News API for RAG & Vector Pipelines

APITube is a news API built for retrieval-augmented generation. It returns full article bodies from 300,000+ sources across 59 languages, each enriched with entities, per-entity sentiment, IPTC categories, source authority, and a story endpoint that groups related coverage — exportable as JSONL or Parquet for embedding pipelines and kept fresh over SSE, WebSocket, or webhooks.

An LLM stops knowing the world at its training cutoff. A RAG pipeline closes that gap by retrieving current documents at question time — and news is the fastest-moving corpus there is. A news API gives the pipeline clean, structured, sourced articles to embed and cite, instead of scraped HTML with boilerplate and duplicates.

Yes. Every article from /v1/news/everything carries the full body along with the title and description, so you can chunk and embed real text rather than a headline. The same enriched schema is returned on every plan, including the free tier.

Three ways. Hold a stream open — Server-Sent Events or WebSocket — and embed articles as they arrive; register a webhook and let APITube POST new matches to your ingestion endpoint; or poll the REST API with date filters on a schedule. All channels accept the same filters, so the index only receives what your pipeline actually retrieves against.

Every APITube article carries a story ID that groups related coverage of one event into a single cluster. Ingest one representative article per story instead of 40 syndicated copies, and your retriever stops returning near-identical chunks for the same query.

Each APITube article carries entities across 9 types linked to Wikipedia and Wikidata, sentiment at the overall, title, body, and per-entity level from −1 to +1, IPTC Media Topics categories with relevance scores, the source with a political bias label and an Open PageRank authority score from 0 to 10, language, publish date, and a story ID. Store them as vector-database metadata and you can pre-filter retrieval by company, topic, sentiment, authority, or recency before similarity search runs.

The REST API exports JSON, JSONL, Parquet, CSV, TSV, XML, RSS, and XLSX. JSONL streams line-by-line into batch embedding jobs, and Parquet loads directly into the data warehouses and dataframe tools most ingestion pipelines already use.

Yes. APITube keeps a historical archive you can query by date range using published-at filters, so you can walk backwards through time windows and batch-ingest the depth your corpus needs before switching to real-time updates.

Not always. APITube runs a hosted MCP server at https://mcp.apitube.io/ that exposes news search as tools an agent can call directly — no vector database required. And the /v1/news/fact-check endpoint runs retrieval-augmented verification server-side: it checks a claim against the live index and returns a verdict with the coverage behind it.

Every article carries its canonical URL, source domain, source authority score, and publish date. Keep them in your chunk metadata and the model can attribute each statement to a linkable, ranked source — which is the difference between a grounded answer and a plausible one.