Two Brothers Built a Simpler AI Agent Framework in Six Weeks. The Security Research Says That Simplicity Is the Point.
When Gavriel Cohen sat down on January 29 and wrote the first line of code for what would become NanoClaw, he was not thinking about fundraising. Six weeks later, NanoCo — the company he runs with his brother Lazer — had a signed term sheet for a $12 million seed round. The round, led by Valley Capital Partners with participation from Docker, Vercel, Monday.com, Slow Ventures, Clutch Capital, and Factorial Capital, closed in late May 2026. The brothers had turned down an earlier acquisition offer of roughly $20 million, according to TechCrunch, choosing the equity round instead.
The speed of NanoCo's fundraise is unusual. The reason behind it may be more significant — and it starts with what the open-source agent ecosystem had just learned about itself.
NanoClaw was built as a direct response to OpenClaw, the dominant open-source agent framework that had been downloaded hundreds of thousands of times and was already embedded in the workflows of major enterprises. The two products represent fundamentally different security philosophies. OpenClaw sandboxing individual tools while leaving the agent itself running with broad system access — a design that, according to a security analysis published by the SUCCESS Lab at Texas A&M University, created an attack surface that scaled with the framework's complexity, reaching 470 catalogued advisories and a documented path to unauthenticated remote code execution. NanoClaw's alternative architecture isolates the entire agent — tools and environment — inside a single sandboxed process, a design intended to foreclose tool-level escalation. OpenClaw comprises roughly 3,680 source files and 434,453 lines of code. NanoClaw, by contrast, is 15 files and approximately 3,900 lines, with fewer than 10 dependencies and zero configuration files. NanoClaw's own documentation puts its understandability at eight minutes, compared to an estimated one to two weeks for OpenClaw.
That architectural difference is what attracted institutional capital. Docker and Vercel — both strategic partners in the round — have integrated NanoClaw into their deployment platforms. Monday.com is listed as a participant. Angel investors include Clem Delangue, co-founder of HuggingFace, and Matias Woloski, co-founder of Auth0. The company says it has forward-deployed engineers embedded at customer sites, and that executives at Amazon, Gap, Google, Meta, SentinelOne, and Accenture are using the framework internally. A NanoCo spokesperson declined to name specific internal champions at those companies.
Andrej Karpathy endorsed NanoClaw on social media, describing it as a return to first principles for AI agents. Singapore's Foreign Minister Vivian Balakrishnan called NanoClaw his "second brain", handling research, daily briefings, speech drafting, and information synthesis. He subsequently invited the NanoCo team to Singapore for an AI innovation meeting.
The company reports more than 29,200 GitHub stars and approximately 250,000 open-source downloads since February 2026. It turned down a second acquisition offer of around $20 million that included an employment arrangement, TechCrunch reported. Valley Capital Partners declined to comment for this story.
There is an important caveat. The SUCCESS Lab analysis was filed against the version of OpenClaw that existed before February 2026 — when OpenAI acquired the framework's creator, Peter Steinberger, and committed to moving OpenClaw into a foundation structure. The new OpenClaw, now under OpenAI's stewardship, may have a materially different security posture. OpenAI did not respond to questions about whether it has addressed the advisories identified in the paper. NanoCo says it has not published an independent security audit of NanoClaw itself.
What is clear is that the demand signal arrived before the formal funding process did. Roughly 50 or more founders and technology executives sent direct messages to the Cohens asking to invest before the round was structured. The brothers say they did not have a pitch deck when they started taking meetings.