Keenable is rebuilding web search for AI agents

Keenable has emerged from stealth with a ³ and $26 million in seed funding led by Accel. The startup says it has indexed more than 100 billion documents and already supplies search infrastructure to AI labs and inference providers during model training and live operation.^{1 2}

The premise is that an AI agent does not search like a person. People want a ranked page they can scan quickly. Agents may need broader document access, structured retrieval and repeated queries while completing one task. If that distinction holds, the search layer beneath AI products could become a meaningful infrastructure market rather than a feature owned entirely by existing search engines.

Traditional search engines optimize heavily for human attention. They rank pages, compress results into snippets and use interaction signals such as clicks. An agent gathering evidence for a multi‑step job may instead need to retrieve many documents, preserve provenance and combine facts that no single page contains.

Keenable co‑founder Andrey Styskin told TechCrunch that grounding chatbot answers in source documents creates a different feedback loop from the one Google learned from human behavior. That is a thesis, not yet independently benchmarked proof. The product will have to show that a specialized index improves answer quality or cost on real agent workloads.¹

The company says its index contains more than 100 billion documents. At that scale, storage is only one expense. The system must crawl changes, remove duplicates, interpret document structure, build searchable representations and answer queries quickly enough for an agent that may make many calls per task.

Styskin said generic enterprise‑search structures can become prohibitively expensive at web scale. Keenable's claimed technical advantage is narrowing the search space quickly for a specific query. The practical buyer question is whether that specialization lowers the total cost of a reliable answer after query fees, model tokens and retries are included.¹

Styskin previously led search, AI and cloud work at Yandex and later worked at Amazon. Co‑founder Matthias Petri is an AI scientist who also worked on search infrastructure for applications including Alexa, according to TechCrunch. Their experience is relevant because building a web index demands distributed systems, ranking, language understanding and constant operations rather than a thin wrapper around another provider's API.¹

Experience does not guarantee distribution. Search infrastructure improves with scale and usage, while buyers resist adding another critical dependency. Keenable must persuade AI companies that its index is sufficiently better, cheaper or more controllable to justify integration.

Accel led the seed round, with participation from Conviction Partners and angel investors, according to TechCrunch. Keenable has about 15 engineers across the United States and Europe and plans to double headcount by year‑end while building a commercial operation.¹

That funding is substantial for a seed‑stage company but modest relative to the cost of crawling and serving the open web. The company therefore needs disciplined infrastructure choices and paying workloads. Its undisclosed production customers make current adoption difficult to verify independently.

Keenable is also developing WebQueryLanguage, a product intended to answer questions by combining information across sources when no individual document contains the full answer. That moves the company from document retrieval toward orchestration and synthesis. The distinction is important for buyers. Retrieval can return sources for a model to interpret; synthesis makes stronger claims about how evidence should be joined. The latter needs careful handling of conflicts, dates and provenance. A useful system should expose which source supports each part of an answer rather than hide uncertainty behind fluent prose.

Keenable competes with search specialists such as Brave and Exa, while Google and Microsoft retain enormous indexes and distribution. Cloud providers can also bundle retrieval with model hosting. TechCrunch notes that large search companies have reduced or constrained some API access as they protect their own AI products, which may create room for an independent supplier.¹

The market will not be decided by index size alone. Coverage freshness, latency, relevance, rights controls, source transparency and unit economics all matter. Developers should run a representative evaluation set and measure answer quality, citation accuracy and total cost rather than choose on a document‑count claim.

The strongest evidence would be published benchmarks on fresh and long‑tail queries, clear pricing, documentation for provenance and named production customers willing to discuss results. Buyers should also ask how the crawler handles publisher controls, removals and corrections. Keenable's timing is credible: agents need dependable access to changing information, and existing search products were not designed around autonomous task loops. The $26 million round gives the company time to prove that a purpose‑built index can deliver a measurable advantage. Until those measurements are public, the 100‑billion‑document figure is scale evidence, not a quality verdict.