Sonar

Sonar is an audio search API for AI agents. It enables natural-language queries against public audio content from news broadcasts, podcasts, social media, radio, earnings calls, and archives, returning ranked clips accompanied by accurate transcripts, timestamps, speaker attribution, and source details.

The service addresses spoken content that has remained unindexed on the web. Audio from discussions, hearings, and broadcasts is typically inaccessible to automated systems. Sonar applies embedding, approximate nearest neighbor search, and reranking to surface relevant results. Agents submit a query through a single endpoint and receive structured outputs that include transcript segments with timing and speaker information, allowing direct quotation or summarization.

Three steps structure its operation. An agent issues a tool call with a question on a topic, event, or speaker. Sonar performs the search across the indexed audio sources and returns ranked clips. Each result supplies the source, speaker, timestamp, and transcript for further agent processing. Example queries in its playground demonstrate retrieval on subjects such as researcher views on large language models, corporate events, legal proceedings, technology developments, and policy matters.

It is delivered as a web API in public beta version 0.4.1. Access requires a manual review process that grants instant API keys upon approval, typically within 24 hours. The service targets developers and teams building AI agents, LLM pipelines, news monitoring tools, podcast intelligence platforms, research or academic projects, and media monitoring solutions. Early users receive a free tier of 500 queries per month without a credit card and entry to a private Slack channel with the founders for direct feedback and roadmap input. The company is based in San Francisco.