Developers building autonomous AI agents currently face a frustrating paradox. While the underlying large language models possess immense reasoning capabilities, their ability to interact with the live web remains tethered to infrastructure designed for humans. Most agents rely on search APIs that return a list of links and snippets, forcing the agent to simulate a human user by clicking, reading, and scraping pages one by one. This process is slow, expensive, and prone to failure, creating a significant bottleneck in the quest for truly autonomous digital workers.

The Infrastructure for a 100 Billion Document Web

Keenable has entered this gap with a $26 million seed funding round led by Accel, with participation from Conviction Partners and several prominent business angels. The company is not building another chatbot or a wrapper around existing models; instead, it is constructing the foundational search index specifically optimized for AI agents. The venture is led by Andrey Styskin, who previously headed the search, AI, and cloud divisions at Yandex, and AI scientist Matthias Petri. Their combined expertise in large-scale retrieval systems is evident in the current scale of the project, as Keenable has already indexed over 100 billion documents.

This index is delivered via an API that is already integrated into the training and runtime pipelines of several AI laboratories and inference providers. By providing a direct path to grounded data, Keenable allows agents to verify facts and retrieve real-time information without the overhead of traditional browser simulation. A primary example of this application is Keenable's recent partnership with Gradium, a voice AI company. In the context of voice interfaces, latency is the primary enemy; the ability to retrieve and synthesize information in milliseconds is the difference between a natural conversation and a disjointed robotic interaction.

To further this capability, Keenable is developing a proprietary tool called WebQueryLanguage. This language is designed to move beyond simple keyword retrieval, allowing AI systems to combine and synthesize information scattered across multiple disparate web sources. When a single source does not contain a complete answer, WebQueryLanguage enables the agent to weave together fragments of data from various origins to construct a comprehensive response. To support this ambitious technical roadmap, the company currently employs 15 engineers across the United States and Europe, with plans to double its headcount by the end of the year.

The Death of the Ten Blue Links

To understand why a dedicated AI index is necessary, one must recognize that the modern web search experience is a relic of human cognitive limitations. Traditional search engines are optimized for a user who lacks the time or attention to read a hundred pages. The result is the classic ten blue links: a curated list of suggestions that a human can skim to find a relevant starting point. However, an AI agent does not skim. An agent can process thousands of words per second and requires raw, structured data to perform grounding—the process of anchoring a model's response in verifiable evidence to eliminate hallucinations.

This shift in the primary consumer of web data has created a strategic opening. For decades, Google and Microsoft have dominated the index market, but they are now facing a classic innovator's dilemma. Their business models are built on the ad-revenue generated by humans clicking those blue links. If they provide an API that allows AI agents to bypass the browser and extract data directly and efficiently, they risk cannibalizing their own traffic and revenue streams. Consequently, the tech giants have begun bundling their search APIs with other services or restricting access to partners, leaving a vacuum for independent infrastructure providers.

Keenable is positioning itself to fill this void by focusing on cost-efficiency and agentic queries. Scanning the entire web is an astronomically expensive operation. The competitive advantage in this space does not come from simply having the data, but from the efficiency of the index structure. By optimizing the search space specifically for the way agents query data—rather than how humans type into a search bar—Keenable aims to provide a more sustainable cost structure for AI companies. While incumbents are slowed by the need to protect their legacy ad ecosystems, Keenable is building a lean, agent-first pipeline that treats the web as a database rather than a directory of pages.

For AI practitioners, this represents a critical shift in the vendor landscape. The dependency on a single Big Tech API for web access has long been a point of failure and a source of unpredictable pricing. The emergence of a specialized indexing layer allows developers to decouple their agent's reasoning capabilities from the search provider's business interests. The real test for this new infrastructure will be the performance of WebQueryLanguage in multi-source synthesis benchmarks, which will determine if agents can finally move past simple retrieval and toward complex, autonomous research.

This transition marks the beginning of a decoupled web, where the index for humans and the index for machines evolve into two entirely different species of infrastructure.