Keenable Raises $26M to Index the Web for AI Agents

Keenable came out of stealth on August 25 with $26 million in seed funding and an unfashionable product: a web search index. The bet behind it is that the indexes we already have were built for people typing queries into a box, and that AI agents need something else. Keenable says its index covers more than 100 billion documents, exposed through APIs for language-based search and content retrieval rather than a consumer search page.

The round was led by Accel, with participation from Conviction Partners, Brightwing Capital and ScOp Venture Capital, plus angel investors drawn from Google and Amazon. The founding team has run search infrastructure at scale before: Andrey Styskin previously led the search, AI and cloud division at Russian search company Yandex, then worked at Amazon on web search infrastructure for AI products. Matthias Petri, a German AI scientist, worked alongside him there.

The technical argument turns on how differently the two kinds of user behave. A person issues one query, skims a page of links and stops. An agent working through a task searches, reads what came back, reasons about the gap, and issues more queries — looping until it has enough to act. That converts search from a one-shot lookup into a continuous, high-frequency workload, and per-call search and page-fetch APIs priced for human traffic get expensive very quickly when something is hitting them continuously. Keenable’s pitch is that it can narrow the search space fast enough to serve that pattern, and Styskin argues agents simply behave better when their answers are grounded in retrieved source documents rather than recalled from weights.

Timing helps. Both Google and Microsoft have retired their public search APIs, steering would-be customers toward selective partnerships and bundled offerings instead. That has left a thin market for anyone who needs web-scale retrieval without a bilateral deal with a hyperscaler. Accel’s Zhenya Loginov framed the gap plainly, noting that AI players have very few options when it comes to web-scale search infrastructure. Keenable lands in a field that already includes Exa and Brave, both selling retrieval to model builders.

The company says it already has production customers among AI labs and inference providers, though it has not named them, and it recently partnered with voice AI company Gradium to supply live information retrieval. Typical workloads it points to are market mapping, pricing monitoring and lead enrichment — jobs where an agent needs current facts rather than a ranked list of links. Next on the roadmap is a product Keenable calls Web Query Language, aimed at letting an AI system assemble an answer across several sources when no single page contains the whole thing.

What the round really prices is a shift in what retrieval is for. When the reader is a model rather than a person, ranking ten blue links matters less than returning grounded passages cheaply, thousands of times an hour — and that turns the index into metered infrastructure rather than a destination.