Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.kontext21.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Performance Issues

If X21 is responding slowly or experiencing performance issues, this guide helps identify and resolve the cause.

​
Slow AI Responses

​
Symptoms

  • Long wait for response to start
  • Slow streaming of text
  • Delayed tool execution
  • Timeouts

​
Common Causes & Solutions

1. Token Limit Approaching Cause: Conversation near 200,000 token limit Check: Look at token counter in status bar Solution: 
📊 Tokens: 185,000 / 200,000 ← Close to limit
  • Start a new conversation
  • X21 will auto-compact, but new chat is faster
2. Large Attachments Cause: Processing big PDFs or many images Check: File size and page count Solution:
  • Reduce PDF to relevant pages only
  • Compress images
  • Attach fewer files per request
  • Split large documents
3. Complex Operations Cause: Operation requires extensive thinking Check: Thinking block duration Solution:
  • Break into smaller steps
  • Simplify prompt
  • Provide more specific instructions
  • Use slash commands for common tasks
4. Network Issues Cause: Slow internet connection to API Check: Network speed, other apps Solution:
  • Check internet connection
  • Close bandwidth-heavy applications
  • Wait for better connectivity
  • Try smaller requests

​
When to Start New Conversation

​
Signs You Should Start Fresh

  1. Token usage high (over 150,000 tokens)
  2. Responses getting slower over time
  3. Unrelated to previous conversation
  4. Context no longer relevant
  5. Performance degraded

​
Benefits of New Conversation

  • Faster responses: Less history to process
  • Better focus: Clear context for AI
  • Token efficiency: Start with low usage
  • Organized history: Easier to find later

​
How to Start New

  1. Click New Chat button
  2. Confirms if ongoing requests
  3. Previous conversation auto-saved
  4. Fresh start with full token budget

​
Conversation Compacting

​
What It Does

When conversation reaches 200,000 tokens:
  1. Summarizes old messages
  2. Preserves recent context
  3. Reduces token usage
  4. Continues seamlessly

​
When It Happens

  • Automatic at token limit
  • No user action required
  • Brief pause during compacting
  • Notice appears in conversation

​
Impact

Benefits:
  • Continue same conversation
  • Maintain context
  • No data loss
Trade-offs:
  • Old details summarized
  • Some nuance may be lost
  • Still slower than new conversation
Recommendation: Start new conversation instead of relying on compacting for better performance.

​
Tool Execution Speed

​
Slow Tool Execution

Causes:
  • Large data ranges
  • Many concurrent tools
  • Complex VBA operations
  • Excel busy with other tasks
Solutions:
  1. Batch operations more efficiently
  2. Smaller ranges per operation
  3. Sequential execution rather than parallel
  4. Close other Excel workbooks
  5. Disable Excel add-ins temporarily

​
Tool Optimization

Efficient patterns: 
✓ Single write_values with 2D array
✗ Multiple write_values for individual cells

✓ Batch formatting operations
✗ Format one cell at a time

✓ Use drag_formula for patterns
✗ Write each formula individually

​
Workbook Performance

​
Large Workbooks

Impact: Slower Excel operations Solutions:
  • Work on smaller worksheets
  • Clear unused data/formatting
  • Disable automatic calculations temporarily
  • Close other workbooks
  • Archive old data

​
External References

Impact: Slower when referencing other workbooks Solutions:
  • Break external links if not needed
  • Copy data into current workbook
  • Close source workbooks

​
System Performance

​
Resource Usage

Check:
  • Task Manager (Windows)
  • Activity Monitor (Mac)
  • Look for high CPU/memory usage
Optimize:
  • Close unused applications
  • Restart Excel if memory high
  • Restart Deno server
  • Reboot computer if needed

​
Disk Space

Impact: Low disk space slows everything Check: Available disk space Solution:
  • Ensure 5+ GB free space
  • Clear temp files
  • Archive old log files
  • Clean up downloads

​
Network Performance

​
Checking Connection

Test:
  1. Visit Anthropic status
  2. Check general internet speed
  3. Test other online services
If slow:
  • Wait for better connectivity
  • Work offline where possible
  • Schedule large operations for better network times

​
Port Issues

Check ports:
  • Port 8000: Deno server
  • Port 8080: VSTO add-in
  • Ensure not blocked by firewall

​
Monitoring Performance

​
Token Usage

Watch the token counter: 
📊 Tokens: X / 200,000
Guidelines:
  • fewer than 50,000: Excellent
  • 50,000-100,000: Good
  • 100,000-150,000: Consider new chat soon
  • 150,000-200,000: Start new chat for best performance
  • over 200,000: Auto-compacting occurs

​
Response Times

Expected:
  • First response: 2-5 seconds
  • Streaming: Continuous flow
  • Tool execution: 1-3 seconds per tool
  • Thinking: 1-10 seconds depending on complexity
Concerning:
  • First response: >10 seconds
  • Streaming: Stuttering or stops
  • Tool execution: >10 seconds
  • Thinking: >30 seconds

​
Best Practices

​
Proactive Management

  1. Start new conversations for new tasks
  2. Monitor token usage regularly
  3. Batch similar operations
  4. Optimize attachments before sending
  5. Close unused workbooks

​
Efficient Workflows

Good: 
1. New conversation
2. Specific prompt
3. Batch operations
4. Review and approve
5. New conversation for next task
Avoid: 
1. Long-running conversation
2. Mix unrelated tasks
3. Large attachments
4. Individual operations
5. Letting tokens max out

​
When Performance Matters

For time-sensitive work:
  • Start fresh conversation
  • Use slash commands
  • Enable auto-approve
  • Prepare data in advance
  • Work during off-peak hours

​
Related Topics