Keenable Exits Stealth With $26M to Index the Web for AI Agents — Why the ‘Ten Blue Links’ Era Is Ending

Introduction: A Search Index Built for Readers That Never Tire

Search engines have spent three decades optimized for a peculiar constraint: humans who can’t spare the time or attention to scan entire webpages. In 2026, the fastest-growing class of web reader doesn’t have that problem. AI agents and chatbots can — and increasingly do — read everything, and that shift is quietly rebuilding the internet’s search infrastructure from the inside.

The latest evidence landed today: Keenable, a startup founded by Andrey Styskin — who previously led search, AI, and cloud at Russian search giant Yandex, and worked on web search infrastructure for Alexa at Amazon — and German AI scientist Matthias Petri, has exited stealth mode with a $26 million seed round led by Accel, with participation from Conviction Partners and business angels.

Their thesis is simple and a little jarring: the web needs a second search index, one built for AI agents rather than for people. “This actually creates a new flywheel that is different from what Google learned from human behavior,” Styskin told TechCrunch — grounded on the observation that AI chatbots perform dramatically better when they can ground their responses in source documents.

What Keenable Built: A 100-Billion-Document Index for Machines

Keenable says it has already built a web search index of more than 100 billion documents, and claims its API is running in production at several AI labs and inference providers — during both training and runtime. The company won’t name its customers, but it recently announced a partnership with voice AI company Gradium to power live information retrieval for spoken interactions.

The technical moat, according to Styskin’s 20 years building search at Yandex and Amazon, is cost. Scanning the whole internet for every query is “enormously” expensive at web scale — enterprise search tools that work fine inside a company break down when pointed at the entire web. The fix is innovating on how fast you can narrow the search space for a given query, with index structures fine-tuned for machine-style retrieval rather than human-style browsing.

As for what it costs to build and maintain an index of that size: “Don’t ask — it is painfully expensive,” Styskin said. The 15-person engineering team, spread across the U.S. and Europe, plans to double its headcount by the end of the year to build out its go-to-market motion.

Why Now: Google and Microsoft Are Closing the Search-API Doors

The timing is not accidental. Accel partner Zhenya Loginov, who led the investment, points out that AI companies now have very few options for web-scale search infrastructure — because Google and Microsoft have been taking steps to shut down their existing search APIs, opting for bundled approaches and selective partnerships instead of feeding potential competitors.

Meanwhile, the demand side keeps growing. Cloudflare data cited by the founders shows AI crawlers responsible for a steadily growing share of search volume — bots reading the web on behalf of models, agents, and answer engines. That is the same undercurrent behind Cloudflare’s Kitesurf giving agents a remote browser and DNS-AID’s attempt to make agents discoverable by domain name: the web is being re-instrumented, layer by layer, for machine visitors.

Styskin argues the innovator’s dilemma makes Google “beatable” specifically on agentic queries — it is “extremely hard” to move people away from Google for human search, but a smaller company can out-innovate a giant that must protect its core business on the queries agents ask.

WebQueryLanguage: Answers No Single Page Contains

The most technically interesting item on Keenable’s roadmap is an upcoming product called WebQueryLanguage — a way for AI systems to answer questions by combining information from multiple web sources, even in cases where no single page contains the full answer.

That’s a meaningful step beyond retrieval. Today’s search APIs return documents; the model does the synthesis. A query language that orchestrates multi-source synthesis at the index layer would move a chunk of that reasoning into the infrastructure itself — closer to what enterprises now expect from RAG pipelines, but at the scale of the open web. For anyone building agents that research, compare, or fact-check, it’s the difference between handing an intern a stack of pages and handing them a research department.

The Competitive Landscape — and the Tools You Can Use Today

Keenable isn’t alone. Brave sells web-scale search via its Brave Search API, Exa built an embeddings-first search engine for AI applications, and Perplexity runs its own retrieval stack to power answers. Google itself is overhauling search for the AI era — and facing its share of backlash for it. If you’re choosing where AI answers come from today, our best AI search engines roundup covers the field.

The more practical question for builders is what a dedicated agent-search layer means for the agent frameworks you already run:

For agent builders, the retrieval layer is becoming a swappable backend decision — exactly the kind of decision frameworks like LangChain were designed to keep flexible. And for people running autonomous agents such as Manus, AutoGPT, AgentGPT, or SuperAGI, better agent-native search directly upgrades the quality of everything those agents do next: research depth, citation quality, hallucination rates, and task completion.

What to Watch Next

Whether Keenable becomes “the next Google for AI agents” or a cautionary tale about index economics, the direction is settled: machine-readable search is now a market of its own. As TechCrunch put it, whether it’s for humans or for agents, the era of the ten blue links may be coming to a close.

Frequently Asked Questions