Keenable raises $26M to build a web search index for AI agents

Keenable.ai exited stealth today with $26 million in seed funding to build a web search index designed specifically for AI agents, not human browsers. For builders running high-frequency agent queries - market mapping, pricing monitoring, lead enrichment - this signals that dedicated agent search infrastructure is becoming a product category worth watching.

Keenable exits stealth with $26M and a 100B-document index

The startup has developed an independent web search index spanning more than 100 billion documents, optimized for the low-latency, high-frequency queries that agents typically run. The seed round was led by Accel with participation from Brightwing Capital, Conviction Partners, and scOp Venture Capital, plus angel investments from Google and Amazon executives. Co-founders Andrey Styskin (former head of search at Yandex) and Matthias Petri (from Amazon AGI) bring deep search and retrieval experience according to SiliconANGLE. The company claims its API already serves several unnamed AI labs in production.

Why dedicated agent search infrastructure matters

Most AI agents today either call traditional web search APIs (built for human queries) or run their own crawlers. Neither approach is ideal for agent workloads that need clean, structured, model-ready data at scale. Keenable's pitch is a task-tuned index that narrows the search space fast based on the query, reducing both latency and cost. As Styskin told TechCrunch, the cost of scanning the whole internet is enormous unless you innovate on index structures for specific tasks TechCrunch coverage. For builders, this could mean fewer infrastructure headaches when building agents that need real-time web data.

What Keenable offers: low latency, historical queries, and a new query language

Keenable provides APIs for natural language searches and cleaned content fetching, designed to deliver "model-ready inputs" with minimal delay. According to the company's website, it achieves under 250ms p95 latency in US East and supports point-in-time historical queries, letting agents search the web as it existed at a specific moment. The startup is also developing Web Query Language, a system that aggregates data from multiple web sources to answer questions even when no single page contains the full answer.

Pricing is set at $1 per 1,000 requests, targeted at frontier-scale users doing 100+ requests per second. That is competitive with some specialized search APIs, but cost will add up fast for agents making thousands of queries per hour.

Uncertainties: pricing, performance, and adoption

Keenable's claims about search quality being superior to Tavily, Exa, and Perplexity are vendor assertions, not independently verified results. The company hasn't named any of its AI lab customers, so independent validation of latency, index freshness, and reliability isn't available. For teams evaluating Keenable, the $1 per 1K pricing is clear, but there are no published SLAs or details on data retention, security, or rate limiting beyond the stated capacity. Builders should trial the API on real workloads before committing to it as a core infrastructure piece.

Overall, Keenable is a credible bet from experienced search engineers, but the agent web search infrastructure category is still early. If you are building agents that depend on fresh, clean web data, Keenable is worth testing, but treat the performance claims as directional until you measure them yourself.