Keenable has emerged from stealth with a $26 million seed round and a plan to build web search infrastructure specifically for AI agents. The startup says its index already covers more than 100 billion documents and that its API is being used in production by several AI labs and inference providers.

Accel led the round, with participation from Conviction Partners and business angels. Keenable’s founders are Andrey Styskin, a former leader of Yandex’s search, AI and cloud division, and German AI scientist Matthias Petri. The company is entering a market where developers increasingly need systems that can retrieve, compare and synthesize information for software agents rather than simply return pages for people to read.

Why Keenable is targeting agent search

Traditional search engines were optimized around human behavior: users scan result pages, open a small number of links and decide which information is useful. AI systems have different requirements. An agent may need to retrieve evidence from multiple pages, ground a response in source documents or gather current information while completing a task.

Styskin told TechCrunch that this creates a different feedback loop from the one built around human search behavior. For AI applications, the value of a result is closely tied to whether it supplies reliable context that a model can use, not just whether it attracts a click.

That distinction is becoming more important as chatbots and AI agents move beyond question answering. Systems that conduct research, monitor changing information or support voice interactions need search infrastructure that can handle broad queries quickly and return material in a form that downstream models can process.

Keenable is also developing WebQueryLanguage, a retrieval product intended to combine information from multiple web sources when no single document contains a complete answer. The company has not disclosed when the product will become generally available.

What is confirmed, and what remains unverified

The funding, the company’s launch from stealth and the identities of its investors were reported by TechCrunch. Keenable claims that its index contains more than 100 billion documents and that its API is already used by several AI labs and inference providers in both training and runtime environments. The startup did not identify those customers, so the scale and depth of that adoption cannot be independently assessed from the available evidence.

Keenable has disclosed one partnership: it is working with Gradium, a voice AI company, to support live information retrieval. That relationship points to a practical use case for the service, since voice applications often need current answers without forcing users to wait for a conventional browsing flow. The evidence does not establish the volume of Gradium’s usage or the commercial terms of the partnership.

The company’s technical and cost claims also come primarily from its leadership. Styskin argued that web-scale search requires specialized index structures that can narrow the search space rapidly, because scanning the entire internet for every request would be too expensive. He described the cost of building a large index as extremely high but said Keenable is working to control spending as it scales.

Those statements are plausible engineering constraints rather than independently verified performance results. No latency, price, precision, recall or benchmark figures were provided in the source material. For prospective customers, the unanswered questions include how often the index is refreshed, how Keenable handles source quality and duplication, and how its results compare with existing retrieval services.

The infrastructure bet behind the round

Keenable is not presenting itself as another enterprise search layer. Styskin said enterprise search systems can become prohibitively expensive when extended to the scale of the public web, making index design and query routing central to the company’s approach.

The startup is led by executives with prior experience in large-scale search. Styskin spent about two decades working in search at Yandex and Amazon, while he and Petri previously worked on web search infrastructure for AI applications such as Alexa. That background gives Keenable a technical rationale for concentrating on retrieval rather than building a general-purpose chatbot or foundation model.

The company currently has 15 engineering employees across the United States and Europe. It plans to double its headcount by the end of the year, using the new capital to expand engineering and develop its go-to-market operation. That hiring plan reflects the unusual capital demands of maintaining a web-scale index while also selling infrastructure to technically sophisticated customers.

For AI developers, the proposition is straightforward but difficult to execute: outsource crawling, indexing and retrieval so product teams can focus on the agent’s workflow. The trade-off is dependency. Customers would need confidence that Keenable’s coverage, freshness, uptime and pricing are stable enough to support applications where a bad or outdated source can directly affect the output.

A gap between bundled search and independent infrastructure

Accel partner Zhenya Loginov, who led the investment, said AI companies have limited choices for web-scale search infrastructure. He pointed to moves by Google and Microsoft to restrict or shut down existing search APIs, arguing that the large platforms increasingly favor bundled products and selective partnerships.

That market interpretation helps explain the funding thesis. Search incumbents have strong distribution and enormous infrastructure, but they may not want to provide an open layer that enables other companies to build competing AI experiences. An independent provider could benefit if developers need neutral access to current web information and if the major platforms continue to keep retrieval tied to their own assistants or cloud products.

Keenable is not alone. Brave and Exa are already pursuing AI-oriented search and retrieval, while Google is changing its own search experience for AI use cases. The competitive question is therefore not whether agent search will exist, but whether an independent company can offer enough coverage and reliability at a lower total cost than the bundled alternatives.

For founders and product teams, that means search selection will become an architectural decision. They will need to compare not only result relevance, but also licensing, source attribution, data freshness, regional coverage, response speed and the cost of repeated agent queries. Those variables can materially change the economics of research agents and live information products.

What to watch next

The first signal will be whether Keenable names additional customers or publishes technical details that substantiate its adoption claims. Customer references, API availability and independent measurements of retrieval quality would help distinguish an early infrastructure business from a well-funded prototype.

Pricing and latency will be equally important. Agent workflows can generate many search requests for a single task, so a service that performs well for occasional chatbot queries may still be too expensive for high-volume automation. Developers will also watch whether WebQueryLanguage can reliably merge evidence across sources without increasing hallucination or attribution risks.

Keenable’s expansion from 15 engineers, its relationship with Gradium and any further partnerships with AI labs or inference providers will show whether the company can turn its search expertise into a repeatable distribution model. Competitor moves from Exa, Brave, Google and Microsoft will provide the broader test of demand for independent agent-oriented search.

Creati.ai perspective

Keenable’s launch matters because retrieval is becoming a bottleneck beneath many AI products. Better models do not eliminate the need for current, well-ranked and traceable information, particularly when agents must act on changing web data. A specialized search provider could give builders a useful abstraction layer, but only if it can make quality and cost predictable.

The company’s strongest evidence so far is its founders’ search background, its reported index scale and the new funding from experienced investors. Its most important claims—production adoption, document coverage and eventual cost advantages—remain largely company-reported. The next phase will show whether Keenable can convert that technical thesis into dependable infrastructure that enterprises and AI labs are willing to build around.