Eight months. That's how long it took Antonio Mallia to go from founding Seltz to closing a $12.5 million seed round.
The San Francisco startup announced the financing on June 24, 2026, with Speedinvest and B Capital co-leading. Italian Founders Fund, United Ventures, and Future Back Ventures—Bain & Company's venture arm—also participated, according to a Fortune exclusive that first reported the deal. According to Italian outlets covering the round, additional backers include 2100 Ventures, Vento Ventures, and Mango Capital, though those names didn't surface in primary U.S. coverage.
Compressed timelines have become something of a signature in AI infrastructure lately. But Seltz's particular wager stands out, if only because of what the company isn't doing.
Most AI search startups—and there are many now—essentially wrap existing search APIs from Google or Bing in a layer of machine-learning polish. Seltz built the whole thing. Crawler, index, retrieval models, ranking algorithms. From scratch.
Why does that matter? Perhaps more than the founders expected when they started.
Rethinking Search for Machines
Traditional web search was architected for humans who wanted ten blue links and a snippet of text. Click, read, repeat. Autonomous AI agents need something else entirely: structured data they can parse and act on without a person in the loop. Tables. Images. Specific passages extracted mid-page.
Seltz performs what it calls "context engineering"—pulling those machine-readable elements out at query time. The system crawls hundreds of millions of pages daily, the company says, and Seltz claims it returns results in under 200 milliseconds. It also claims near-real-time freshness for U.S. news, with new stories indexed within roughly an hour of publication, per a June 1 product blog post announcing a Zapier integration.
Those are bold numbers in a field where freshness and latency have historically traded off against each other. Whether Seltz can sustain that performance as it scales—well, that's the infrastructure bet investors are making.
Mallia, the founder and CEO, holds a PhD in information retrieval from NYU, worked as an applied scientist on Amazon's AGI team, and spent time as a research scientist at Pinecone. Before that, search research at Bloomberg. He launched Seltz in October 2025.
The current team numbers around 15 people, roughly half full-time, scattered across the San Francisco Bay Area, Pisa, and Leipzig. Fully remote.
A Crowded, Well-Funded Field
Seltz is hardly alone in reimagining search for the age of autonomous agents. The category has attracted extraordinary capital in recent months, and the competitive landscape is getting dense.
Parallel Web Systems—led by Parag Agrawal, the former Twitter CEO—hit a $2 billion valuation in April 2026 after a $100 million Series B from Sequoia. Exa raised $85 million at a $700 million valuation back in September 2025. Then in February 2026, Nebius agreed to acquire Tavily for $275 million upfront, with another $125 million tied to milestones.
Differentiating these players isn't straightforward, at least not from the outside. The technical approaches vary widely: some own their stacks, others rely on API wrappers. Neural retrieval versus traditional methods. Breadth of coverage versus vertical specialization.
Seltz is positioning on full infrastructure ownership and sub-200ms query latency. The company already has public API documentation, integrations with major AI frameworks, and an MCP server endpoint that went live in December 2025. In April, it appeared on Orthogonal's unified API marketplace for agent skills. Then in June, AWS showcased the startup at VivaTech 2026, highlighting a migration to AWS infrastructure alongside partner Devoteam.
The Go-To-Market Inflection
The $12.5 million will fund further development of the search stack and team expansion. But more immediately, it signals Seltz's first serious push into enterprise sales.
A LinkedIn post from early June advertised for a "Founding Head of GTM"—go-to-market, in startup parlance—a role that marks the shift from pure product development to revenue generation. The company has also assembled a roster of angel investors and advisors from Google, Ramp, Cohere, Synthesia, and Databricks, plus academics from NYU and the University of Glasgow's information retrieval labs.
Whether owning an entire search stack proves economically viable at scale remains an open question. The infrastructure costs are real. The competition is well-capitalized and, in some cases, much further along.
But Seltz's thesis is straightforward enough: the next generation of autonomous agents needs search infrastructure rebuilt from first principles. Not retrofitted from the era when humans typed queries into a box and scanned results on a screen.
For now, at least, they have the capital to find out if that's true.
