Keenable Web Index for AI Agents Lands $26M Seed Led by Accel

A search engine whose users are machines

Keenable came out of stealth on August 25, 2026, with a $26 million seed round led by Accel. It is building a web index meant to be read by AI agents, not by people.12 Conviction Partners and a group of angel investors joined the round, and Accel partner Zhenya Loginov led the deal.1 One report says some of the angels came from Amazon, Google and Databricks.6

The company says it has indexed more than 100 billion documents. That number comes from Keenable itself and has not been independently checked.1 Its API is reportedly already in production at several AI labs and inference providers, which use it both to train models and to answer live queries.29 Keenable has not named those customers. It has confirmed a partnership with voice AI company Gradium for live information retrieval.89

The founding team is a big part of the pitch. Co-founder Andrey Styskin previously led search, AI and cloud at Yandex. His co-founder, Matthias Petri, is a German AI scientist.210 Styskin and Petri also worked together on web search infrastructure for AI products, including Amazon's Alexa.8 Reports describe the founders' backgrounds slightly differently. One calls both of them former Yandex search executives5, while another describes them as Yandex and Alexa search engineers.7

What the product is

Keenable sells three things on top of its index. The first is a search API that returns ranked results. The second is an endpoint that fetches a page's full text so a model can quote it directly. The third is a set of developer tools, including a Model Context Protocol (MCP) server that lets coding agents like Claude Code and Cursor query the index.1 The company calls the main use case "grounding": tying a chatbot's answer to a real source document instead of a summary of a summary.2

The product still in development is Web Query Language, which is meant to answer questions by combining information from several pages when no single page has the full answer.19 Keenable also claims its approach makes web search about ten times cheaper for AI workloads. It says this is because the system narrows its search based on what the agent is asking.6 Published pricing puts a million queries at about $4,000 on pay-as-you-go and about $1,000 at the enterprise rate, with the first 100,000 requests free.1

Styskin's argument is short: search engines were built for people who can't read every page, and agents can.4 The practical difference is that a human wants ten links and a snippet. An agent doing market mapping, price monitoring or lead research wants to read dozens of full pages quickly and cheaply.6

Why the timing matters

The deal is about access as much as technology. Loginov says AI developers now have very few options for web-scale search infrastructure. In his account, Google and Microsoft are bundling search into their own AI products and choosing partners carefully instead of offering open APIs.89 Several outlets present Keenable as a way around that problem for AI companies that cannot or will not build their own crawler.69

Keenable is not first to this market, and its round is the smallest among recent deals. Exa raised a $250 million Series C at a $2.2 billion valuation in May 2026. Parallel Web Systems raised a $100 million Series B at a $2 billion valuation in April. Nebius bought Tavily in February for $275 million, potentially rising to $400 million.1 Brave Search and Exa are also selling search APIs to agent builders.7

Coverage disagrees on how far along Keenable is. One analysis describes Accel's investment as a seed-stage bet on infrastructure before the product existed.5 That doesn't fit with reports that the API is already in production at AI labs29, or with the observation that the product was live before the company announced itself.1 The production claims are better supported. Keenable is more accurately described as a working product with a short track record than as a pre-product bet.

What analysts do agree on is the main risk. Keenable's advantage depends on how good and how complete its index is, and both take years and a lot of money to build.5 Fifteen engineers run the index today, and the company plans to roughly double that to 30 by the end of 2026.15 Google's index is maintained by an organization many times larger.1 One assessment says $26 million is enough to prove the idea with early customers, but keeping a competitive index over time will need much larger future rounds.7

The publisher side: clicks are already falling

For publishers, Keenable is part of a bigger shift: more of the web is being read by machines and fewer pages are being visited by people. The evidence on Google traffic has become hard to dispute. Chartbeat's 2026 Publisher Playbook found that Google Search referrals to news and media sites fell 40.2% between July 2025 and July 2026. Search's share of pageviews across its network dropped from 9% in July 2024 to 5% in July 2026.13 Chartbeat attributes the drop mainly to AI-generated summaries, which hit simple, generic articles hardest.13

The strongest causal evidence so far comes from a randomized field experiment with 1,065 desktop Chrome users. When Google showed an AI Overview, clicks from search results to outside websites fell 39.8% and searches ending with no click rose 34.5%. Clicks on sponsored results did not change.18 That last finding matters: the researchers say AI Overviews replace visits to websites specifically, rather than shifting attention around the results page.18

Some widely quoted numbers need context. Ahrefs found a 58% lower click-through rate for the top-ranking page on keywords that trigger an AI Overview, up from 34.5% in its April 2025 study.1718 Critics point out that this measures the top result on affected keywords, not average traffic loss across all sites. Pew's 8%-versus-15% click rate measures something different again.17 All of these studies point the same way, but the numbers can't be directly compared.

Google's own financial results show the effect. In Alphabet's first-quarter 2026 earnings, Google Network revenue, which comes from ads on third-party sites, fell 4% to $6.97 billion, while revenue from Google's own search pages grew 19%.17 In May, Google added five changes to AI Overviews intended to send more traffic to publishers, including a "Further Exploration" links section and subscription labels. Those changes came while Google faced Penske Media's antitrust lawsuit and a European Publishers Council complaint to the European Commission.14

The hope that chatbots would make up the difference hasn't worked out. Chartbeat puts AI referrals at 0.01% of publisher traffic, and that share drifted lower through 2026.13 An earlier Chartbeat release said ChatGPT referrals had more than tripled but still made up less than 1% of referrals.16 The two readings fit together: growth from a tiny base doesn't change the overall picture.

What this means for answer engine optimization

So far, answer engine optimization (AEO) has mostly meant getting cited by the big consumer assistants. Agencies and software tools track citations in ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews.2127 Common advice is to put a short, direct answer at the top of each section, add structured data, and build depth on a topic.2226 Some vendors go further. Scrunch, for example, serves AI crawlers a machine-friendly version of a page while human visitors see the original.27

Indexes like Keenable's add another layer to the problem. If an agent builds its answer from a third-party index, the sources in that index are the ones that get cited, and sites missing from it effectively don't exist for that agent, however well they rank on Google.8 Pages built mainly for human navigation, with many menus and few self-contained factual statements, may also do poorly in retrieval systems.8

The practical conclusion is that discovery is splitting up. A publisher used to optimize for one crawler, Googlebot. Now it also has to consider OpenAI's crawlers, Google's AI features, and a growing set of independent agent indexes such as Keenable, Exa and Parallel, each with its own way of choosing pages. That makes it harder to stay visible, and much harder to measure. One AEO guide notes that AI platforms offer nothing like Search Console's analytics.26

Our read

Keenable is a reasonable infrastructure bet with a clear commercial case. Google and Microsoft are restricting access, AI labs need affordable data to ground their answers, and a team with Yandex search experience has a credible chance of building it. Its success depends mostly on whether its index proves good enough at scale, not on how it markets itself.

The broader point is less comfortable for publishers. Every new agent-focused index makes reading the web more efficient for machines, while traffic data shows those machines send almost no visitors back. Keenable doesn't cause that imbalance, but it is part of a market built around it. Until there is a working way to pay publishers whose pages agents read, being cited by more indexes will add to a site's visibility without adding much revenue.