Documentation
Run a complete invoice example, inspect the missing-PO finding and adapt the working request to your records.
Start the API quickstart for copy-and-run cURL, Python and Node.js commands, or create an account and try the console. The example deliberately returns a false verdict with HTTP 200: the request succeeded and found a control exception.
Start with a practical result
| Your goal | Start here | What you will have |
|---|---|---|
| Try a review without writing code | Open the console | A sample request, its returned findings and downloadable results using your signed-in account. |
| Make your first API call | Follow the quickstart | A downloadable request and runnable cURL, Python and Node.js examples. |
| Use an API client | Download the Postman collection | The same invoice request with an empty key variable and the API base URL already set. |
| Choose the right check set | Explore use cases | The supported records, required context, worked examples and limits for your workflow. |
| Resolve a specific problem | Visit Help | Practical guidance for accounts, the console and interpreting an unexpected result. |
Connect the finding to the audit objective
Choose the assertion you need to test, the control or evidence that supports it, and how a reviewer will resolve exceptions. An invoice that reconciles arithmetically can still lack a required purchase-order reference. Preserve that distinction when planning substantive testing and documenting the audit trail.
Audit concepts explains assertions, supporting evidence, control exceptions and materiality. Materiality remains the reviewer's planning and evaluation judgment; the API does not use it to suppress findings or decide which exceptions to waive.
Build the integration
The API reference is organized into separate guides for each stage of an integration. Request and response bodies use JSON. The versioned endpoint base is https://openauditor.ai/api/v1.
| Guide | What it covers |
|---|---|
| Authentication | Create and rotate a key, authenticate requests, distinguish console sessions from integrations and inspect the running configuration. |
| Verify an item | Construct a request, supply related records and reference sets, and choose question and scope options. |
| Verification results | Read verdicts and assertion outcomes, resolve evidence pointers, and distinguish certainty, probability and risk. |
| Batch audits | Queue a population with shared context, follow job status, filter and page results, then retain or delete the audit. |
| Review marks | Record whether a human reviewer judged a completed audit's verdict correct, without changing the original finding. |
| Schemas and discovery | Discover registered item types and inspect the generated request, response and context contracts. |
Understand the decision before automating it
A result applies to the checked scope and supplied evidence. A missing vendor record, an omitted comparison population or an unqualified model question can leave an otherwise ordinary item unresolved. Keep that unresolved path visible instead of treating every successful HTTP call as approval.
The requested certainty_target sets the decision gate's error budget. Returned certainty is separate evaluation evidence for a verdict group when available; it can be unmeasured or below the requested target. Neither the target nor a displayed probability establishes accuracy on your records. Inspect the individual assertions even when the final verdict is decided.
- How it works explains direct checks, qualified questions and the rules for reaching a verdict.
- Evaluation results shows published coverage, unresolved workload and error bounds with their synthetic-data limits.
- The results guide explains what to keep with a finding so a reviewer can trace it back to the request.
Operate and maintain the workflow
Read the operational guides before moving from a trial to a larger population. Match your retry behavior to the returned error, size batches with their required context, and retain only the evidence your review process needs.
- Limits and quotas covers request sizes, item counts, rate limits and account usage.
- Errors and recovery distinguishes malformed requests, authentication failures and temporary service problems.
- Security explains credential handling and request protections.
- Privacy and retention explains stored data and audit deletion.
If you are starting now, test the supplier-invoice example and compare its returned checks with the recorded walkthrough before expanding the integration.
Further reading
The OpenAuditor blog includes practical guides and investigations with recorded results. Build a batch audit review workflow covers population preparation and complete result retrieval. 99.5% of which decisions? examines the difference between a gate target and error measured among the decisions it accepts.