Seltz Raises $12.5 Million to Rebuild Web Search for AI Agents

SAN FRANCISCO, Calif. - Seltz has raised $12.5 million in seed financing to build web-search infrastructure specifically designed for AI agents, signaling a shift in how engineers approach the fundamental architecture of the internet's discovery layers. The funding round was co-led by Speedinvest and B Capital, highlighting significant venture interest in the specialized infrastructure required to support autonomous software. Other participants in the round included Future Present, Italian Founders Fund, United Ventures, Vento Ventures, Mango Capital, 2100 Ventures, and Future Back Ventures, which serves as the venture investment arm of Bain & Company.

Led by Chief Executive Officer Antonio Mallia, a veteran of Amazon’s search engineering efforts, Seltz enters a market currently undergoing a paradigm shift. For three decades, web search has been optimized for human cognition, prioritizing the presentation of a 'ten blue links' interface that requires a person to click, evaluate, and synthesize information manually. Mallia’s thesis is that the rise of large language models and autonomous agents necessitates a complete departure from this human-centric design, moving instead toward a machine-to-machine exchange of indexed information.

The core problem Seltz aims to solve is the inherent friction between legacy search engines and agentic workflows. Conventional search engines are architected for a 'one-and-done' interaction model where a human types a two-or three-word phrase. In contrast, AI agents often submit queries that are significantly longer, more complex, and highly specific. These agents frequently operate in parallel, executing multiple sub-tasks simultaneously to fulfill a single user request. This high-volume, high-complexity traffic puts a unique strain on existing search APIs that were not built with such programmatic intensity in mind.

To address these requirements, Seltz is developing a vertically integrated search stack from the ground up. This includes a proprietary crawler to discover web pages, a specialized index to organize the data, and advanced retrieval models to select and deliver the most relevant results. By owning the entire stack rather than layering their technology on top of existing providers, Seltz claims it can exert greater control over the freshness of its data, the speed of its response times, and the structural format of the output, which is designed to be consumed directly by software rather than rendered for human eyes.

The decision to build an independent crawler and index is a bold move in an industry where indexing the entire web is often considered the most capital-intensive hurdle for any new entrant. The sheer scale of the web makes coverage a critical metric; if an agent misses a primary source because a niche search engine failed to index it, the agent's utility is compromised. Seltz’s strategy suggests a belief that the efficiency gains of a purpose-built architecture for agents will eventually outweigh the massive overhead associated with maintaining a modern web index.

Industry analysts have noted that the competitive landscape for search is currently more volatile than it has been in decades. While Google and Bing command the lion's share of traditional traffic, the rise of search-augmented generation has created an opening for startups to define new standards. Seltz’s focus on structured results—information packaged in formats like JSON that agents can parse immediately—addresses a major pain point for developers who currently spend significant compute and engineering time transforming messy HTML into something an LLM can understand.

As part of its early outreach, Seltz has published benchmark claims regarding its accuracy and latency, including specific performance results for its news index. While these early metrics suggests a technical foundation capable of competing on speed, experts in the field caution that initial benchmarks are only a starting point. Potential enterprise customers and developers will likely require rigorous independent testing across their own unique and varied queries before committing to a new infrastructure provider in a mission-critical capacity.

Search quality is famously difficult to quantify through a single score because it involves multifaceted challenges such as source authority, de-duplication, and the filtering of web spam. Furthermore, the web is a living entity where pages change at different rates; a search engine's ability to provide high-fidelity results depends on its ability to re-crawl volatile pages while maintaining a deep archive of static ones. Seltz will have to demonstrate that its retrieval models can handle these nuances at scale without sacrificing the structured precision that agents require.

The $12.5 million in seed capital is earmarked for several key growth initiatives, including the scaling of its index to covers more of the searchable web and the expansion of its engineering team. Finding a balance between the vastness of the web and the specific information needs of AI agents will be a primary focus as the company moves from its current stage into broader commercial availability. The startup also plans to use the funds to put its product in front of a wider audience of AI developers who are currently building the next generation of enterprise automation tools.

The competitive environment Seltz faces is formidable. It is entering a space occupied not only by incumbent general-purpose search giants but also by a new wave of specialized startups targeting the RAG (Retrieval-Augmented Generation) market. These competitors range from vector database providers to other API-first search libraries. The differentiation for Seltz lies in its belief that the underlying technical requirements for agent traffic are fundamentally distinct enough to justify an entirely new infrastructure stack rather than a wrapper over existing indices.

A significant variable in Seltz's long-term viability will be its cost structure. In a typical agentic workflow, a single complex task might trigger hundreds of individual searches as the agent explores different avenues to reach a goal. For this model to be economically sustainable for developers, the search provider must deliver results at a price point and reliability level that can scale predictably. If the cost per query is too high, or if the results lack the freshness required for time-sensitive tasks, developers may be forced to stick with less specialized but more established providers.

The broader context of this fundraise reflects a growing investor appetite for 'picks and shovels' in the AI era. As the hype around individual consumer-facing bots begins to settle, venture capital is increasingly flowing toward the infrastructure that makes these bots more capable. By targeting the discovery layer, Seltz is positioning itself as a central utility for the autonomous economy, where the primary users of the internet are no longer humans, but the software programs acting on their behalf.

Looking ahead, the enterprise market’s adoption of AI agents will serve as the true test for Seltz’s infrastructure. If agents become the primary way that businesses interact with external data, the demand for high-structured, low-latency search will likely explode. Seltz is betting that its specialized focus will allow it to outpace more general tools in providing the 'fresh' and 'trustworthy' data that these systems demand for accurate reasoning.

As Seltz expands its footprint, the industry will be watching to see if it can maintain its promised performance levels while dealing with the realities of the open web. The ability to distinguish between high-quality data and the noise of AI-generated content on the web will be another hurdle for the company's crawler and index. Despite the challenges, the backing of a diverse group of investors ranging from European funds like Italian Founders Fund to Silicon Valley staples like Mango Capital suggests a global recognition of the problem Mallia and his team are attempting to solve.