Radar: Podcasts Become Data for AI Agents

What Has Changed: Agents Now Listen

On August 26, 2026, Particle launched Radar, a podcast search engine that makes audio readable by AI agents. The system transcribes speech and interprets its meaning.

The startup, founded by former Twitter engineers, has redirected its focus toward indexing spoken conversations. Radar extracts key quotes and highlights from episodes.

According to TechCrunch^[1], Radar transcribes over 130,000 podcasts, making it the largest transcription service in existence. The index covers all Apple Top 200 podcasts across 135 verticals, with 20,000 episodes added every day.

The product grew out of a feature in Particle's old news-reading app, which selected relevant podcast clips. The team recognized the value and extracted that capability into a dedicated API.

The Mechanism: Audio Transformed into a Structured Layer

Most agents and API services index the textual web. Radar adds an audio layer to that landscape.

CEO Sara Beykpour describes the problem in technical terms: agents remain blind to audio until something or someone transcribes it. Radar occupies precisely that gap.

Transcripts carry speaker labels and rich metadata. The system recognizes the entities discussed: people, companies, brands, products, and topics. It then tracks mentions across episodes and sends alerts via email, Slack, or webhook. This transforms an unstructured audio stream into a queryable dataset. An agent no longer needs to listen to an episode: it queries a text index with entities already extracted and linked.

Who Pays: Hedge Funds and Research Platforms

Beykpour identifies hedge funds as the highest-volume customers integrating the API directly. They are seeking data their agents are still unable to see.

Other customers include AI research platforms and data resellers. Exa, an API search provider for agents, is among Radar's partners.

The commercial signal is clear. Demand comes from players who treat audio as a source of information alpha, and API availability shifts the build-versus-buy calculation for anyone building intelligence pipelines. A hedge fund that previously ignored spoken content can now integrate it with a single API call. The barrier to entry is no longer transcription, but the discipline with which the data is consumed.

The Risk: A Transcript Is Untrusted Input

A transcript retrieved from Radar enters a RAG pipeline like any other document. It carries the same credentials as the user querying it.

A transcribed audio artifact is sufficient to deliver a hostile instruction to a downstream agent. Prompt injection remains the most underestimated attack vector among enterprise AI teams. Most production RAG architectures treat documents as trusted input, and that approach is wrong.

A speaker can utter a sequence designed to manipulate the model processing the transcript. The attack surface is the spoken content itself. Unlike a document downloaded from a known source, a podcast episode is public content that anyone can produce. The vector requires no access to internal systems: publishing an episode is enough.

The retrieval system inherits the permissions of whoever is querying. Hostile content executed with those credentials can access resources that an external attacker would struggle to reach directly.

The Cascade: Multi-Agent Pipelines Without Circuit Breakers

Imagine a chain: Radar transcribes, one agent summarizes, a second agent decides. The output of each becomes the input of the next.

Production multi-agent systems without explicit circuit breakers will fail in cascade. This is mathematics, not prediction. The hallucination cascade documented in research applies to any pipeline where independent validation is absent.

An incorrect transcription of an entity, a name, a figure, a ticker, propagates along the entire chain. Each agent inherits the error from the previous one and amplifies it. Without an independent validation checkpoint, no agent in the chain has any way to distinguish correct data from data corrupted upstream.

Available and Production-Ready Are Different Things

Radar is available today as a commercial product with paying customers. That status warrants a precise distinction.

An available service solves a data access problem. A production-ready service for a regulated context requires audit trails, accuracy guarantees, and treatment of input as an attack surface.

Transcription quality determines the quality of every downstream decision. A hedge fund acting on a mistranscribed quote takes on direct financial risk. Pipeline fault tolerance therefore becomes a governance issue, not merely an engineering one.

Build vs. Buy: The Communication Layer Is the Moat

Protocol standardization will be the real competitive moat in agentic AI, more so than model performance. Whoever controls the communication layer between agents controls the architecture.

Radar is available as an API. This makes it a component to integrate rather than a capability to rebuild internally. Transcribing 130,000 podcasts and maintaining the index requires expensive infrastructure. Replicating that volume internally means sustaining compute and maintenance costs that few teams can justify.

The risk to evaluate is architectural lock-in. A proprietary API for audio intelligence creates a dependency. Procurement must ask what data portability guarantees and exit terms accompany the contract.

Three Questions for Enterprise AI Teams

Before integrating an audio intelligence layer, three questions define the operational scope.

1. Does your RAG system treat transcripts as untrusted input with explicit execution boundaries?
2. Does your multi-agent pipeline include circuit breakers that halt the propagation of an error downstream?
3. Does the contract with the audio vendor define data portability and clear exit conditions?

The answers determine the nature of the integration: competitive advantage or architectural trap. A team that answers these questions vaguely is accumulating technical debt.

Decisions for the Next Planning Cycle

For the CTO and Chief Digital Officer, the arrival of audio intelligence as an API shifts the stack evaluation. Audio becomes a first-class data source and must be incorporated into the existing threat model.

For the Head of Engineering, the priority is RAG pipeline hardening. Treating every retrieved artifact as potentially hostile is the production-grade baseline.

For the CFO, investment in an external API reduces initial infrastructure costs while introducing a recurring dependency risk. The Technology Procurement Committee should renegotiate portability terms before signing.

Radar is available and commercially active. The specification to adopt is security discipline around audio, more than any single vendor product.