Playground · Vindex docs

Try prompts against your policy before anything is enforced. There are two playgrounds: this one, for a single prompt, and the dashboard's, for a whole file.

Here

Paste a test key and a prompt. You get the decision and its explanation back without leaving the page.

Try a prompt

No key: sample response

YES · allowSample · "a lighthouse at dusk, oil painting"

Allowed: no check reached its review line.

Minors, sexual0.01nothing to do
Adult content0.01nothing to do
Real people0.01nothing to do
Copyrighted IP0.02nothing to do

Response JSON

{
  "id": "dec_8c1187e9ea1a4b04a5adb40f19868d19",
  "outcome": "allow",
  "tags": [],
  "checks": {
    "minors_sexual": {
      "score": 0.0085,
      "review_score": 0.01,
      "fired": []
    },
    "adult": {
      "score": 0.01,
      "review_score": 0.01,
      "fired": []
    },
    "real_person": {
      "score": 0.01,
      "review_score": 0.01,
      "fired": []
    },
    "copyrighted_ip": {
      "score": 0.02,
      "review_score": 0.02,
      "fired": []
    }
  },
  "policy": {
    "id": "pol_cb805d8999bd453f8f2b0ffd04f05b59",
    "version": 1
  },
  "engine_version": "2026.09",
  "latency_ms": 383,
  "dry_run": false,
  "mode": "test",
  "shadow_mode": false
}

It takes test keys only. The key stays in this tab (sessionStorage) and is only ever sent to api.getvindex.com.
Every run is a dry run: logged as a test decision, with no review item and no webhooks.
Without a key it shows a real response, captured from the API, marked as a sample.

In the dashboard

Playground runs many prompts at once against your own policy:

Paste prompts, one per line, or upload a CSV with a prompt column. An optional expected column (allow, review or block) scores the policy against what you expected.
Compare two versions of a policy side by side, including one you haven't published, to see what a change would do before it's live.
Export the results. Cells that would run as spreadsheet formulas are defused, because exported prompts are often your users' text.

Every run is a dry run with a test key, paced under the rate limit at about 9 checks a second: 2,000 rows against two versions take about seven and a half minutes. Rows that hit a 429 or 503 wait for Retry-After and go again.

A prompt blocked by the minors check is never shown in the results or the export, only a placeholder.

Doing the same with the API

A playground run is a normal check with two extras: the X-Vindex-Dry-Run: true header, and policy_version to pin a version. Dry-run & policy_version has both.