Seltz Raises $12.5 Million Seed Round to Build a New Search Infrastructure Layer for AI Agents

As AI agents become increasingly capable of performing research, executing workflows, and making decisions autonomously, a growing challenge has emerged: the web itself was never designed for machines. While today’s large language models can reason, summarize, and generate content, they still rely heavily on retrieval systems originally built for humans clicking links. Seltz, a startup focused on rebuilding web search infrastructure specifically for AI systems, has announced a $12.5 million Seed round to address that gap. The funding was led by Speedinvest and B Capital, with participation from several venture firms and a group of advisors and angel investors drawn from companies including Amazon, Google, Cohere, Databricks, Ramp, and Synthesia.

Why AI Agents Need a Different Kind of Search Engine

For decades, web search has been optimized around human behavior. Search engines return ranked lists of links, snippets, and advertisements designed to help people navigate websites. AI agents, however, have fundamentally different requirements.

Rather than browsing multiple pages, agents need structured information delivered quickly, with clear source attribution and enough context to support reasoning. The challenge becomes even more significant when agents perform multiple searches during a single task, where latency compounds and retrieval quality directly impacts performance.

Seltz argues that existing AI search solutions often function as wrappers around traditional search engines such as Google, Bing, or Brave. While those systems work well for human users, they were not built to serve as knowledge infrastructure for autonomous software. The company’s approach is to build a retrieval system designed from the ground up for machine consumption rather than human browsing.

Building the Entire Retrieval Stack

Founded by Antonio Mallia, whose background includes research roles at Amazon, Pinecone, and Bloomberg, Seltz has taken the more difficult path of building and operating its own infrastructure stack.

Instead of relying on third-party search providers, the company owns the entire retrieval pipeline, including web crawling, knowledge extraction, indexing, retrieval, and ranking. This level of control allows Seltz to optimize every stage of the process specifically for AI workloads. According to the company, owning the stack also enables tighter control over quality, performance, source verification, and future product development.

The company’s Web Knowledge API provides developers with real-time access to web information that has been processed and structured for use by large language models, retrieval-augmented generation (RAG) systems, and autonomous agents. Rather than simply returning search results, the platform focuses on delivering context-rich information with source references that can be consumed directly by AI systems.

A Growing Market for AI-Native Infrastructure

The funding comes amid increasing interest in foundational AI infrastructure. Much of the recent AI boom has focused on model providers, chips, and agent frameworks, but retrieval infrastructure is emerging as another critical layer of the stack.

One reason is that AI systems are increasingly expected to work with fresh information. Training data quickly becomes outdated, particularly in areas such as news, regulations, pricing, technical documentation, and enterprise data. As a result, real-time retrieval is becoming essential for many production AI applications.

Seltz is positioning itself as a provider of this retrieval layer. Rather than competing directly with consumer search engines, the company is targeting developers, AI-native startups, frontier AI labs, and enterprises building systems that require reliable access to current information. This reflects a broader shift toward specialized infrastructure designed specifically for machine-to-machine interactions rather than human interfaces.

Performance and Benchmarking Become Competitive Advantages

Search quality is notoriously difficult to measure, particularly when evaluating systems intended for AI agents rather than human users. To address this, Seltz recently launched the Dynamic News Search Benchmark (DNSB), a public benchmark designed to evaluate retrieval quality and latency across different AI search providers.

The company reports strong results in both retrieval speed and effectiveness. Earlier product updates highlighted retrieval latency measured in hundreds of milliseconds, a significant factor for agents that may perform multiple searches during complex reasoning tasks. Faster retrieval not only improves user experience but can also reduce compute costs by shortening the amount of time models spend waiting for external information.

The emphasis on benchmarks reflects a broader trend across AI infrastructure, where measurable performance metrics are becoming increasingly important as enterprises evaluate competing platforms.

Expanding Beyond Search Results

While web search remains the company’s core focus, Seltz has already begun expanding the capabilities of its platform. Recent product releases include publication-date filtering, domain filtering controls, integrations, and tools designed to help developers manage the quality and freshness of retrieved information. The company has also released features aimed at making real-time web knowledge more accessible within AI workflows and agent frameworks.

These additions suggest that Seltz sees retrieval not as a standalone feature, but as a broader knowledge infrastructure layer that can be integrated across enterprise systems, applications, and autonomous workflows.

The Broader Implications of AI-Native Retrieval Infrastructure

Technologies like Seltz point toward a broader evolution in how AI systems interact with the internet. Traditional search engines were designed around human workflows, where users evaluate links, compare sources, and determine relevance themselves. AI agents, by contrast, require retrieval systems that can rapidly surface trustworthy information in formats optimized for machine reasoning and automation.

The development of dedicated retrieval infrastructure could also influence the next generation of AI applications. Faster access to fresh information may improve the performance of agentic systems operating in areas such as enterprise intelligence, financial analysis, scientific research, cybersecurity, and real-time decision support. At the same time, greater emphasis on source attribution and verification could help address ongoing concerns around hallucinations, outdated information, and AI transparency.

As organizations deploy increasingly autonomous systems, the retrieval layer may become a critical component of the AI stack, sitting alongside models, vector databases, and orchestration frameworks. The companies building these foundational technologies are helping establish the infrastructure that could support more reliable, accountable, and information-aware AI systems in the years ahead.