ScreenerBot
DocsAssistantTroubleshooting
ScreenerBot
ScreenerBot

Need Help?

Join our community for support, updates, and trading discussions.

Telegram ChannelTelegram GroupFollow on X (Twitter)Contact Support

AI Analysis Troubleshooting

Common issues with AI-powered analysis in ScreenerBot and how to diagnose and resolve them.

Quick Diagnostic Steps

Before You Start

If AI analysis isn't working as expected, follow these steps to diagnose the issue:

  1. Check the Dashboard: Look at the Events page for AI-related errors
  2. Check Console Logs: Run ScreenerBot in terminal to see detailed error messages
  3. Verify Configuration: Ensure your config.toml has correct API keys and settings
  4. Test Provider: Try a simple API call to your provider outside ScreenerBot to verify it works

Common Issues & Solutions

Issue: API Key Invalid or Rejected

Error: "Invalid API key" or "Unauthorized" or "Authentication failed"

Possible Causes:

  • •API key was copied incorrectly (missing characters, extra spaces)
  • •API key was revoked or expired on the provider's dashboard
  • •Using wrong key format for the provider (e.g., using OpenAI key for Groq)
  • •Provider account has billing issues or exceeded quota

Solutions:

  • Log into provider dashboard and verify the API key is active
  • Generate a new API key and update your config.toml
  • Ensure no extra spaces before/after the key in config file
  • Check billing status on provider dashboard (some require payment method even for free tier)

Issue: Rate Limit Exceeded

Error: "Rate limit exceeded" or "Too many requests" or "429 error"

Possible Causes:

  • •Rate limit in config is set higher than provider allows
  • •Too many tokens being screened simultaneously
  • •Provider has tier-based limits and you hit daily/monthly cap

Solutions:

  • Lower rate_limit in config (e.g., Groq free tier: use 20 RPM instead of 30)
  • Configure fallback providers: filtering_providers = ["groq", "deepseek"]
  • Upgrade to paid tier on provider dashboard for higher limits
  • Wait until rate limit window resets (typically 1 minute for RPM limits)

Issue: AI Analysis is Very Slow

Symptom: AI filtering takes 5-30 seconds per token, slowing down screening

Possible Causes:

  • •Using slow provider (e.g., OpenAI GPT-4 instead of Groq)
  • •Timeout setting is too high, waiting for unresponsive provider
  • •Network latency or slow internet connection
  • •Using large/complex model (e.g., GPT-4o instead of GPT-4o-mini)

Solutions:

  • Switch to Groq (fastest) or DeepSeek for filtering: filtering_provider = "groq"
  • Lower timeout: timeout = 5 instead of timeout = 30
  • Use smaller models: gpt-4o-mini instead of gpt-4o
  • For Ollama (local), use smaller models: llama3.2 instead of llama3.1:70b

Issue: AI Analysis Not Running

Symptom: Tokens are being filtered but AI analysis doesn't appear to run

Possible Causes:

  • •AI filtering not enabled in config: filtering_enabled = false
  • •Provider not configured or disabled: enabled = false
  • •Tokens failing numerical filters before reaching AI analysis
  • •ScreenerBot not restarted after config changes

Solutions:

  • Verify [ai.analysis] filtering_enabled = true in config
  • Ensure provider is enabled: [ai.providers.groq] enabled = true
  • Restart ScreenerBot after config changes (AI config requires restart)
  • Check Events dashboard for AI-related logs to confirm it's running

Issue: Request Timeout Errors

Error: "Request timed out" or "Connection timeout"

Possible Causes:

  • •Provider is experiencing downtime or high load
  • •Network connectivity issues on your end
  • •Timeout setting is too low for the provider/model
  • •For Ollama: local model is too large for your hardware

Solutions:

  • Increase timeout: timeout = 30 instead of timeout = 5
  • Configure fallback providers to handle downtime automatically
  • Check provider status page for outages (e.g., status.groq.com)
  • For Ollama: use smaller model or allocate more RAM to Ollama

Issue: AI Gives Incorrect Recommendations

Symptom: AI marks good tokens as scams or approves obvious scams

Possible Causes:

  • •Using small/weak model (e.g., llama3.2 instead of llama3.1-70b)
  • •Model hasn't been trained on crypto/DeFi scam patterns
  • •Scam is using novel technique AI hasn't seen before
  • •Confidence threshold for auto-blacklist is too low

Solutions:

  • Upgrade to larger/better model: GPT-4o-mini, Claude-3.5-Sonnet, or llama-3.1-70b
  • Use advisory mode (filtering_enforce = false) and manually review
  • Increase auto-blacklist confidence: auto_blacklist_confidence = 0.9
  • Combine AI with RugCheck for better scam detection coverage

Provider-Specific Issues

Groq

Common Issue: Rate limits hit quickly

Free tier: 30 RPM. With multiple tokens screening simultaneously, easy to hit limit.

Solution: Set rate_limit = 20 or configure DeepSeek as fallback.

Ollama (Local)

Common Issue: "Connection refused"

Ollama service not running or wrong base_url.

Solution: Run ollama serve in terminal. Set base_url = "http://localhost:11434" in config.

OpenAI

Common Issue: Expensive costs

GPT-4o can cost $0.01+ per analysis. With 100 tokens/day, that's $1/day.

Solution: Use gpt-4o-mini (10x cheaper) or switch to Groq/DeepSeek for filtering.

DeepSeek

Common Issue: Slower than Groq

DeepSeek has generous quotas but 2-3x slower response time than Groq.

Solution: Use Groq for filtering (speed), DeepSeek for entry/exit analysis (less time-sensitive).

Debugging Tips

  • Enable Verbose Logging:Run ScreenerBot with RUST_LOG=debug environment variable to see detailed AI request/response logs.
  • Test Provider Independently:Use curl or Postman to test API calls to your provider outside ScreenerBot. Confirms if issue is with provider or ScreenerBot config.
  • Check Events Dashboard:The Events page in dashboard shows all AI-related events, errors, and analysis results. Useful for diagnosing what AI is doing.
  • Restart After Config Changes:AI provider configuration requires a full restart. Changes to analysis features (filtering_enabled, etc.) may hot-reload depending on the setting.

Still Need Help?

Get Support

If you're still experiencing issues after trying these solutions:

  • •Join the Discord community for real-time help from other users
  • •Check GitHub Issues for known bugs or submit a new issue
  • •Review provider documentation for provider-specific issues
  • •Include error logs, config (without API keys), and provider name when asking for help

Next Steps

Configuration Guide

Review configuration settings to ensure everything is set up correctly.

Provider Reference

Detailed setup instructions and troubleshooting for each provider.

Best Practices

Learn optimal configuration and usage patterns for AI analysis.