.

What we do not claim

We do not publish client names without permission, results we have not measured, or certifications we do not hold.