NanoClaw's founders refuse a $20M buyout and raise a $12M seed for secure AI agents

NanoClaw began as a security fix. Gavriel Cohen, a former Wix engineer, wrote its first line of code on January 29, 2026, because the OpenClaw-style agents powering his and brother Lazer's AI marketing agency were powerful but dangerously unguarded. Instead of running the agent with access to every service and credential, he isolated the whole agent inside a container and released the result under the MIT License on January 31.

The project went viral within weeks: AI researcher Andrej Karpathy praised it, and Singapore's Foreign Minister Vivian Balakrishnan called it his 'second brain' in a post that snowballed. The brothers declined a six-figure offer to sell the project and then a roughly $20 million acquisition, shut down their agency, and closed an oversubscribed $12 million seed led by Valley Capital Partners — under six weeks from the first line of code, with Docker, Vercel, Monday.com and Slow Ventures among the backers.

NanoClaw's core logic is intentionally small — about 500 lines of TypeScript versus roughly 400,000 in OpenClaw — so a security team can audit it. Every agent runs in a MicroVM-based Docker sandbox, raw credentials never reach it, and sensitive actions such as sending email or deleting files trigger human approval cards in Slack, Teams or WhatsApp.

By May 2026 NanoClaw had roughly 250,000 downloads and 30,000 GitHub stars, and executives at companies like Amazon, Google, Meta, Gap, SentinelOne and Accenture were using it. NanoCo, a Tel Aviv company of about ten employees, began selling per-agent, per-month managed deployments and 'forward-deployed engineer' implementation services, positioning itself as the first institutionally funded company in the 'claw' space.