Search infrastructure built for human browsing may not be the right foundation for software agents that can inspect far more material before producing an answer. Keenable is betting on that difference with a web index and retrieval system designed specifically for AI applications.
The startup has emerged from stealth with a $26 million seed round led by Accel and Conviction Partners, alongside several angel investors. Its founders are Andrey Styskin, who previously led search, AI and cloud work at Yandex and also worked at Amazon, and Matthias Petri.
A web index built for machine readers
Keenable says its index already covers more than 100 billion web documents. Rather than presenting ranked links to a person, the company exposes retrieval through an API so an AI system can gather information during training or while answering a request.
That distinction shapes both the scale and format of the product. A person usually inspects only a small results page. An agent can process a much larger set of documents, compare sources and extract details before returning a compact response. Keenable therefore describes its system as search infrastructure for machines rather than another consumer search page.
The API is already in production at several unnamed AI laboratories and inference providers, according to the company. Keenable has also announced a partnership with voice AI startup Gradium, where live retrieval can supply fresh information during spoken interactions. Those deployments show intended uses, but the source does not provide customer benchmarks or disclose the laboratories involved.
Because the customers remain unnamed, readers cannot yet compare how retrieval behaves across training, text generation and latency-sensitive voice applications. Each setting may reward a different balance between breadth, speed and freshness.
Retrieval becomes part of an agent’s runtime
Keenable is developing proprietary retrieval technology and plans a Web Query Language that can combine information from multiple sources. The aim is to let developers express richer information needs than a short keyword query and receive material structured for downstream model processing.
The timing reflects a broader constraint identified by the company’s backers. An Accel investor said Google and Microsoft had closed or restricted search APIs that developers once used, leaving AI companies with fewer independent ways to retrieve the public web. That is the investor’s account of the market, not an independently demonstrated measure of access or performance.
Keenable is not alone. TechCrunch identifies Brave and Exa among companies building retrieval products for AI. The competition means a large index is only one part of the test. Developers will also care about freshness, relevance, latency, source coverage and the cost of processing enough documents to improve an answer.
Funding a costly infrastructure ambition
Operating a web-scale index requires substantial computing, storage and crawling capacity. Keenable’s seed round is unusually large for an early-stage company, reflecting the expense of building infrastructure before a mass-market application exists. The company currently employs 15 engineers across the United States and Europe and plans to double headcount by the end of the year.
Styskin has framed the ambition as becoming the next Google for AI agents. That comparison conveys the size of the opportunity Keenable sees, but it remains a goal rather than an established position. The company has not published comparative evaluations showing that its retrieval improves agent accuracy, nor has it disclosed pricing or revenue.
The most consequential idea is that search may split into different products for people and machines. Human interfaces optimize for quick judgment and a manageable list. Agent infrastructure must optimize for programmatic access, broad evidence gathering and predictable integration. Keenable now has capital and an initial index to test that proposition; its undisclosed production deployments will need to turn into measurable results if it is to justify the scale of its claim.
