Realset AI and Flatkey, two companies founded by Hunter Guo, have jointly raised a $10 million Series A. Both are based in San Jose, California. They work on two different bottlenecks in AI: getting data that can’t be scraped, and getting access to a crowded field of models without juggling dozens of accounts.
“The easy data is gone,” Guo says. “What is left is the physical world, and you cannot scrape it.”
Realset: data from the physical world
Realset produces training data for frontier models and embodied AI, meaning robots and agents that act in the real world. It has three divisions. Realset Body captures first-person and third-person video of skilled workers doing hands-on manipulation tasks, the raw material robot learning needs. Realset Field builds reinforcement learning environments from real workflows in e-commerce, customer support and logistics. Realset Judge provides expert evaluation and monitoring of AI agents running in production.
In the fourth quarter Realset will launch an Open Household Manipulation Benchmark that evaluates robot policies including π0, OpenVLA, GR00T and Octo. Its workspace runs in six languages.
Flatkey: one key, 100+ models
Flatkey launched in July 2026. It signed up more than 10,000 developers in two months. It gives developers one API key and one bill for more than 100 official models (from OpenAI, Anthropic, Google, DeepSeek, Kimi and GLM, plus video generators like Seedance) and over 1,000 tools, including search, browsers and data enrichment. Prices run 60–90% of official list prices. It’s OpenAI-compatible, and plans start at $10 a month.
“AI development is becoming less about choosing one model and more about combining models, data and tools,” Guo says. Between them, the two companies cover both of those needs.