Skip to content

Troubleshooting

Common issues and their solutions.

Agent Not Responding

Issue: Agent does not generate a response or shows an error.

  • Verify at least one language model is configured under the agent's Brain module
  • Check that the model's credential is valid under Settings > Credentials
  • Ensure the Large Context Processing Algorithm is selected
  • Check Batch Monitoring (User Icon > Batch Monitoring) for any failed jobs

Documents Not Appearing in Responses

Issue: Uploaded documents are not being used by the agent.

  • Verify document status is Ready in the Knowledge Base
  • Confirm the agent has a Retrieval tool or Context tool configured under the agent's Tools & Artifacts module
  • Check that Document Locks don't restrict access for the user's role
  • Ensure the Knowledge Base is selected in the tool configuration

Voice Features Not Working

Issue: Mic button not appearing or TTS not playing.

  • Verify Speech to Text and/or Text to Speech are enabled in the agent's Voice module
  • Check that ASR/TTS models are configured under Settings > AI Models
  • Ensure the browser has microphone permission granted
  • For Realtime Voice, verify SIP/Websocket configuration and that the Realtime Voice Model is selected

Crawl Job Failing

Issue: Web crawl job shows "Failed" status.

  • Check the URL is accessible and not blocked by the target website
  • Verify the User Agent string is not being blocked
  • For Confluence/SharePoint, confirm credentials and permissions are correct
  • Try reducing Max Pages and Depth for initial testing
  • Use Review Documents mode to inspect individual pages before indexing
  • For failed uploads, use the Retry Failed action on documents in "Upload Failed" state

Widget Not Loading on Website

Issue: Embedded widget doesn't appear on the website.

  • Verify the embed script snippet is correctly placed in the page HTML
  • Check that the website domain is in the Allowed Domains list (the agent's Widget module)
  • Ensure CORS is configured to allow the domain
  • Test using the Bookmarklet first to verify agent functionality

Pre-Processor Tags Not Applied

Issue: Conditional behavior based on tags is not working.

  • Verify the pre-processor is assigned to the agent under the agent's Pre Processors module
  • Check that the pre-processor JavaScript logic is correct (test with known inputs)
  • Confirm the tags returned by the pre-processor match the tags expected in shortcuts/instructions
  • For Prompt Classifier, verify the Acceptance Threshold is not too high

DLP Policy Not Working

Issue: Sensitive data is not being anonymized.

  • Ensure the DLP policy is set as Active (only one policy can be active per workspace)
  • Use the Test feature on the DLP Policies page to verify detection
  • Check that the Acceptance Threshold is appropriate (default: 0.8)
  • Verify keywords are not in the Whitelist Keywords field

Contact Support

If the issue persists after troubleshooting, contact the support team:

  • Email: support@fifthelement.ai

Collect Useful Evidence

Before escalating, record the selected workspace, agent or Knowledge Base, time, intended role, exact action, expected result and actual result. Include a request/job identifier where available, but remove API keys, connection secrets, personal details and private document content.

Symptom Check first Next step
Knowledge query misses an expected source Document readiness, selected Knowledge Base and filters Repeat a known-source query and review member/role document keys; see Search Agents.
Background work is not completing Current status and response in the job view Distinguish waiting, required user action and failure; see Jobs API. Avoid automatic reruns.
MCP client cannot retrieve a document Knowledge Base, approval period and intended identity Review MCP Connections and client configuration before requesting more access.
A report is empty or unexpected Measure, Time Range and active filters Compare with known retained test data; see Agent Reports.
Speech input or playback fails Browser microphone/audio settings and selected ASR/TTS model Separate input transcription from output playback; see Voice workflows.
A tool reports a failure Statistics and the matching conversation/debug record Check the specific invocation and configuration rather than relying on the aggregate failure count.
Async chat polling returns 404 Singular /chat/request/{id} route, request ID, age and identity Check the async contract; the schema also uses 404 for expired or inaccessible requests.