Keenable Raises $26M to Build a 100B-Document Search Index for AI Agents

Keenable emerged from stealth on August 25, 2026 with $26 million in seed funding led by Accel, with Conviction Partners and a group of business angels also participating. The company is building a web search index explicitly architected for AI agents rather than human browsers. As the agentic era reshapes data infrastructure, the assumption that agents can simply bolt onto Google or Bing APIs is collapsing: both Google and Microsoft have been moving to shut down their existing search APIs to avoid cannibalization, opting instead for bundled, partner-selective arrangements that leave independent AI builders without a reliable, cost-efficient path to web-scale retrieval.

Andrey Styskin, Keenable's co-founder, spent 20 years building search infrastructure at Yandex — where he led the search, AI, and cloud division — and later at Amazon, where he and co-founder Matthias Petri worked on web search infrastructure for AI applications including Alexa. His diagnosis of the incumbent problem is structural: enterprise search solutions not purpose-built for web scale become prohibitively expensive the moment a query requires scanning the full index. "If you do not fine-tune your index structures for a specific task, the cost of serving and scanning the whole internet is enormous because of the volume," Styskin told TechCrunch. The solution is narrow-search-space selection — ranking and filtering the relevant index slice per query before any deep scan occurs, a technique that requires index innovations rather than just faster hardware.

Human-optimized search engines learned ranking signals from click behavior and dwell time. Styskin argues that AI agents generate a qualitatively different signal set — grounding responses in source documents rather than selecting a single link — and that this creates "a new flywheel that is different from what Google learned from human behavior." The constraint isn't model speed; it's retrieval design, a distinction that matters for agent architectures that decompose tasks across tool calls.

Keenable's index currently spans more than 100 billion documents. The API is already running in production at several AI labs and inference providers for both training-time and runtime retrieval, though Keenable has not disclosed customer names. The one confirmed partnership is with voice AI company Gradium, which uses Keenable to support live information retrieval.

Beyond the index, Keenable is developing a proprietary retrieval layer called Web Query Language (WQL). The design goal is specific: enable AI systems to synthesize answers by pulling from multiple web sources simultaneously, even when no single source contains a complete answer. This is a materially different retrieval contract than keyword search or vector similarity lookup. Competitive context includes Brave and Exa, both of which have entered the agent-oriented search space, though neither has publicly disclosed an index at the 100-billion-document scale Keenable claims. Google itself is also overhauling its search experience for the AI era.

Keenable currently employs 15 engineers distributed across the U.S. and Europe. The $26 million seed will be used primarily to double that headcount by end of 2026, with expansion focused on go-to-market functions. Styskin is candid that index construction costs are severe — "don't ask — it is painfully expensive" — and that the company is pacing capital deployment carefully against that constraint.

Accel partner Zhenya Loginov's investment thesis — that AI players have very few viable options for web-scale search — reflects a pattern visible across the stack: purpose-built tooling for agent workflows pulling ahead of retrofitted enterprise solutions. Whether Keenable's index scale and WQL retrieval model are sufficient to hold that position against a re-architecting Google remains the central open question for the company's next phase.