Skip to main content

​
Quick Health Checklist

Use this checklist for rapid system verification: ✔️ Socket Connection Socket indicator shows green dot (on Elitea logo, top-left corner) The socket connection indicator shows the real-time connection status between your browser and the ELITEA server:
  • Green dot = Connected and ready Connected
  • Red dot = Disconnected - check your network connection Disconnected
✔️ Permissions Can create entities available by permission scope Verify that you can create agents, toolkits, credentials, and other entities based on your assigned permissions. If you cannot create certain items, contact your administrator to review your permission scope. ✔️ Credentials Test connection shows “successful” status After creating credentials for external services (Jira, Confluence, GitHub, etc.), use the “Test Connection” button to verify they work correctly. A successful test confirms your API tokens and authentication settings are valid. Connected Connected ✔️ MCP Integration MCP status shows connected (if using MCP features) If you’re using Model Context Protocol (MCP) features, verify the MCP status indicator shows as connected. This ensures proper communication with MCP servers for enhanced functionality. If not connected, you need to run your MCP server first. See how to create and use MCP client for setup instructions.
  • Connected and ready
Connected
  • Disconnected - MCP server not running
disconnected ✔️ Indexes Tab Availability Indexes tab is visible in Toolkit Configuration (for supported toolkits)
  • Indexes tab is automatically disabled if required prerequisites are not met. To access this tab in your toolkit configuration, ensure:
    • PgVector Configuration: Vector storage must be configured at project level (Settings → AI Configuration)
    • Embedding Model: An embedding model must be selected and configured (Settings → AI Configuration)
    • Index Data Tool: The “Index Data” tool must be enabled in your toolkit configuration
  • Delete button (in Indexes tab) is disabled if Remove Index tool is not selected in toolkit configuration
  • Run tab (in Indexes tab) is disabled if at least one search index tool (Search Index, Stepback Search Index, or Stepback Summary Index) is not selected in toolkit configuration
For detailed setup instructions, see How to create and use indexes.

​
Common Issues: Quick Fixes

​
Agent & LLM Problems

Symptoms:
  • Agent starts working but stops mid-response
  • Answer appears truncated or incomplete
  • Complete answer visible in “thinking step” but not final response Agent truncated response Agent truncated response example
Root Cause: Max completion token limit reachedSolution:
  1. Open agent configuration (⚙️ settings icon)
  2. Locate “Max completion tokens” field
  3. Increase value (e.g., from 2000 to 4000 or higher)
  4. For nested agents: Check each agent’s token limits
  5. Save and retry
Solution screenshotRelated: Troubleshooting Guide §1
Symptoms:
  • “This conversation violates our policy”
  • “Message filtered due to prohibited content”
  • “Rejected as a result of our safety system” Content policy error
Root Cause: Message triggered LLM safety mechanismsSolution:
  1. Review your prompt for sensitive topics
  2. Rephrase instructions that might look like prompt injection (e.g., “Ignore user input”)
  3. Clearly state the purpose of security-related instructions
  4. Avoid prohibited content per your provider’s policy
Related: Troubleshooting Guide §8
Symptoms:
  • “Hit token rate limit. Minute limit: X / Y tokens”
  • Error appears during heavy usage or repeated requests Rate limit error Rate limit error example
Root Cause: Too many requests to LLM or embedding modelsSolution:
  1. Wait 1-5 minutes and retry
  2. Switch to a different LLM model if available
  3. Reduce request frequency or batch operations
  4. For shared environments: Coordinate with team to spread usage
Related: Troubleshooting Guide §9
Symptoms:
  • “This model’s maximum context length is X tokens. However, your messages resulted in Y tokens”
  • Error when processing large files or long conversations Context length error
Root Cause: Input exceeds model’s context window limitCommon Limits:
  • GPT-4o: 128,000 tokens
  • Claude Sonnet: 200,000 tokens
  • GPT-5: 400,000 tokens
Solution:
  1. Reduce input size: Break large files into smaller chunks
  2. Use summarization: Summarize content before passing to LLM
  3. Use indexing: Instead of passing full documents, use toolkit indexing and semantic search
  4. Clear chat history: Start a new conversation to reduce context
Related: Troubleshooting Guide §10

