Astelia

Meet Astelia

Astelia is an AI-native cybersecurity company whose exposure management platform helps large organizations identify and eliminate reachable vulnerabilities with viable attack paths into their environments. By correlating network context with vulnerability data, Astelia narrows tens of millions of findings down to the ~1% attackers can actually reach.

As AI enables attackers to exploit newly disclosed vulnerabilities faster, Astelia uses agentic AI to reduce the time between disclosure and defensive action. Delivering that capability depends on continuous access to current vulnerability information.

Ori Shahar, Astelia’s AI Team Lead, and his team needed a reliable way to find, retrieve, and retain the web data that Astelia's models analyze for customers.

The challenge: Turning fragmented vulnerability data into reliable answers

Answering questions about a vulnerability requires combining public web information, third-party data, and details from the customer’s environment. Shahar's team uses these sources to determine whether a vulnerability is relevant, if a proof-of-concept (PoC) applies to a specific configuration, and which remediation or mitigation is appropriate.

To produce consistent, grounded answers, Shahar and team needed control over the information supplied to Astelia's models. This would allow the team to analyze the same data with multiple models, assess confidence in the findings, and apply deterministic analysis.

Rather than build and maintain scrapers for thousands of changing sources, Shahar and team went looking for a reliable way to retrieve and retain relevant web content.

Why Tavily: Reliable data with greater control

Tavily was the first API Shahar's team evaluated. Its broad source coverage and structured results allowed the team to gather information across diverse cybersecurity technologies without building custom scrapers. They could also retain and distill the data before sending it to an LLM, giving them control over context size, model selection, and downstream analysis.

The solution: Continuously updating vulnerability intelligence

Shahar's team placed Tavily at the beginning of Astelia's data collection pipelines.  What started with gathering information about newly published vulnerabilities expanded to collecting and analyzing PoCs, remediations, and mitigations.

Because information about a vulnerability is often incomplete when its ID first appears, Astelia’s pipelines use Tavily to check for newly released information at regular intervals. The team can then validate findings against the retained source data, apply deterministic analysis, and continually improve the reliability of guidance that Astelia provides customers.

The outcome: Vulnerability analysis at the speed of disclosure

With Tavily, Shahar’s team can continuously label an exploding amount of newly disclosed vulnerabilities relevant to enterprises all over the world. Today, the pipeline has achieved:

• Approximately 350,000 vulnerabilities labeled
• Around 200 new vulnerabilities processed on a typical day, increasing to as many as 1,000 on peak days, like Microsoft Patch Tuesday
• An estimated 20 hours of scraper work saved each month, including 10 hours adding new scrapers and 10 maintaining hundreds of site-specific scrapers

For each vulnerability, the pipeline gathers information about it, available patches, remediation guidance, and mitigation options. When details are incomplete, the pipeline continues checking for new information as it becomes available. This is especially important when a PoC exploit is released on the same day as a vulnerability, giving attackers an immediate path to target exposed assets.

What’s next: Building the next layer of vulnerability intelligence

Shahar and his team continue to build backend pipelines that answer new questions about vulnerabilities and customer environments.

As that work expands, Tavily will remain a data source wherever the team needs to retain and control the information its models analyze.

Give your security workflows a clearer view of the web. Talk to our team.