Seltz raises $12.5M to build a search engine that AI agents can actually use

Seltz, a startup rebuilding web search for AI agents, has raised $12.5 million in seed funding. The round was led by European venture firm Speedinvest and global investor B Capital, with participation from the Italian Founders Fund, United Ventures, and Future Back Ventures, the venture arm of Bain and Company. The company was founded in October 2025 and is betting the next big search company is one that nobody reads the results of.

Most AI search products you have heard of are thin wrappers. OpenAI, Perplexity, Anthropic, and Mistral all add an LLM on top of someone else's search index, usually Google, Bing, or Brave. Seltz is going the other way. The company owns the entire stack: its own web crawler, its own search index, its own retrieval models, and its own ranking layer. "The old search methods don't work because they were architected for humans," Seltz founder and CEO Antonio Mallia told Fortune. "The information the agent needs is actually not in the snippet. It's in the body of the web page, it's in things like tables, images, and other forms of representation that can be useful for an LLM or for an agent."

That pitch is the throughline of the new round. AI agents do not browse the way people do. They fire off long, precise, machine-formatted queries. Research agents can send dozens or hundreds of queries in parallel. They need machine-ready answers, not a list of blue links designed to make a human click through. A human can read a snippet and infer the rest. An LLM cannot. Seltz scores individual passages and extracts the specific table, text, or image an agent actually needs, in what Mallia calls context engineering. The system crawls hundreds of millions of pages a day and returns results in under 200 milliseconds.

The dependency mess in agent search

The opportunity is the gap between how search is sold today and how agents consume it. Google's index is the largest in the world because its user base is the largest in the world, with billions of daily users feeding it signal that no other product sees. That scale is a moat, and it is also a constraint. Google is optimized to return a human-friendly top ten. Seltz is optimized to return machine-usable evidence.

The dependency story is messy enough that there is room for a new entrant. In December, Google sued SerpApi, a service that scraped Google results and counted OpenAI among its reported customers, for allegedly circumventing its anti-bot protections. Anthropic and Mistral are reported to lean on Brave's index to power web search in their chatbots. OpenAI and Perplexity have been reported to be building their own crawlers and indexes, but neither has reached the kind of coverage that Google takes for granted. Mallia frames the moment as the same kind of platform shift that the early 2000s were for Google. "The revolution is back again," he said, "this time driven by transformer models and the AI workflows that increasingly do the searching themselves."

The competitive set is real, and it is well funded. Parallel, the AI search startup founded by former Twitter CEO Parag Agrawal, recently raised $100 million at a $2 billion valuation. Exa pulled in $85 million for its AI-native search engine. Tavily, another AI search company, was acquired by AI cloud operator Nebius for up to $400 million earlier this year. Seltz is the smallest of that group by headcount and by capital, and the most direct about what is being built: a search engine whose first customer is an LLM, not a person.

What Seltz is actually shipping

The product surface is narrow on purpose. Seltz crawls the public web, indexes pages at passage granularity, and exposes a retrieval API that returns the specific bit of evidence an agent needs to ground a response. The pitch to enterprise customers is that the retrieval is fast enough to sit inside an agent loop, and that the company owns enough of the index that it can ship features the wrappers cannot, like passage-level attribution and structured output that is honest about which part of which page a given claim came from.

The team is built for that. Mallia did his PhD in computer science at New York University, with a focus on information retrieval. He worked as an applied scientist on Amazon's artificial general intelligence team and as a research scientist at the vector database company Pinecone before founding Seltz. Many on the fifteen-person team hold PhDs in information retrieval, and the rest are veterans of Amazon's AI efforts. The company is incorporated in the United States, works fully remote, and is split between the San Francisco Bay Area and research hubs near the University of Pisa in Italy and the University of Leipzig in Germany. Seltz's advisers and angel investors include executives from Google, Ramp, Cohere, Synthesia, and Databricks, along with academics from information retrieval labs at NYU and the University of Glasgow.

The seed round funds the next stage of that build. Mallia said the capital will go toward continuing to develop the search stack, hiring, and the start of an enterprise sales effort. The company is going after the same customers the wrappers are going after, including developer platforms building research agents, e-commerce companies that need live product data inside an LLM, and enterprise search teams that need a retrieval layer they can audit. The pitch is that buying retrieval from Seltz is closer to buying a database than buying a Google contract, which is the kind of relationship enterprise software buyers prefer.

Seltz versus the rest of the AI search field

The seed number is small against the recent AI search rounds, and that is the point. Parallel raised a $100 million Series A at a $2 billion valuation. Exa is a Series B. Tavily got acquired for up to $400 million. Seltz is at the same starting line those companies were at two years ago, with a smaller team, a focused product, and a thesis that owning the index is the only moat that holds up in a world where every AI lab needs retrieval it can trust. The investor list backs the thesis: Speedinvest has been early on European infrastructure, B Capital has a track record of enterprise AI, Bain's venture arm is plugged into the consulting base that is buying agent platforms, and the Italian Founders Fund plus United Ventures give Seltz deep ties to the European research community the company is hiring from.

The broader context is that AI search is one of the few categories in the 2026 AI stack where the winners have not been picked yet. OpenAI is the default chat surface, Anthropic is the default coding surface, Google is still the default web surface. The agent retrieval layer underneath all of those is fragmented, dependent, and getting expensive. Seltz is betting that the next platform shift in search is not another wrapper but a rebuilt index, and that the customer is going to be a model, not a person. For a look at the wider agent stack this kind of retrieval piece is being added to, the Best AI Coding Agents in 2026 reference page tracks the tools that are now depending on retrieval like this to do real work. The full deal is detailed in Seltz's exclusive with Fortune.