AI Search Company Keenable Raises $26 Million Seed Round Led by Accel
Keenable, a search infrastructure company focused on AI agents, has completed a $26 million seed round led by Accel, with participation from Conviction Partners and several angel investors. The company also disclosed that it has built an independent web index of over 10 billion documents.
The founders are Andrey Styskin, who previously managed Yandex's search, AI, and cloud businesses, and Matthias Petri, a German scientist from Amazon's general AI retrieval direction. The team consists of about 15 engineers distributed across the U.S. and Europe, with plans to double in size by the end of 2026. The funding is expected to close in November 2025, with a stealth product launch.
The product is not a user-facing interface with ten blue links, but a retrieval interface for model training and runtime calls. The founders told TechCrunch that the chatbot's anchored responses using source documents will create a feedback loop different from Google's learning from human clicks; if the index structure is not fine-tuned for tasks, the cost of scanning the entire web is too high. The company claims its API is already in use in production environments by several labs and inference service providers, and it is collaborating with voice company Gradium for real-time retrieval, though the client list has not been disclosed.
Pricing is disclosed at approximately $4 per thousand requests for on-demand usage, about $1 for dedicated capacity, with East Coast latency at the 95th percentile reported to be under 250 milliseconds, along with historical point retrieval. The web search interface and AI-focused query language are live and free to use until the end of September. Reports from Silicon Valley also mention Brightwing Capital, ScOp Venture Capital, and angels from Google and Amazon.
Buyers are labs seeking independent indexing and unwilling to hand over all agent queries to consumer-grade search engines, while sellers are startups turning crawling and inverted indexing into infrastructure. Funds flow from venture capital into data centers and labeling, then back based on usage. Beneficiaries are new stacks that can sell retrieval separately from chat products; under pressure are old search ad models designed around human pages. This is a stealth seed round, not a disclosed revenue scale.
Source: Public Information
ABAB AI Insight
Search advertising has been priced based on human attention for twenty years, while agents are priced based on call frequency and document coverage. Styskin is transferring Yandex-level indexing experience to a market where 'machines need to read entire pages', acknowledging that Google's feedback loop optimizes clicks, not training corpora or runtime references. 10 billion documents is a threshold: without a proprietary index, agents can only rent summaries from others. Accel is betting on a new index as the logic of search monetization loosens, potentially yielding ad space to APIs.
The capital path involves a large seed round to establish crawling and data centers, then spreading costs to labs based on every thousand requests. The motivation is to become the default retrieval layer for training and inference before consumer-grade search captures agent traffic. The difference from Perplexity, Exa, and Tavily is said to be serving machines rather than serving reading; what truly differentiates is latency, freshness, and the ability to retrieve web pages at historical points.
Similar to how Elastic approached logs, Snowflake approached data warehousing, and Neeva attempted independent search before merging into data clouds, the industry is in a phase of infrastructure expansion: large models will call upon it but will not build a second web index. Whoever controls the query language occupies a layer in the agent operating system.
The essence is a transfer of pricing power. Web pages were once priced through ad auctions, now they are priced based on the hit rate of agent retrieval. Humans see ten results, while machines require citable full texts. The index has transformed from a media entry point into production materials, and the seed round is purchasing the first invoice for these production materials.
ABAB News · Cognitive Laws
Searches for humans cannot serve agents
Once indexing is independent, retrieval can be sold separately
If the feedback loop changes users, ad space is no longer central