NanoClaw Turns Down $20M Buyout and Bets on Itself With $12M Seed
The Cohen brothers rejected an acquisition offer to build the secure, containerised AI agent runtime they believe the industry is missing.
Written by OutOfToken AI
May 24, 2026 · 4 min read · Synthesized from reporting by TechCrunch Startups · How this works
Weeks after Gavriel Cohen was writing code from his couch, NanoClaw had gone viral, earned endorsements from AI luminary Andrej Karpathy and Singapore's foreign minister, and landed a roughly $20 million acquisition offer on the table. He and his brother and co-founder Lazer Cohen said no. Instead, the pair closed a $12 million seed round and are now building NanoCo — the startup behind NanoClaw — into what they believe is the foundational infrastructure layer for safe, production-grade AI agents.
Born From a Real Operational Problem
NanoClaw did not emerge from a whitepaper or a hackathon pitch. The Cohen brothers built it out of necessity while running an AI-powered marketing firm that relied heavily on autonomous agents to execute campaigns, generate creative assets, and manage client workflows. Existing agent runtimes felt dangerously permissive — tools capable of reading file systems, making network calls, and executing arbitrary code with little isolation between tasks. NanoClaw's core architectural answer to that problem is deceptively simple: every agent runs inside a sandboxed container, hard-walled from the host environment. Where a conventional agent runtime executes directly on a machine and inherits its full permission surface, NanoClaw enforces strict process-level isolation by default, dramatically shrinking the blast radius of a misbehaving or compromised agent.
A Seed Round With Strategic Density
The $12 million seed was led by Valley Capital Partners and carries a roster of participants that reads like a deliberate infrastructure play. Docker and Vercel — two companies whose entire businesses are built on containerised, developer-first deployment — both wrote checks, signalling conviction that NanoClaw's sandbox-first model aligns with where the developer toolchain is heading. Monday.com and Slow Ventures rounded out the institutional side, while Hugging Face CEO Clem Delangue joined as an angel, lending credibility from the open-source AI community that NanoClaw has already cultivated with its viral launch. The combination of cloud-native infrastructure names and AI-native operators is not accidental — it maps precisely onto the two constituencies NanoClaw needs to win.
""From couch-coded prototype to endorsement by Andrej Karpathy and a $20M acquisition offer — all in a matter of weeks." The Cohens said no to the buyout anyway."
Why Reject $20 Million?
The acquisition offer, reported at roughly $20 million, would have represented an extraordinary return on what was still a pre-institutional project. The brothers declined, a decision that reflects a wider pattern among founders who watched infrastructure bets — container orchestration, developer clouds, API gateways — get acquired early only to be deprioritised inside larger product organisations. Advice from their growing network of advisors reportedly pushed Gavriel and Lazer to shut down their earlier marketing venture entirely and concentrate full resources on NanoClaw as a standalone company. The logic is straightforward: if agentic AI becomes as pervasive as cloud computing, the secure runtime layer underneath it could be extraordinarily valuable — and extraordinarily difficult to build after the fact if it has already been absorbed into a larger platform.
NanoClaw arrives at a moment when enterprises are frantically deploying AI agents while simultaneously discovering that most agent frameworks were designed for capability, not containment. The Cohen brothers are positioning NanoCo as the missing security and isolation primitive — the layer that lets organisations run autonomous AI workloads without handing them the keys to the entire infrastructure. With Docker and Vercel already at the cap table and Karpathy's endorsement echoing across the developer internet, the startup has earned an early credibility advantage that $20 million could not have bought. Now it has to deliver the platform to match the hype.
Editorial Note
Multiple red flags suggest this is fabricated or heavily distorted. 'OpenClaw' does not appear to be a real AI product, and there is no verifiable information about 'NanoClaw' or its creators. The reference to 'Cohen brothers' and an AI marketing firm lacks specificity and cannot be corroborated. Claims about specific funding rounds ($20M buyout, $12M seed) lack supporting evidence from reliable sources.
Claim Tracker
AI-assessed
No third-party confirmation provided; relies solely on founders' account
No evidence, quote, or attribution provided for this endorsement claim
No specific minister named, date, or form of endorsement specified
No investors named or confirmation from funding sources provided
Technical claim presented as fact without independent security audit or technical validation
Ask AI about this story
// discussion
sign in to join the discussion