Bigdata.com

What it is

The platform RavenPack launched in October 2024 and now sells alongside its licensed datasets. It is a retrieval layer, not a feed — you send a query and get back the passages that answer it, with citations, rather than a stream to index yourself.

Four services sit behind one API. Search is retrieval only, returning ranked chunks with no synthesis. The Research Agent is a conversational RAG endpoint that reasons over the same index. The Workflows API runs parameterised templates for reproducible, batched research. A Monitors API that turns a saved search into a recurring event feed is documented and marked coming soon, which is worth reading as "not yet" rather than as a capability.

Retrieval is chunk-level. Each returned chunk carries a relevance score, a sentiment score from -1.00 to 1.00 and entity detections with character offsets, and documents are pre-chunked so a story mentioning four companies can be sliced per company. Entities are knowledge-graph objects with their own identifiers — companies, ETFs, organisations, people, places, products, topics and sources — which is how you pin a query to a company rather than to a name.

Pricing

Per token, with two shapes and no seats. Individuals self-sign up, get free credits (the amount is not stated publicly), and top up a wallet afterwards. Enterprises commit an annual spend and draw it down at a discount.

The per-token rate is the number that is not published. The docs point at the Usage page inside an account for content-tier and model prices, so there is no way to price a workload before you have a key. What the pricing page does publish is an estimator with worked runs — seven example agent workflows between $0.57 and $1.82 each — and one of them broken down by content type, from premium news at $0.46 down to regulatory filings at $0.01. A content set you lean on heavily can be switched to a flat subscription that drops its per-token content charge; that price is not published either.

Data & coverage

The store lists nine sets and dates most of them: public web news at over 10 million new documents a month on a rolling five years of history; premium news from 200-plus licensed sources including Benzinga, the Financial Times, MT Newswires and Risk.net, with content since 2000; filings across 90,000 companies in more than 50 countries since 2010; and fundamentals with 30-plus years. Press releases from 25,000 companies, analyst research from 100-plus providers, 200-plus expert interviews a year, 4,400-plus podcast shows and ten-or-more alternative datasets covering jobs, ESG and supply chain fill out the rest.

Documents are pre-processed in over 13 languages and translated into English for search. Indexing is real-time. There is no quote or bar data here at any latency.

Integrations

REST at api.bigdata.com/v1, a Python SDK (bigdata-client, published by the Bigdata Team, 2.21.0 in May 2026), and first-party MCP connectors for Claude, ChatGPT and Microsoft Copilot — which is the intended path, not an afterthought. A developer platform hosts playgrounds for the endpoints. You can also upload your own files and have them indexed alongside, priced as storage.

Limitations

The licence is the thing to read before you design anything. Section 4.2 of the terms grants a "Grounding Right" to use the content at inference time only. You may keep what your model wrote from it; you may not store, cache, retain or archive the underlying content past the moment of consumption, may not pass it to a third party, and may not use it to train, fine-tune or evaluate any model. Each retrieval is a separate one-time right with no standing licence. That rules out building your own index off the back of it, which is the usual reason to buy a news API. The same terms make you warrant that any model you connect runs under zero data retention, and bar you from configuring it to re-expose, proxy or resell the content onward — the constraint follows the content into whatever you plug in at the other end.

Public news ages out at five years, so the depth advertised in the marketing copy belongs to the premium sets, and what you can reach depends on which ones your account has.

The vendor's own pages disagree on broker research. The store lists analyst research from over 100 providers, "including tier-1 brokers"; the docs FAQ says Bigdata does not include broker research because of entitlement complexity, and that you can upload your own where you are entitled to it — which there is a connector for. The store is the page RavenPack has to keep current, but settle it before you count on it.

The sentiment model's training data stops in 2023, per the docs.

Alternatives

AskNews is the closest thing in shape — licensed news retrieval aimed at LLM pipelines — and publishes its per-credit rates. For the same company's contracted analytics feed rather than a retrieval endpoint, see RavenPack News Analytics. If what you actually want is filings and transcripts to read rather than to ground a model, AlphaSense and Quartr sell that directly.

Compare it against the rest of the market data APIs.