Tool reference

Use XData to explore investment signals across podcasts, video interviews, and newsletters. Search an investment theme, inspect the matching passages, then retrieve the full source to develop and test your idea.

`search`

Semantic and keyword search over transcripts, articles and X posts. Results come best first, each with up to two verbatim passages (speaker and time offset for recordings) and the url to cite.

Param	Use
`query`	A question, topic or "exact phrase". `""` browses the newest items.
`source`	Show, newsletter or X account (see `list_sources`).
`speaker`	Only passages this person spoke, in transcripts with verified speaker labels.
`types`	`podcast` · `video` · `article` · `x_post`.
`sector`	Sector code or name, e.g. `TEC_SEM` or `Semiconductors and chipmaking`.
window	`published_after / published_before` (ISO dates).
`sort`	`relevance` (default) or `recent`: the relevant documents, newest first.

Continue with next_cursor. A note appears when nothing matches closely.

`fetch`

The complete text of a transcript or article by id, with its title, citable url, source, date and speakers. Transcripts read as "Speaker: words" paragraphs with advertising removed. Documents over 100,000 characters come in parts (fetch(id, part=2)). Articles list their figures with an image_id.

`list_sources`

Shows, newsletters and X accounts with item counts and latest coverage. Monitored feeds are checked every 15 minutes.

`get_image`

A figure from an article, by the image_id that fetch lists.

`request_ingestion`

Queue a public episode, YouTube, RSS, or newsletter URL. Returns an existing content_id or new job_id. Text takes minutes; audio is transcribed on-device overnight, searchable the next morning. Content joins the shared corpus.

`get_ingestion_status`

Poll a job: queued → working → completed for text; needs_local → local_fetching → prefetched → completed for overnight audio.

Recipes

Earnings-season sweep

search("NVIDIA earnings guidance", published_after=…) → compare the passages → fetch the strongest sources.

Track a person

search("AI spending", speaker="Brad Gerstner") → his own words, with time offsets, across shows and dates.

Latest on a theme

search("AI capex payback", sort="recent") → the relevant discussions, newest first.

Ingest and wait

request_ingestion(url) → for queued audio, advise waiting until morning. For text, poll get_ingestion_status every ~20s for up to 5 minutes.

Machine-readable: server card · llms.txt · setup guide