Data API

Authentication

Two secret tokens per site, generated separately in Setup → Data API and each shown only once. One scope model everywhere — REST and MCP:

Send either as a bearer token:

Authorization: Bearer sk_your_secret_token

Unlike the publishable widget key, both tokens are secrets — keep them server-side, never in client JS. Each is scoped to exactly one site and individually rotatable from the dashboard; the old token stops working immediately.

Endpoint

GET https://feedback.latentedge.io/api/v1/export/feedback

Query parameters

Example request

curl -s "https://feedback.latentedge.io/api/v1/export/feedback?limit=50" \
  -H "Authorization: Bearer sk_your_secret_token"

Response

Every response is wrapped in { success, data }. For this endpoint data holds the page:

{
  "success": true,
  "data": {
    "feedback": [
      {
        "id": "0f2c…",
        "siteId": "c7b3…",
        "type": "bug",
        "message": "The export button 404s",
        "rating": null,
        "pageUrl": "https://yoursite.com/reports",
        "status": "new",
        "createdAt": "2026-06-27 19:04:11"
      }
    ],
    "nextCursor": "MjAyNi0wNi0yNyAxOTowNDoxMXww0f2c",
    "hasMore": false
  }
}

• nextCursor — pass it back as ?since= on your next call. It is always returned so you can keep polling for rows created after the latest one.
• hasMore — true when more rows are immediately available (keep paging right away); false when you've caught up.
• Personal fields (IP address, user agent) are not included.

Polling pattern

Store nextCursor between runs and pass it as since. You'll only ever receive rows you haven't seen, in stable order. Poll at whatever interval suits you — empty polls are cheap, so there's no need to hammer it. A minute is plenty for most workflows; if you need lower latency, ask us about webhooks.

// Poll for new feedback on an interval and persist the cursor to resume later.
const ENDPOINT = 'https://feedback.latentedge.io/api/v1/export/feedback';
const TOKEN = process.env.FEEDBACK_READ_TOKEN; // sk_...
let cursor = loadCursor(); // your own storage; null on first run

async function poll() {
  const url = new URL(ENDPOINT);
  if (cursor) url.searchParams.set('since', cursor);
  url.searchParams.set('limit', '100');

  const res = await fetch(url, { headers: { Authorization: 'Bearer ' + TOKEN } });
  if (!res.ok) throw new Error('export failed: ' + res.status);

  const { data } = await res.json();
  for (const item of data.feedback) {
    handleFeedback(item); // do your thing
  }
  if (data.nextCursor) {
    cursor = data.nextCursor;
    saveCursor(cursor); // resume from here next run
  }
}

setInterval(poll, 60_000); // once a minute is plenty

Updating feedback status

Triage from your own tooling with the write token: mark an item triaged once you've reviewed it or resolved once it's addressed.

curl -s -X PUT "https://feedback.latentedge.io/api/v1/export/feedback/FEEDBACK_ID" \
  -H "Authorization: Bearer wk_your_write_token" \
  -H "Content-Type: application/json" \
  -d '{"status": "triaged"}'

A read token gets a 401 here; an id belonging to another site is a plain 404.

Question batches

The outbound channel: publish a set of questions, send the share link to a client or stakeholder, and they answer in the browser with no account. These endpoints let your code (or your agent) send batches and read the answers back.

curl -s "https://feedback.latentedge.io/api/v1/export/questions" \
  -H "Authorization: Bearer sk_your_secret_token"

Rate limits

Each site's token is limited to 60 requests per minute. Over that you get a 429 response — back off and retry on your next interval. Polling once a minute uses one request, so you have plenty of headroom.