Particle gives AI agents ears with a 130K-podcast search engine

AI agents are notoriously good at scraping billions of web pages, but the second someone says something out loud into a microphone, those models are completely in the dark.

Particle, the startup founded by former Twitter engineers, is fixing that blind spot. On Wednesday, the team launched Radar, a specialized search engine and API built to transcribe, understand, and index spoken podcast conversations specifically for AI consumption.

Shining a light on 130,000 podcasts

As Particle co-founder and CEO Sara Beykpour put it, most AI search crawlers focus entirely on text. Audio remains a massive black box unless someone actually transcribes and structures it.

Radar has already indexed more than 130,000 podcasts, which Particle claims makes it the largest transcribed audio repository around. That covers Apple Top 200 shows across 135 verticals, with another 20,000 episodes ingested every single day.

Instead of dumping raw transcripts, Radar adds speaker labels and extracts structured entities like companies, executives, brands, and products. It pulls out self-contained audio clips with timestamps so you can jump straight to the exact quote.

Users can set up custom alerts delivered to Slack, email, or webhooks. You can filter down to when specific guests discuss specific topics on high-ranking shows. There is even a podcast ad tracker that monitors where certain brands sponsor shows and how their ad spending trends over time.

The hedge fund arbitrage

The idea started as a side feature inside Particle’s AI newsreader app to surface relevant audio snippets alongside articles. But the team quickly realized the enterprise value of the underlying pipe was infinitely bigger than the consumer reader.

Surprisingly, hedge funds have become Radar’s heaviest API users. For traders, executive chatter on casual two-hour podcasts often contains valuable market alpha long before it reaches financial reports.

AI search services are moving in too. Agent search platform Exa has already partnered with Radar to feed podcast audio data into its own pipelines. Radar offers web access, standard APIs, and direct integration via MCP.

Pricing starts at $29 monthly per user, with an enterprise tier at $399 per month for 20 seats. Particle plans to expand next into YouTube videos and broadcast news clips.

Why it matters

We think pivoting away from consumer news aggregators into audio infrastructure is a brilliant survival move. The web text corpus is exhausted, but hours of unfiltered technical discussions, founder gossip, and market sentiment live exclusively in audio files.

For researchers, market intelligence teams, and automated agents, getting reliable, timestamped audio quotes via a plug-and-play MCP interface saves hundreds of hours of manual scrubbing. At $29 a month, testing it out is an easy decision if you track company news.

However, audio parsing at this scale carries real operational friction. Speech-to-text accuracy degrades rapidly with specialized jargon and overlapping speakers, which can easily hallucinate bad data downstream. There is also the inevitable intellectual property question: as podcast networks realize their subscriber content is fueling wall-to-wall algorithmic trading, licensing battles seem almost guaranteed.