Particle, the AI newsreader startup created by former Twitter engineers, is abandoning its original mission to pursue a different opportunity: making the conversations hidden inside podcasts searchable and accessible to artificial intelligence agents. On Wednesday, the company unveiled Radar, a podcast search engine that not only converts audio into text but also interprets the content's meaning, allowing it to surface key quotes and notable moments. The shift has already drawn commercial interest from hedge funds seeking data their AI systems can't currently access, according to Particle co-founder and CEO Sara Beykpour.
Radar transcribes more than 130,000 podcasts, making it the largest transcribed podcast service in operation. The index covers all of Apple's Top 200 podcasts spanning its 135 categories, with 20,000 episodes added to the database every day. The transcriptions include speaker identification and detailed metadata, as the system recognizes entities such as individuals, corporations, brands, products, and subjects being discussed. The service can monitor mentions of these entities across all podcasts and dispatch notifications when they appear, either immediately or bundled as daily or weekly summaries delivered through email, Slack, or webhook. Users can apply filters to customize alerts, configuring the system to notify them only when specific guests appear and address a particular subject, for example.
"Hedge funds have been the highest-volume customers that are directly integrating with the API," Beykpour told TechCrunch. While journalists and researchers could benefit from the tools, other significant paying customers have included AI search platforms and data resellers, with the search API provider for AI agents, Exa, numbered among Radar's partners. The concept originated from one of the Particle news-reading app's most popular features, which had used its API to find interesting podcast clips that it then displayed alongside related news stories in the app's feed. According to Beykpour, "agents are generally blind to audio; they can't see it unless something or someone has transcribed it."
The company explains that the product's real value lies not in its web interface but in its API and MCP, which lets AI agents and other businesses tap into the same intelligence through code. Radar can pull out relevant self-contained clips with timestamps, enabling users to both listen to and read the remarks made, with the system pre-selecting notable clips for users who don't want to consume entire episodes or read summaries. The service also tracks which topics were mentioned in podcasts, who or what was referenced and when, listener ratings and reviews, the episode's advertisements, and more, including a dedicated podcast ads search engine that can locate every episode where a particular company advertises and monitor how it changes over time. Radar is priced at $29 monthly per seat, with a $399-per-month plan for businesses that includes 20 seats, while API users face custom pricing based on their requirements. In the future, the company plans to broaden the service beyond podcasts to handle other audio formats, such as YouTube videos and news clips. The move reflects a strategic bet that podcast intelligence represents untapped territory in the AI agent economy, where text-based crawlers dominate but audio remains largely invisible. For enterprises deciding whether to build or buy transcription capabilities, the tradeoff hinges on whether speed to market justifies dependence on a third-party index in a space where proprietary data access could become a competitive moat.