# Integration docs tested against live APIs every day

> A test suite per service checks what our integration docs claim against the real API daily, so agents hit fewer failed calls. Plus August's new apps.

Aug 31, 2026. Topics: Integrations, Platform. https://www.generalinput.com/changelog/2026-08-31-integration-docs-verified-live

Agents call your integrations from our documentation of each API, so a wrong field name or a retired endpoint becomes a failed run. We now check that documentation against the live APIs. Each service gets a test suite whose probes encode what our docs claim, and a daily run replays them. The finding we care about is a faithful probe failing because the API disagrees with the docs.

### How it works

- **Two modes.** Verify replays the committed suite deterministically, with no model involved, so it's cheap enough to run daily. Author runs a maintainer agent to write or repair a suite, but never in response to a failure, which would hide real drift.
- **Evidence, not status codes.** A 200 can carry an error body and a 403 can mean a higher plan tier, so each result is classified as passed, failed, needs a human, or requires a paid plan.
- **Safe test data.** Probes may create the records they need, but never spend money, message a person, or touch data they didn't create.
- **Scope today.** The catalog documents 34,054 operations across 557 services, and 201 services with test accounts put 11,169 operations in reach.

### Also shipped

- New integrations in August include 1Password, Bitwarden, SignWell, DocuSeal, Hootsuite, Sprout Social, GoHighLevel, Paylocity, Snowplow, and Circleback, 16 with polling triggers. See the [catalog](/apps).