API reference
Extract
Pull named fields out of a document: each value as written, with its position and a confidence.
POST/v1/extract
Base URL https://api.wity.alphanimble.com. Same authentication as decisions: Authorization: Bearer wity_…. For how to write fields, see Extract.
Request body#
text
stringrequired
The document, 1 to 60,000 characters of plain text.
fields
objectrequired
1 to 16 fields: name (up to 64 characters, no double quotes) → what the field asks for, in plain words (up to 300 characters).
types
object
Optional. Field name →
date (YYYY-MM-DD), duration or number: those fields come back normalised.Note
Unknown top-level fields are rejected with 400.
Response#
{"fields": { "name": ["PRIYA NAIR"], "college": ["National Institute of Technology, Calicut"], "skill": ["Kafka", "Python", "SQL", …] },"entities": [{ "field": "name", "text": "PRIYA NAIR", "start": 0, "end": 10, "confidence": 0.9924 },{ "field": "college", "text": "National Institute of Technology, Calicut", "start": 571, "end": 612, "confidence": 0.8933 },{ "field": "skill", "text": "Kafka", "start": 265, "end": 270, "confidence": 0.9604 },…],"timing_ms": 503,"usage": { "input_tokens": 301, "output_tokens": 0 }}
fields
object
Every field you asked for, in your order, with its values (
[] when the document has none).entities
array
The same values with
field, text, character offsets start / end (end exclusive) and confidence (0 to 1), by field and then by position. For a typed field, text is the normalised value and the offsets mark where it was read.timing_ms
number
Time spent on the server.
usage
object
input_tokens: the document and the field list, counted once, billed at $0.042 per million. output_tokens is always 0.Errors#
400
invalid
Empty text or over 60,000 characters, no fields or more than 16, a name or description out of bounds, or a type other than date, duration or number. The body says which.
401
auth
Missing or unknown key.
402
credit
Your balance is zero. See Errors, limits & billing.
429
busy
Your key's rate limit, or too many extractions running on the server. Retry after
Retry-After seconds.Note
Long documents take longer: use a client timeout of at least two minutes.
