Skip to main content

Troubleshooting

Common issues and solutions.

Known Limitations

These are expected behaviors, not bugs. See Introduction for the full list.
IssueExpected?Solution
Cold start delay (10-20s)YesFirst request after idle is slow. Subsequent requests are fast.
529 quota errorsYesAuto-retry usually fixes it. If persistent, wait 30-60 seconds.
Web search not workingYesGoogle grounding protocol differs. Use without web search.
High latency (South America, etc.)YesServer is in Singapore. Farther = slower.

Connection Issues

Cause: Cannot reach the proxy server.Solutions:
  1. Check your internet connection
  2. Test the endpoint:
    curl YOUR_ENDPOINT_URL/health
    
  3. If that fails, try again in a few minutes
Cause: Request taking too long.Solutions:
  1. Try once again — could be cold start
  2. If first request, wait 10-20 seconds (cold start is normal)
  3. Check your internet speed
Cause: First request after idle spins up the backend.Solutions:
  1. This is normal — wait 10-20 seconds
  2. Subsequent requests will be fast
  3. If every request is slow, check your internet connection

Authentication Issues

Cause: API key not recognized.Solutions:
  1. Double-check your API key for typos
  2. Ensure no extra spaces before/after the key
  3. Contact admin on Discord: presumptiousbastard
Cause: Authentication failed.Solutions:
  1. Verify your API key is correct
  2. Check the Base URL format:
    • Claude Code: YOUR_ENDPOINT_URL
    • Cursor: YOUR_ENDPOINT_URL/v1
    • Cline/Roo: YOUR_ENDPOINT_URL (no /v1)
  3. Contact admin if key was recently created

Quota & Rate Limit Issues

Cause: Temporary quota limit hit.Solutions:
  1. Simply retry — auto-retry usually works
  2. If persistent, wait 30-60 seconds
  3. Switch to gemini-3-flash temporarily to preserve Claude quota
Cause: Heavy usage in short time.Solutions:
  1. Wait a few minutes
  2. Use gemini-3-flash for simple tasks
  3. If you consistently hit limits, contact admin about usage patterns

Model Issues

Cause: Invalid model name.Solutions:
  1. Use exact model names:
    • claude-opus-4-5-thinking
    • claude-sonnet-4-5-thinking
    • claude-sonnet-4-5
    • gemini-3-flash
  2. See Available Models for full list
Cause: Using model without -thinking suffix.Solutions:
  1. Use claude-opus-4-5-thinking or claude-sonnet-4-5-thinking
  2. Models without -thinking don’t use extended thinking mode

Tool-Specific Issues

Install Claude Code first:
npm install -g @anthropic-ai/claude-code
Your config isn’t loading. Check:
  1. ~/.claude-proxy/settings.json exists
  2. You’re using claude-proxy alias (not just claude)
  3. Run source ~/.zshrc to reload
  1. Close VS Code completely
  2. Reopen and reconfigure
  3. Check VS Code has write permissions
  1. Restart Cursor after adding model
  2. Check model is toggled “on” in settings
  3. Try removing and re-adding the model
Add modalities config to your opencode.json:
{
  "provider": {
    "anthropic": {
      "options": {
        "baseURL": "YOUR_ENDPOINT_URL"
      }
    }
  },
  "modalities": ["text", "image"]
}

Still Stuck?

DM on Discord: presumptiousbastard — happy to help with setup or issues.