troubleshooting

Common issues

Work through the most common OpenbirdAI problems — blank views, empty answers, failed connections and credit messages — in the right order.

· OpenbirdAI documentation

Most OpenbirdAI problems are one of four things: the wrong workspace, a missing connection, an unclear question, or a genuinely unavailable capability. This page is the triage guide; the pages it links to are the fixes.

Start here

  1. Confirm you are signed in and the correct workspace is active.
  2. Reload the page once — a stale view is a common and harmless cause of an apparently broken screen.
  3. Test a general question with no private data. If that works, the problem is authorisation or data, not the product.
  4. Test a narrow private-data question against one connected service.
  5. Read the error text; it usually names the failing stage.

Symptoms and likely causes

  • Blank or empty main area → the view did not finish loading, or a drawer is covering it. Close drawers, scroll to top and reload.
  • Answer is empty → check workspace, connected account, and date range. See No data in an answer.
  • Answer ignores instructions → the capability may be unavailable; the product states limitations instead of guessing.
  • Connection shows connected but fails → reconnect and approve all scopes. See Troubleshoot a connection.
  • "Not enough credits" → check which limit was reached. See Credits and limit errors.
  • Layout looks clipped or unreadable → reset browser zoom to 100% and reload; zoom below 100% on a narrow window can compress the reading column.

Things that are not bugs

  • An answer that cites nothing for a general question — there was nothing to retrieve.
  • A capability that reports itself unavailable — a documented limitation, not a failure.
  • Different results for the same question later — the underlying data changed.
  • A missing button — your plan or role may not include it.

Collecting information before reporting

Note the approximate time, the workspace, the exact error text, whether a general question works, and whether a narrow private-data question works. Never include tokens, passwords or provider secrets in a report.

Common questions

How do I tell whether it is authorisation or the model?

If a general question works and a private-data question returns empty, it is authorisation or data, not the model.

Should I report a slow response as a bug?

Record it, but first rule out a very long conversation, which legitimately costs more and takes longer.

Was this page helpful?