Our Seed Investment in Keenable: Search Infrastructure for better AI Agents

Ask an AI agent a question, and it will confidently answer it. The answer will seem great, but if you know the topic deeply, you immediately see the gaps. It’s not necessarily wrong, but it’s under-researched, under-reasoned, and just very average. Why is that?

We believe that the gap is search infrastructure. A great answer doesn’t always appear in the first link provided by regular search engines. It requires looking through a plethora of sources, accessing unindexed documents, and cross-referencing — fast and cheap. It’s a manageable task for agents that can and must retrieve information very differently from us; however, search engines built for humans fall short. A new search infrastructure built from the ground up for AI agent needs must emerge — that’s where Keenable fits in.

Agentic Search Infrastructure, Built by Industry Veterans

Keenable aims to build the world's largest independent web index, created from the ground up with agents in mind. The company is founded by Andrey Styskin, former CEO of Yandex Search and a director at Amazon AGI, and Matthias Petri, previously a principal applied scientist at Amazon AGI.

At Yandex, Andrey built a 200-billion-document web index that had to fight and win vs. Google twice — first on desktop and then on mobile. Over 15 years at Yandex, Andrey grew from an engineer to the CEO of Search, eventually running an organization with 7,000+ employees. At Amazon AGI, Styskin and Petri built the search infrastructure for Amazon AI agents and saw firsthand how different the requirements for search are in the AI era.

Adapting Search from the Ground Up

Agents search differently than people do. Owning the full infrastructure stack lets Keenable build retrieval tools around those differences, rather than bolting them on later. Web Query Language, Keenable’s new web-query language for AI, lets agents pull and reason from thousands of live sources at once. Keenable’s real-time search API helps agents deal with queries more quickly—fast enough for voice agents to answer mid-sentence, while making search more economical by reducing prohibitive search costs.

Keenable plans to work with both model builders and inference providers, and the team has already secured commercial contracts with multiple AI labs.

Today, Keenable emerged from stealth, and we announced we’re leading their seed round, along with participation from Conviction and angels from Amazon, ClickHouse, Databricks, Google, Snowflake, and SpaceXAI.

This is a team that has already built search from scratch and run it at scale, and they’re well prepared to reimagine search for the AI era. Welcome to the Accel family, Keenable.