​
Toolkit & Integration Problems

Symptoms:
  • “The calling user does not have permission to view the content”
  • Indexing stops or fails partway through
  • Some items indexed, others skipped Indexing access error Indexing error example
Root Cause: User token lacks access to some itemsSolution:
  1. Use filters to exclude restricted items:
    • Jira: Add JQL filter in toolkit configuration
    • Confluence: Add CQL filter in toolkit configuration
  2. Verify token permissions: Ensure API token has read access
  3. Retry indexing: Some failures due to items deleted during indexing
Related: Troubleshooting Guide §11
Symptoms:
  • “Exceeded 30 redirects”
  • Some toolkit tools work, others don’t (e.g., can’t fetch images)
  • Connection repeatedly fails API redirect error
Root Cause: Incorrect API version settingSolution:
  1. Open toolkit configuration
  2. Locate API Version field
  3. Set correct version:
    • Atlassian Cloud: API version 3
    • EPAM Jira/Server: API version 2
  4. Save and test connection
API version settingNote: Partial functionality doesn’t guarantee correct configuration!Related: Troubleshooting Guide §12
Symptoms:
  • “Unauthorized (401)” error
  • “Permission denied” or “Access forbidden”
  • Toolkit connection fails Authentication error
Root Cause: Incorrect authentication configuration or invalid tokensSolution:For Atlassian (Jira/Confluence):
  • Cloud: Use Basic Auth with API key + username
  • Server/EPAM: Use Bearer Token with API key only Authentication setup
For All Toolkits:
  1. Verify credentials in Credentials menu
  2. Click “Test Connection” button
  3. Check for extra spaces in API keys/tokens (common issue!)
  4. Ensure token hasn’t expired
  5. Verify user has necessary permissions
Related: Troubleshooting Guide §13
Symptoms:
  • Toolkit returns small number of results
  • You know more items exist but they’re not shown
  • Lists appear incomplete
Root Cause: Toolkit “Limit” field set too lowSolution:
  1. Go to Toolkits menu
  2. Edit the affected toolkit
  3. Find “Limit” field (may be in Advanced Settings)
  4. Increase value (e.g., from 10 to 100, or 50 to 500)
  5. Save configuration
  6. Retry your query
Toolkit limit configurationRelated: Troubleshooting Guide §14

​
Pipeline Configuration Problems

Symptoms:
  • “Entry point is not defined in the schema”
  • Pipeline fails immediately on start
  • Error appears before any execution Entry point error
Root Cause: No starting node designated in pipelineSolution:
  1. Open your pipeline
  2. Identify the first node that should execute
  3. Click three dots (⋮) on that node
  4. Select “Entry Point” option
  5. Save and retry
Entry point solutionRelated: Troubleshooting Guide §2
Symptoms:
  • Node ignores provided input
  • Responses are general instead of specific to your data
  • Variables seem to not work Nodes not receiving input Variable typo example
Root Cause: Variables not referenced in task fieldSolution:
  1. Open node configuration
  2. In the Task field, add variable references:
    • Using f-string: Analyze this data: {input_data}
    • Using variable syntax: Select variable from dropdown
  3. Verify variable names match exactly (check for typos!)
  4. Variables in “Input” section must be referenced in task to be used
  5. Save and retry
Correct variable usageRelated: Troubleshooting Guide §6
  • Keep API tokens and credentials up to date
  • Monitor token usage and rate limits
  • Use descriptive names for agents, toolkits, and credentials
  • Check the end of error messages for actual error details
  • Look for node names in pipeline errors to identify problem areas
  • Use Test Connection button before assuming toolkit issues
  • Check for typos in variable names and configuration fields
  • Review recent changes to identify what might have broken