What makes Samsung invest in a new Israeli startup by Amazon alumni?
Israeli startup Xpander, founded by former AWS employees, has raised $7.5 million. The company has developed a platform for managing AI agents, providing security and flexibility without vendor lock-in.
The entrepreneurs of Xpander | Source: Xpander
Quite a few organizations today are integrating AI agents to streamline processes and save time, but they quickly discover a real headache: lack of control over the agents, security issues, and lock-in within the closed garden of a single model. A new Israeli startup by Amazon alumni was revealed today (Monday) with a solution that promises to return power and flexibility to the users' hands, and it has raised $7.5 million for this purpose.
Bringing expertise from Amazon to AI
The Israeli startup Xpander has developed a platform for managing agents for organizations, ranging from managing permissions and connecting to necessary systems to monitoring them to ensure everything is running properly. "At the core is what we call Universal Harness: a vendor-agnostic execution layer that runs any agent from any vendor in a centralized environment that the organization holds," explains David Toyzer, CEO and co-founder of Xpander, in a conversation with Geektime.
Toyzer explains that Xpander's product is a Kubernetes environment that runs within the client's existing infrastructure, in their private cloud, on-prem, or in the startup's cloud. "Because the agents and their linked sandboxes run as containers in the client's Kubernetes, the platform meets their existing security requirements instead of setting up a new infrastructure and putting it through another round of approvals," he added.
Security and Control
At Xpander, they place significant emphasis on security, specifically on the issue of identities. When an agent accesses the organization's systems, it must identify itself to the model and the edge systems as the user themselves with their specific permissions, rather than through a shared key. "With us, it works end-to-end, and login details are kept in a vault and injected only at the moment of calling the tool, so the model never sees the secret," explains Toyzer. Every run takes place in an isolated environment, and every action is fully documented — including approvals, failures, and cost per task.
Above all this, Xpander runs Omni — a "super-agent" that operates the infrastructure. It helps every person in the organization build and operate agents within the permissions granted by the admin: it builds the agent, connects connectors, defines memory and timings, tests, launches, and then continues to operate: reading logs, fixing failed runs, and choosing the right model for each task according to cost and quality. Omni was also built to run inside popular tools like ChatGPT and Claude via an official connector, allowing employees to continue working in the tools they love while the organization controls the connectors and exposed data.
What is Xpander's secret sauce?
Toyzer: "Our secret sauce is that we built the only platform whose entire business model is built on the absence of any lock-in — and we proved that you don't pay for this freedom in performance. Our competitors sell you a layer you can't leave; we sell the layer thanks to which you are never stuck. This is critical because the market is changing, and a company that committed to one model lab can suffer greatly in a short time frame when a new Open-weight model reaches frontier performance and they aren't able to work with it."
Xpander was founded by David Toyzer (CEO), Ran Sheinberg (CPO), and Muriel Fahima (CTO). Toyzer and Sheinberg are alumni of AWS — Amazon's cloud division — and encountered the challenges they now aim to solve during their time there. The Seed funding round of $7.5 million was led by Pico Venture Partners and included Samsung Next, Emerge Ventures, and SeedIL. The startup has 6 employees at its development centers in Israel and the United States.
