Nebius buys Tavily for agentic search
- Nebius announced it acquired Tavily to integrate real-time web access into its production AI stack, aiming to move agents beyond static demos. - Combined with Nebius' Token Factory for high-performance reasoning, the deal lets agents search, verify, and operate on live information in real time at scale. - This acquisition signals a push to make tool-connected agents use current web data in production stacks. (x.com)
Nebius said on February 10 it had agreed to acquire Tavily, a search startup built for AI agents, and fold its web access tools into Nebius’s AI cloud stack. The companies said Tavily’s search, extraction and crawling tools would sit alongside Nebius Token Factory, the inference platform Nebius has been pitching as the reasoning layer for production agents. (nebius.com) That makes this less about a generic acqui-hire than about owning another layer of the agent stack. Nebius said Tavily gives customers the ability to build agents that can navigate the web, verify facts and execute tasks against live information, rather than relying only on a model’s static training data. Tavily founder Rotem Weiss said the company was built because “AI agents need reliable, real-time access to the web,” and said the web access layer had to be “safe, reliable, and at scale.” (nebius.com) The practical point is where the product sits. Nebius described Token Factory as the high-performance inference layer and Tavily as the grounding layer that supplies current web data for factual accuracy. In a May 20 Nebius blog post, the company said the combined setup is meant to help developers build agents that can “reason, verify information and act on live context” inside the production AI stack. (nebius.com) Nebius has also been signaling that this is part of a broader software buildout, not a one-off deal. In its May 13 Q1 2026 shareholder letter, the company listed Tavily among acquisitions meant to deepen its position in retrieval and agentic workflows, alongside separate moves around inference execution and model optimization. The same letter said Nebius AI cloud revenue rose 841% year over year to $390 million in the quarter, giving context for why it is adding more application-layer tooling on top of compute. (assets.nebius.com) Tavily, for its part, said the product would keep operating under the same brand and that existing customers would see no immediate change in the API or data policies. Nebius said Weiss and the Tavily team would join the company and continue leading development of the product. (nebius.com) The larger industry thread is that agent companies increasingly want bundled infrastructure: model serving, orchestration, retrieval and live web access in one place. Nebius is explicitly making that pitch. Roman Chernin, Nebius’s co-founder and chief business officer, said the company was building “the complete platform” for customers building AI products, agents and services, and said owning the search layer would let developers spend less time stitching together vendors. That is the clearest read on why this deal matters: Nebius is trying to turn web-connected, verifiable agent behavior into a native feature of its cloud platform rather than an external integration. (nebius.com)
