NanoCo says its NanoClaw platform isolates AI agents in separate sandboxes and routes requests for access or action through a human-approval gateway. The Israeli start-up is pitching those controls as a way to let businesses use agents without handing them an unexamined key to company systems.
Watch Desk analysis
What happened
NanoCo’s commercial lead, Lazar Cohen, told The Jerusalem Post that the company grew out of the founders’ experience using OpenClaw and questioning whether agents could be trusted with company data. NanoClaw is the company’s open-source agent platform; NanoCo also sells an enterprise offering with governance and management controls.
Cohen says NanoClaw has a smaller codebase than some alternatives and runs agents in isolated sandboxes. A gateway outside the agent environment evaluates requests, such as accessing information or sending communications, for approval or denial by a human operator. The enterprise platform is deployed in a customer’s cloud environment, according to Cohen, and the company says customers can move agents, stored knowledge and workflows between providers.
Why it matters
The pitch addresses a real operational tension: agents are useful when they can do things, but each new permission can also widen the consequences of a mistake. Sandboxing and approval gates offer a concrete answer to that problem, at least in design. Whether they work as advertised in varied deployments is a separate question.
Cohen told the Post that NanoClaw had attracted hundreds of thousands of users and more than 30,000 GitHub stars. He also said NanoCo raised a $12 million seed round in 2026. Those adoption and funding figures come from the company’s account in the interview, not an independent assessment of security or product performance.
Our read
This is more interesting than another promise to make agents “enterprise-ready”: NanoCo describes specific boundaries around where agents run and which actions need human approval. That is a useful design proposition, not a security verdict. Buyers should ask how permissions are configured, what gets logged and how the controls behave when an agent encounters an unexpected task. The word “sandbox” is reassuring; the implementation is where the story lives.
What to watch
- Whether NanoCo publishes technical details or independent audits of NanoClaw’s isolation and gateway controls.
- How enterprise customers configure approval rules and audit agent actions.
- Whether customers can move workflows between providers as easily as the company says.
Discussion spark: For workplace AI agents, should access to sensitive data and the ability to send messages require human approval by default, or should teams be able to delegate those actions once the system is tested?
Sources and evidence
- Israeli AI start-up NanoCo brings governance and security to enterprise AI agents – The Jerusalem Post (28 October 2026, 07:05 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.