Connect a detector

Point Sift at a real detection API

Sift does not pretend to detect AI on its own. To get real, validated (still probabilistic) classification, you can connect your own detector API. This page shows whether one is configured and documents exactly how to set it up without ever leaking a key.

Current status

Checking…

How secrets are handled

The API key is stored in a server-side environment variable and read only by the server when it calls your provider. It is never sent to the browser and never committed to the repository. The .env file is git-ignored; only the .env.example template (with no real values) is safe to commit.

When you press "Run connected detector", the text you pasted is submitted to the configured provider. That hand-off is disclosed in the interface first, and Sift never stores or logs the text.

Environment variables

DETECTOR_API_URLrequired to enable

Full URL of a detection endpoint. The proxy POSTs JSON { text, model? } here with an Authorization: Bearer header.

DETECTOR_API_KEYoptional

Optional bearer token for that endpoint. Read on the server only.

DETECTOR_MODELoptional

Optional model or reference label forwarded in the request.

DETECTOR_TIMEOUT_MSoptional

Request timeout in milliseconds (default 30000).

Setup in one paragraph

Run the app locally, copy .env.example to .env, set DETECTOR_API_URL to your endpoint (and a key if it needs one), rebuild, and restart. The check page will then read it as connected and let you run real classification with full provenance. The full flow is on the local version page.