Skip to main content
A flow is the central session object in Atlas. Every time you verify documents for a loan application, you create a flow that groups all related uploads under a single flow_uuid. That identifier follows the request from initial setup through OCR extraction, cross-checks, and final result delivery — whether via callback or polling.

​
The flow lifecycle

Atlas processes documents asynchronously. After you upload files, Atlas queues them for OCR extraction, runs pairwise cross-checks across documents, and then delivers the completed results to your callback_url or makes them available via the /verification/extracts endpoint.
1

Initialize the flow

Call POST /verification/flow/init with your application_id, product_type, auth_url, and callback_url. Atlas returns a flow_uuid that you use for all subsequent calls in this session.
2

Upload documents

Call POST /verification/multiupload with the flow_uuid and an array of file_urls. Each entry pairs a pre-signed S3 URL with a document_id you supply for tracking. Atlas returns a file_uuid and batch_id for each uploaded file.
3

Atlas processes documents

Atlas downloads each file, classifies the document type, runs OCR extraction to produce structured field output, and then executes cross-checks — pairwise comparisons of matching fields across documents (for example, dealer_address on DELIVERY_ORDER vs. INVOICE).
You do not need to specify document types at upload time. Atlas auto-classifies each file. If a document cannot be classified, the OCR entry returns an error_code of UNRECOGNISED.
4

Receive results

Atlas posts the completed CallbackData payload to your callback_url. If you prefer polling, call GET /v3/verification/status to check progress, then retrieve the full payload from GET /v3/verification/extracts once master_status is SUCCESS.

​
Flow status fields

The status endpoint returns four independent status fields, each tracking a different stage of the pipeline.
Poll master_status to determine when results are ready. When it reaches SUCCESS, the full payload is available via /verification/extracts.

​
The flow_uuid

The flow_uuid is the primary key for the entire session. You must include it in every subsequent request:
  • POST /verification/multiupload — flow_uuid in the request body
  • GET /verification/status — flow_uuid as a query parameter
  • GET /verification/extracts — flow_uuid as a query parameter
Store the flow_uuid immediately after calling /flow/init. There is no way to retrieve it later if it is lost.

​
Async processing and callbacks

Atlas delivers results asynchronously. When processing completes, Atlas sends a POST request to your callback_url containing the full CallbackData object. Your endpoint must return a 2xx response to acknowledge receipt. If you cannot use callbacks (for example, in a server-side polling workflow), use the status endpoint to poll until master_status is SUCCESS, then fetch results from /verification/extracts.
Do not poll /verification/status more frequently than once every five seconds. For most flows, processing completes within 30–60 seconds depending on the number of documents.

​
Cross-checks

After OCR extraction completes, Atlas automatically runs cross-checks — pairwise field comparisons across documents in the same flow. Each cross-check compares a field from a source document against the corresponding field on a target document and returns a similarity_score from 0 to 100. Cross-checks are returned in both the callback payload and the /extracts response under the cross_checks array.
The name field is a stable identifier for the check (for example, DEALER_ADDRESS, CUSTOMER_NAME) that you can use to consistently locate a specific comparison in your processing logic regardless of which documents are present. See Confidence Scores and Extraction Quality for guidance on interpreting similarity_score values.