Keenable Raises $26M—but Its 100B-Page Index Has No Named Customers

Keenable emerged from stealth on August 25, 2026, with a web-search service for AI systems and a $26 million seed round led by Accel, with Conviction Partners and business angels participating. The same launch account found that several AI labs and inference providers use the API in production during training and runtime, but Keenable did not disclose their identities.

Keenable’s official product page lists more than 100 billion documents, latency below 250 milliseconds at the 95th percentile in US East and pricing from $1 per 1,000 requests for dedicated capacity at 100 requests per second or more. Those figures describe a large, commercially accessible retrieval system, but they are vendor disclosures rather than independently reproduced measurements; the title’s “100B-page” shorthand refers to the document count, not an audited count of distinct and current web pages.

What Keenable launched and who funded it

Keenable provides search and content-retrieval APIs backed by an independently maintained web index. Its role is below the model layer: the service retrieves source material that an AI application can process during training or while responding to a live request, rather than operating as a consumer search engine or generating the final answer itself.

SiliconANGLE’s launch coverage identifies Brightwing Capital and scOp Venture Capital among the investors alongside Accel and Conviction Partners. It also describes natural-language search, cleaned-content retrieval and historical point-in-time queries, with market mapping, price monitoring and lead enrichment presented as target uses.

The product therefore combines three elements that are often sold separately: a web corpus, query infrastructure and page-content retrieval. That distinction matters when comparing Keenable with search APIs that return only ranked links, or with crawling tools that fetch pages after another service has selected them.

The scale, price and performance remain company disclosures

The available fact sheet is unusually direct about the size of the corpus and the entry price for high-volume access. It is less complete on the evidence needed to judge how those specifications behave across different workloads: no independent crawl audit, reproducible latency study or neutral freshness assessment has been published with the launch.

Index size alone does not establish useful coverage. A document total can include historical versions, near-duplicates and material that is stale or rarely relevant; developers evaluating retrieval quality would also need to know how quickly pages are discovered, how often existing records are refreshed and how the ranking performs across representative queries.

The advertised lowest rate also has a throughput condition and is tied to dedicated capacity. It is not a universal pay-as-you-go price, so a direct comparison with another search API would need to align traffic volume, hosting model, retrieved content and any additional fetching required after search.

The latency figure carries similar limits. A regional percentile is more informative than an unspecified average, but it does not reveal performance from other locations, under sustained customer load or for requests requiring different retrieval operations. Until an outside party reproduces the result, it should be read as an operating specification supplied by the vendor.

Why the index is designed for machine-scale retrieval

Keenable is betting that AI systems create a different search workload from people. A person normally scans a short result list and opens a small selection of pages; an agent can make repeated queries within one task, retrieve full source material and search again as it encounters missing information.

That pattern changes the economics of retrieval. Query charges multiply across automated workflows, while delays accumulate when later searches depend on earlier results. Low per-request cost and predictable response times consequently become infrastructure concerns rather than conveniences at the interface layer.

An independent index also gives Keenable control over crawling, ranking and the format returned to models. The trade-off is operational: maintaining broad coverage requires continuous crawling, storage and query-serving capacity, and the launch does not disclose enough data to compare those processes with established search providers.

Machine-oriented retrieval does not automatically imply better results. Agents may consume more documents than people, but they still depend on relevance, freshness and source quality. A larger corpus can widen the available evidence while also increasing the importance of ranking and deduplication.

Production use is claimed, but adoption cannot yet be examined

The unresolved question is whether Keenable’s technical footprint has translated into referenceable commercial adoption. The disclosure of production use is more specific than a statement of customer interest, yet unnamed accounts do not allow outsiders to assess deployment scale, contract value, retention or how central the API is to a customer’s product.

“Production” can cover materially different arrangements. An integration may serve a narrow feature or a limited share of traffic, while another may supply retrieval throughout a model’s training and runtime workflow. Customer names, workload descriptions or independently documented deployments would clarify which kind of adoption Keenable has achieved.

The same evidence gap applies to performance. The company has disclosed a large index, a conditional high-volume price and a regional latency target, but not independently measured freshness, relevance or sustained-load results. These omissions do not disprove the specifications; they limit what prospective users and competitors can verify from the launch.

Keenable therefore enters the market with substantial seed financing and a clearly defined product for machine-driven web retrieval. What remains missing is public proof connecting that scale to identifiable customers and independently reproducible operating results—the evidence needed to distinguish a technically ambitious launch from established adoption.

Also read:

• Blog Link Building Still Works—But Its Old Shortcuts Now Count as Spam
• SEO in AI Search: The Fundamentals Still Win, but the Metrics Changed
• Business Blogs Still Matter—But Search Traffic Can’t Be the Only Goal