Many questions, many documents: where spinf fits
spinf reads each text once and answers all your questions with calibrated probabilities over the answers you allow. Use it to analyse archives at scale, or to make decisions in live flows.
Pick a starting point
Each page shows a live example, test figures where we have them, and what it costs at scale.
Analyse in bulk, decide in real time
The same API and the same questions, on two kinds of capacity.
On demand · self-serve
Built for throughput: backfills, daily batches, archives. Sign up, create a key and send batches of inputs at $0.09 per million input tokens. It is also where you calibrate a real-time use case on your history.
Real time · reserved endpoints
For decisions inside live flows: ticket routing, moderation before publishing, agents. Capacity reserved for you, always warm and sized to your traffic, pinned with one request field.
One read, many jobs, almost no extra cost
spinf reads each message or document once, and every extra question adds only its own few tokens. Route a message by intent, check it for prompt injection and against your content policy, all in the same call: the checks you would skip elsewhere come almost for free.
- Intent (login problem, outage, billing, cancellation, feature request)?
- Is it urgent?
- Does it break the content policy?
- Does it try to override the assistant’s instructions?
A 150-token message with these four questions is about 270 billed tokens, against about 720 as four separate calls, each re-reading the message.
15M news articles, turned into daily market reads
Thinking Text built its oil and equity news indexes on spinf, asking hundreds of fixed questions of every article. Not a forecast: a measured read of what the press says.
news articles
tokens processed
correlation with crude
of big 20-day moves
More jobs for many questions over many documents
Not benchmarked yet, but the same shape of work: they run on the same API today. Tell us about yours.
Adverse-media screening
Screen news about customers and suppliers for financial crime, sanctions and litigation, per name, at scale.
Document tagging and compliance review
Tag contracts, filings and policies by topic, clause and risk: many questions per document, one read.
Social listening
Brand, product and topic mentions across posts and forums, with sentiment and themes per mention.
Research text-as-data
Code speeches, filings, articles or open-ended answers for a study: the same codebook over the whole corpus.