September 2025 Summaries

Exa has introduced exa-code, a web-scale context tool designed to enhance coding agents by providing highly relevant and dense web context, specifically optimized for handling libraries, APIs, and SDKs, thereby reducing hallucinations in code generation tasks by large language models (LLMs). This tool leverages Exa's AI-focused search engine to efficiently extract code examples from over a billion webpages, prioritizing examples that maximize recall and quality to deliver concise and informative results. Evaluations have demonstrated that exa-code is effective at preventing hallucinations, as it minimizes the token count required while maintaining high relevance, and it is available for free to the public on platforms like Smithery. Exa aims to address the knowledge constraints faced by LLMs, thereby allowing them to generate code with greater accuracy and competency.

Exa, a modern AI-oriented search engine, recently secured $85 million in Series B funding led by Benchmark, valuing the company at $700 million, and aims to redefine how AI systems access web-based information. Founded in 2021, Exa was developed to surpass traditional search engines like Google by offering a search platform specifically designed for AI applications, providing high-quality, ad-free results, complete page content, and customizable search functions. With features like zero data retention and a focus on high-compute, comprehensive searches, Exa is positioned to become the go-to search engine for AI systems, which require more precise and rapid access to information than human users. The company plans to use the new funding to expand its GPU capacity, enhance its indexing and processing capabilities, and grow its team to further develop a search engine that perfectly organizes the world's information, thereby entering the next stage of its strategic growth.