Who said what?When?Answered.

Most AI search tools rely on second-hand text: news articles, social posts, or one-off transcripts. Regest indexes primary spoken audio from multiple public sources: congressional and regulatory hearings, keynotes and product launches, executive interviews, and long-form remarks that move markets and headlines. Podcast and video appearances sit in the same index, so context spans formats in one place.

New content is ingested and speaker-labelled daily. What sets the dataset apart is the combination of freshness (updated continuously, not a static archive), breadth (many source types in one searchable corpus), and retrieval depth (sentence-level chunks with speaker, programme, date, and timestamp so you find the statement itself, not a wall of text). Adverts and bumpers are stripped at ingestion; every utterance is attributed to a named speaker where the pipeline can verify it.

Add Regest as a custom MCP connector in Claude Desktop or OpenAI Codex (desktop agent app). Copy one URL from the Setup page, sign in once, and your agent can call the index from chat. You can also search directly at regest.com/search: same corpus, no connector required.

Our retrieval layer is built for agentic reasoning: low-latency natural-language input that matches search intent and surfaces the relevant statement, even when you do not know the exact wording. Agents receive speaker, timestamp, programme context, and the exact passage on every hit, so they can cite on record instead of paraphrasing from memory.

Fact-check a claim: give an agent a statement from social media or a draft article. It searches for source-attributed quotes that confirm or contradict it, with speaker and timestamp on every hit.

Track how a position evolved: ask how a public figure's stance on a topic changed over the last year. Regest surfaces statements across multiple appearances, not just the latest headline.

Extract unreported quotes: after a long interview or hearing, find substantive lines that have not been picked up elsewhere. Useful for newsrooms and analysts working from the primary record.

Ground AI workflows: connect once via MCP; Claude, Codex, or your own stack pulls verified spoken context on demand.

Journalists, researchers, and AI teams use regest when the answer has to be who said what, when, on record.

Generic transcription gives you a wall of text. Web search gives you articles about what someone said. Regest indexes what was actually said, at sentence level, with who said it and where to find it in the original recording.

Regest is a data product built for retrieval: specialised conversational diarization labels who is speaking, adverts and bumpers are stripped at ingestion, and every utterance is indexed with rich metadata. We maintain a broad, continuously updated index of public spoken conversations so search stays precise and agents can cite with confidence.

For the public corpus, regest runs specialised conversational diarization tuned for interviews and dialogue: voice matching, context clues, and verification against known speaker profiles. Wrong attribution is worse than missing content. If the pipeline cannot verify a speaker, the label stays unknown rather than guessed.

Every searchable quote carries a speaker label, so agents and humans always know who said what, on record.

Public index: browsing and searching the public corpus does not require uploading anything.

Your queries: never used to train models, never sold, never shared with other customers. MCP access is read-only retrieval over the indexed corpus, not a pipeline that exfiltrates your source material.

Finding the statement you need, when you need it, is hard when it only exists in hours of audio. Regest prepares and indexes public spoken audio at sentence level, so search can be precise and effective at surfacing hidden statements.

Our system transcribes accurately, removes ads where present, and intelligently labels who is speaking. Vector representations are stored so finding a specific sentence stays fast, even across large collections.