Keenable's $26M seed: your AI agent needs its own search index

Every agent you build that answers a question about the outside world — competitor pricing, a supplier's lead time, whether a part is discontinued — is making a search call to somebody. Most operators never ask who. Keenable just raised $26 million to be that somebody, and the raise is a decent excuse to look at the search layer under your own automation.

What actually happened

Keenable came out of stealth on August 25 with a $26 million seed round led by Accel, partner Zhenya Loginov, with Conviction and angel investors participating. The company is building an independent web search index of more than 100 billion documents, tuned for machine consumers rather than people.

The founders are not tourists. Andrey Styskin ran search, AI, and cloud at Yandex and spent roughly two decades on search there and at Amazon. Matthias Petri worked with him on Alexa's web search infrastructure. The team is 15 engineers across the US and Europe, with plans to roughly double by year end. The API is already in production at several AI labs and inference providers, used during both training and runtime. A "Web Query Language" for synthesizing answers across multiple sources is on the roadmap.

The market gap is the interesting part. Google and Microsoft have wound down general-purpose search APIs rather than cannibalize their own results pages. That left agent builders choosing between scraping, Brave, Exa, and a short list of others. Styskin's pitch is cost structure: index the web generically and scanning it is ruinously expensive, so you fine-tune the index for the task.

Why the search layer matters for your business

Here's the failure mode we see. A shop builds a research agent, wires it to whatever search API the framework tutorial used, and ships. Six months later the vendor triples pricing, or deprecates the endpoint, or starts returning results that are quietly worse. The agent gets dumber and nobody can point at a commit that caused it.

Search is a dependency, not a detail. Treat it like your payment processor: know who it is, what it costs per thousand calls, and how fast you could swap it. That means a thin interface in your own code — one search(query) -> results function your agents call — instead of scattering a vendor SDK through forty files.

Keenable being new is not a reason to avoid it. It is a reason to make sure you could leave.

Sources: TechCrunch, Dealroom.