When something goes wrong.

Most of the time, Kanonik works quietly in the background and you barely notice the machinery. When something does look off, the cause is usually small and the fix is simple. The most important thing to know up front is this: nothing changes your record until you approve it, and your record is always the truth. An earlier chat message is not. So when you are unsure, the answer is never to guess from something your AI said a while ago. It is to check your audit log, or ask your AI to read the current state for you, because it reads the live record.

Here are the situations you are most likely to run into, what is happening in each, and what to do.

An approval link says it is expired

Approval links are short-lived on purpose. Each one works once and only for a short window, so that a stale link can never be used to change your record later. If you open one and it says expired, nothing was lost. Nothing is written until you click, so an expired link simply means that link did its job or its time ran out, not that any change slipped through or went wrong.

What to do: ask your AI to send a fresh approval link, then approve it promptly while it is still live.

A large batch is taking a moment to seal

When you approve a big set of changes at once, Kanonik records your approval right away and then seals the items into your record in the background. It works this way so the page does not sit and time out while a large batch is written. Your approval is already captured. The items land shortly after.

What to do: give it a moment, then refresh your audit log. You will see the items appear there as they finish sealing.

The page shows a spinner or a timeout after you approve

Sometimes the page keeps spinning, or shows a timeout, even though your approval went through. In most cases the work completed on the server and only the page failed to catch up. Approving again is usually unnecessary and can be confusing, so it is worth checking first.

What to do: open your audit log to see whether the change is recorded, or ask your AI to confirm the current state. Once you know where things stand, you can decide whether anything still needs doing.

You are not sure whether something landed

It is easy to scroll up, read an earlier message, and assume that is how things stand now. Do not rely on that. A message from earlier in the conversation describes a moment that has passed, not the current state of your record.

What to do: ask your AI to check the current state. It reads the live record and will tell you what is actually there. You can also look in your audit log yourself. Either way, you are reading the truth rather than a memory of it.

Your AI seems to be redoing things you already approved

If your AI starts proposing changes you have already approved and recorded, it is usually working from the earlier part of the conversation rather than from your current record.

What to do: ask it to re-read the current state first, then act only on what is still outstanding. Once it is looking at the live record, it will skip what is already done and pick up only the work that remains.

A first change will not go through on a brand-new workspace

A new workspace needs an approval channel set up once before it can record any changes. This is a normal one-time setup step. Until it is in place, a first attempt to approve and record something has nowhere to land, so it will not go through.

What to do: open the dashboard, set up the approval channel in your settings, and then retry. After that one-time step, changes record normally.

When in doubt

The pattern under all of these is the same. Your record is the source of truth, and the two reliable ways to read it are your audit log and a fresh check by your AI. An earlier chat message is never the source of truth. When something looks wrong, start there, and the picture usually clears up on its own.

More help

Browse every article in the Help center, where you can also ask the Kanonik assistant directly. For anything else, email support@kanonik.ai and a person who works on the product answers.