Keenable wants to rebuild web search for AI agents

Introduction

Most search engines were designed around human behavior. People generally scan a small number of results and rarely read an entire page, so ranking, snippets, and click signals have shaped how search infrastructure is built. AI agents change that assumption. They can process far more text, use search as part of a task, and ground an answer in several source documents rather than simply presenting a list of links.

Keenable is building its business around that shift. The startup was founded by Andrey Styskin, who previously led search, AI, and cloud activities at Yandex, and German AI scientist Matthias Petri. It has now exited stealth with a $26 million seed round led by Accel, with participation from Conviction Partners and business angels.

Key points

• A large index for machine use. Keenable says its web index contains more than 100 billion documents. Its API is already being used in production by several AI labs and inference providers during both training and runtime, although the company has not disclosed their names.
• The central problem is retrieval efficiency. At web scale, scanning everything for every query would be prohibitively expensive. Keenable says it is developing index structures and retrieval methods that can narrow the search space quickly according to the task and query.
• A cross-source query layer is planned. The company is working on a product called Web Query Language. It is intended to help AI systems combine facts from multiple web sources when no single document contains a complete answer.
• Live information is an early use case. Keenable has partnered with voice AI company Gradium to support real-time retrieval. The startup currently has 15 engineering employees across the United States and Europe and plans to expand the team with its new funding.

Why it matters

Keenable is entering a field that already includes Brave and Exa, while Google is also redesigning search for the AI era. Its pitch is not simply to add a chatbot to a conventional search page. Instead, it is trying to improve the underlying index, query routing, and cost structure for systems that may issue many searches as they work.

The market context is important. According to Accel partner Zhenya Loginov, Google and Microsoft are taking steps to limit or reshape existing search APIs, partly to avoid undermining their own products. That leaves AI companies with fewer independent options for web-scale retrieval and creates room for specialist infrastructure providers.

The economics remain a major obstacle. Maintaining a huge index, refreshing content, ranking sources, and serving low-latency queries all require substantial resources. Styskin acknowledged that building the index is extremely expensive, so the company is trying to control its pace of expansion. Retrieval quality also involves more than coverage: systems must handle conflicting pages, source reliability, and answers that can be checked.

Keenable may benefit if AI agents make search a continuous backend function rather than an occasional consumer action. But its ambition to become “the next Google for AI agents” will depend on proving customer demand, operating efficiency, and sustained index quality. The broader direction is already clear: web search is being rebuilt not only for people who click links, but also for software that searches, reasons, and acts.

Source: TechCrunch AI