Keenable, a startup founded by former Yandex search executive Andrey Styskin and AI scientist Matthias Petri, came out of stealth with a $26 million seed round led by Accel, with participation from Conviction Partners and several angel investors. The company’s pitch is that web search built for humans does not serve AI systems well, and that agents need their own retrieval layer.
The bet matters because the two companies that dominate general web search, Google and Microsoft, have been closing off their search APIs rather than opening them to AI competitors, according to the TechCrunch report on which this account is based. That leaves a gap for infrastructure vendors willing to sell raw retrieval to labs that do not want to build a crawler and index from scratch.
Keenable puts its index at more than 100 billion documents and claims a handful of AI labs and inference vendors have already folded the API into their training and runtime workflows. That figure and that customer claim come from the company itself. TechCrunch’s report does not name a single lab using the product, and Keenable declined to disclose who its customers are. Readers should treat both numbers as unverified until an independent party confirms them.
Gradium, a voice AI outfit, is the sole customer relationship Keenable will confirm by name, tied to a live information retrieval integration. That is a modest proof point set against an unnamed roster of several labs. Named partnerships tend to surface in company materials precisely because they can be checked, while everything else remains a claim without an auditor.
Styskin’s argument for why this is hard to replicate rests on his background: 20 years building search infrastructure at Yandex and Amazon, including work on Alexa’s retrieval stack with Petri. He says the economics of web-scale search only work if you narrow the query space before scanning, rather than treating every query like an enterprise search problem. He also does not dispute that the underlying cost is severe, telling TechCrunch that building the index has been “painfully expensive.”
Keenable is not alone in this space. Brave and Exa already sell search APIs aimed at AI applications, and Google itself is rebuilding its own search product for an AI-query era. Keenable’s differentiation claim, an index tuned specifically for how AI systems query rather than how humans do, is also the claim its competitors make about themselves. The company has not published a benchmark comparing retrieval quality or cost against those rivals.
The startup’s next product, a “Web Query Language” meant to let AI systems combine partial answers from multiple sources into one response, is not yet shipping. With 15 engineers today and a plan to double headcount by year end, Keenable is still small relative to the ambition Styskin describes: becoming, in his words, “the next Google for AI agents.”
For operators evaluating retrieval vendors for agentic products, the near-term decision is not whether Keenable’s index is real. Index size claims of this kind are common in the category and rarely audited before a company has paying reference customers willing to be named. The more useful question over the next quarter is whether Keenable, Brave, or Exa will publish a named customer and a retrieval-quality benchmark that is not self-reported, since that is what will actually separate the category’s vendors from its pitches.
Based on reporting by Anna Heim for TechCrunch, published August 25, 2026.
