Troubleshooting
Agent is not answering
What to check when an agent or chatbot stays silent, shows an error message or gives a generic reply.
An agent can fail to answer in the builder’s test chat, in a chatbot on your site, or through the API. The cause is almost always one of the checks below. Work down the list, because the first ones are the most common and the quickest to rule out.
Symptoms#
- You send a message and nothing comes back.
- The chat shows an error instead of an answer.
- A chatbot visitor sees “Oops! Something went wrong. Please contact website owner!”
Likely causes, in order#
1. The workspace is out of credits#
Agents cannot run without credits. Look for the Read-only mode notice and check the credits figure at the bottom of the sidebar. API callers get “Insufficient credits to perform this action. Please recharge your credits.” Chatbot visitors see “Sorry, you’ve reached your payment limit.” See out of credits.
2. The subscription ended or the workspace is read-only#
If the plan was cancelled, the workspace is inactive and runs are refused with “This workspace is inactive because its subscription ended.” Only the owner can choose a plan again.
3. An app the agent needs is not connected#
If a tool needs an account you have not connected, you see a warning such as “Slack integration is required, but was not found” or “Component is not integrated”, and a run that reaches that tool fails with “is not configured. Please set up the integration first.” Open Integrations and connect it. See integration disconnected.
4. Memory is on with no category#
In the agent builder, Memory needs a category. The builder warns “Pick a category, or turn Memory off. The agent refuses to run while Memory is on with none picked.” Pick a category under Advanced, or switch memory off. See memory and chat history.
5. Your changes are saved but not published#
Edits save as a draft. People using the live agent, including a chatbot on your site, see the last published version until you click Publish changes. A trigger shows “Publish to activate” until you do. See share and publish.
6. The chatbot is switched off or rate limited#
On Chatbots, open the chatbot and check that Chatbot enabled is on. Visitors who send too many messages see “Too many requests. Please try again later or sign in to continue.” The limits are under Rate Limiting in the chatbot settings.
7. A step or tool failed#
A tool or step in the agent can fail even when the model itself works. Open History, find the session, switch on the Error messages filter and read the failing event. See a run failed.
Fix it#
- Send one short test message in the builder’s chat and note the exact text of any error.
- Match that text to the list above and apply the fix.
- Publish if you changed anything, then test again from the place that failed, such as your site.
Test where it failed
Reproduce the problem in the same place your users hit it, whether that is the test chat, a chatbot on your site or an API call. Each has its own settings, and a chatbot also has its own rate limits.