BY NUMBER7 AIInvoice extraction & document processing API
Bring document results into your application.
Submit a file or supported document URL, retain the returned job ID and retrieve structured results. Your application owns downstream review, approval and action.
Example invoice · LAB-101File → job → result
01Submit the sourcePOST /api/v1/upload
Upload the document. Processing is asynchronous; upload acceptance is the start of a job.
02Keep its identifierdata.jobId
Retain the returned job ID. Use it to retrieve the result, or associate a verified completion callback with the same job.
03Read the job resultdata.fileStatus
Check the terminal status. A completed document result is ready for your application’s review and business controls.
Follow the documented job flow with LAB-101. Test the API with your own credentials to inspect the returned result.
Published v1 contract
Make one request. Keep one job identifier.
curl --fail-with-body --request POST \
'https://api.number7ai.com/api/v1/upload?document.type=invoice&document.version=v1' \
--header "Authorization: Bearer $N7_API_KEY" \
--form 'file=@invoice.pdf' \
--form 'fileName=invoice.pdf' \
--form 'fileType=application/pdf'Read and persist data.jobId from the response. Upload acceptance means a job exists; it does not mean processing is complete.
curl --fail-with-body \
'https://api.number7ai.com/api/v1/results/YOUR_JOB_ID' \
--header "Authorization: Bearer $N7_API_KEY"Check data.fileStatus. done means a result is available; failed needs error handling.
The examples use the published request and status names. The completed business-data schema must come from the verified contract for your integration; this site does not invent an invoice response.
A clear ownership boundary
Document completion starts your business workflow.
Document processing
Returns the documented structured document result and processing status.
Business controls and action
Checks required accounting fields, routes uncertain cases and owns downstream review, approval and action unless separately implemented.
Extraction completion alone does not establish a valid vendor, approved bill or successful accounting posting. Keep source evidence and the returned job identifier attached to the work.
Asynchronous by design
Follow the job through to a terminal result.
POST /uploaddata.jobIddata.fileStatus| Status | What your application does |
|---|---|
uploading · processing · scanning · ai_processing | Continue bounded polling or wait for the configured completion webhook. |
done | Retrieve and validate the structured result against the requirements of your workflow. |
failed | Handle the documented job failure and surface the next action. |
Implementation checkpoints
Plan for the cases around the happy path.
Keep credentials server-side
Use Bearer-token authentication. Scope credentials to the agreed account and environment.
Bound the wait
The guide recommends polling about every 20 seconds with a timeout. Do not poll indefinitely.
Read the polling guidanceVerify completion callbacks
Authenticate the raw payload before parsing it, then handle duplicate delivery safely.
Read webhook verificationReconcile uncertain uploads
If a timeout may follow acceptance, reconcile before resubmission. Do not assume an idempotency header exists outside your agreed contract.
Read errors and retries
Try the same source files
A document boundary can matter as much as a field.
The example pack includes a four-page PDF containing two invoices and one credit note. Pages 2–3 belong to one bill. Compare the actual output of the API you evaluate with the answer key.
Compare document boundaries and accounting amounts, then record the API’s result on the same files.
Next step
Discuss the contract your application needs.
Map result fields, operational limits, webhook handling and downstream ownership with the product team.