Radar podcast search: your agents are blind to audio
Particle launched Radar, a podcast search API and MCP server over 130,000 shows. The lesson for operators: agents can only use media somebody indexed first.
Particle launched Radar on August 26 — a podcast search API and MCP server sitting on top of 130,000-plus transcribed shows. The product is interesting. The premise underneath it is the part worth stealing: your agents cannot use anything nobody transcribed, and most small businesses are sitting on hours of it.
What actually happened
Per TechCrunch, Particle — an AI newsreader startup founded by ex-Twitter engineers and led by CEO Sara Beykpour — shipped Radar as a search engine over podcast audio it says is the largest transcribed index of its kind. It covers 130,000+ podcasts, adds about 20,000 episodes a day, and spans the Apple Top 200 across 135 verticals.
The consumer app is not the business. The API and the MCP server are. They expose transcripts with speaker labels, extracted entities (people, companies, brands, products), self-contained clips with timestamps, ad and sponsorship data, and alerts that fire to email, Slack or a webhook when a term gets mentioned. Pricing runs $29/month per seat, $399/month for a 20-seat business plan, and custom pricing on the API.
The customer list tells you what it is really for: hedge funds are the highest-volume direct API integrators, alongside AI search platforms like Exa and data resellers. Beykpour's framing was that agents are "generally blind to audio."
Why the index layer matters for your business
Radar solves this for public podcasts. Nobody is solving it for yours.
Run through what your business generates every week that no agent can touch. Sales calls on Zoom. Support calls in a phone system that keeps recordings for 90 days and does nothing with them. Voicemails. The Loom your ops lead recorded explaining the returns process, which is now the only documentation that exists. All of it is knowledge, none of it is queryable, and pointing an LLM at it does nothing because there is no index to point at.
The build order is the same one Radar used, and it is not exotic. Transcribe with speaker labels. Extract entities and timestamps so a result can cite where it came from. Put it behind an API or an MCP server so your agents reach it the same way they reach everything else. Then the question "what did we promise that customer in June?" has an answer that does not require a human scrubbing through a recording.
The alerting piece is the underrated half. Knowing a competitor got named on a call last Tuesday beats searching for it next quarter.
Key takeaways
- Radar launched August 26 with 130,000+ transcribed podcasts, adding ~20,000 episodes daily
- The real product is a search API plus an MCP server, not the consumer app
- Exposes speaker labels, entity extraction, timestamped clips, ad data and webhook alerts
- Pricing: $29/month per seat, $399/month for 20 seats, custom API pricing
- Hedge funds are the highest-volume API customers, alongside AI search platforms and data resellers
- The transferable pattern: transcribe, extract entities and timestamps, expose over MCP
Your call recordings are dead weight until something indexes them. We build transcription-to-MCP pipelines over your own audio, so your agents can cite the call instead of guessing. See what we've shipped or tell us where your recordings pile up.
Sources: TechCrunch.
- #mcp
- #ai-agents
- #search
- #unstructured-data
- #automation
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
GLM-5.3-Flash: MIT open weights that read your documents
Z.ai shipped GLM-5.3-Flash under MIT — 320B total, 18B active, native vision, 1M context. The first cheap open-weight model aimed at document and screenshot work.
Read itW4 Games raises $18M: open core beats a royalty on your stack
Tencent led an $18M Series B into W4 Games, which sells enterprise support for MIT-licensed Godot. Buy the support, not the license — the pattern generalizes.
Read it