Former Yandex search chief Andrei Styskin has launched Keenable — a search index built specifically for AI agents. The project raised $26M from Accel and Conviction.
Why search built for humans doesn't work for agents
The team's starting premise: traditional search was always built for humans, and that format doesn't fit machines. Here's why.
A model knows a lot — until you ask about something you actually understand yourself. The answer usually isn't a lie, it's just averaged out. Weights are compressed, often-outdated memory. Anything fresh, narrow, or factual, the model has to fetch from outside.
That's why LLMs already have search bolted on almost everywhere: ChatGPT, Claude, Perplexity, agents. The question isn't "does it need search." The question is which kind.
Google and most search APIs are built for humans: one query, ten links, you open them yourself, you clean out the noise yourself. Unlike a human, an agent needs:
- dozens of short queries instead of one broad one
- low latency
- cheap repeat calls
- text that's already cleaned up for a model, not raw HTML
The same logic is behind Google's agentic browser built on Gemini 3 — companies are reworking human-facing tools one by one for how agents actually use the web.
Expensive search and Keenable's bet
Search is expensive right now, so agents get trained to economize. They cram everything into one query at once — company, revenue, funding rounds, dates, competitors. That works worse than a series of precise questions. Keenable is built exactly for this: an agent should be able to search as easily as it "recalls" information from its weights.
The company's bet is its own index of 100+ billion documents, an API, and a query language built for agents — not "another Google," but infrastructure that models can use often and cheaply. It's part of a broader trend: search isn't the only thing being rebuilt for AI agents — Google recently unveiled the Universal Commerce Protocol for agent-driven purchases.
What to keep in mind
That's a strong thesis. And an expensive one. Only a handful of players can maintain a full web index — most "search for AI" products on the market are wrappers over someone else's search. If Keenable pulls it off on price, speed, and context cleanliness, agents get noticeably stronger. If not, the market stays on the labs' built-in search and the current APIs.
Announcement: https://x.com/styskin/status/2092265673041084505
Website: https://keenable.ai