Getting Started
Dashboard Guides
Trading Guides
Reference
Telegram Integration
Need Help?
Join our community for support, updates, and trading discussions.
AI-Powered Filtering
Use AI to automatically analyze and filter tokens during the screening process. Detect scams, assess token quality, and make smarter filtering decisions with ScreenerBot's AI integration.
What is AI Filtering?
Intelligent Token Analysis
AI Filtering adds an intelligent layer to your token screening process. Instead of relying purely on numerical thresholds (liquidity, volume, holder count), the AI analyzes tokens holistically by considering:
- •Token Metadata Quality - Name, symbol, description patterns that indicate scams
- •Holder Distribution - Concentration of holdings, wallet patterns
- •Liquidity Patterns - Liquidity to market cap ratios, sudden changes
- •Volume Analysis - Organic vs wash trading indicators
- •Historical Context - Similar token patterns from past scams
The AI provides a pass/fail recommendation that can override your numerical filters, preventing risky tokens from reaching your trading engine.
How It Works
Token Discovery
ScreenerBot discovers a new token through your configured sources (DEX pools, billboards, etc.)
Numerical Filtering
First, your standard numerical filters are applied (liquidity > X, volume > Y, holders > Z)
AI Analysis
If the token passes numerical filters, it's sent to your configured AI provider for analysis
Decision
The AI returns PASS, FAIL, or UNCERTAIN. Based on your settings, the token either proceeds to trading consideration or gets rejected/blacklisted.
Enabling AI Filtering
Configure AI filtering in your config.toml:
# AI Filtering Configuration
[ai.analysis]
filtering_enabled = true
filtering_provider = "groq" # or "deepseek", "openai", etc.
# Optional: Configure fallback providers
filtering_providers = ["groq", "deepseek", "gemini"]
# Auto-blacklist tokens that AI marks as scams
auto_blacklist_enabled = true
auto_blacklist_confidence = 0.8 # 0.0-1.0 thresholdFiltering Modes
Advisory Mode (Default)
AI provides recommendations but doesn't block tokens. You see AI analysis in the dashboard but final decision is yours.
filtering_enforce = falseEnforcement Mode
AI recommendations are enforced. Tokens marked as FAIL are automatically rejected and don't proceed to trading consideration.
filtering_enforce = trueAuto-Blacklisting
When enabled, tokens that AI identifies as scams with high confidence are automatically added to your blacklist, preventing future consideration.
Confidence Threshold
Set how confident the AI must be before auto-blacklisting:
- •
0.6- Aggressive (blacklist on suspicion) - •
0.8- Balanced (recommended) - •
0.95- Conservative (only obvious scams)
What AI Analyzes
Red Flags (Fail Indicators)
- Suspicious token names (pump, moon, scam keywords)
- Missing or low-quality metadata (no image, no description)
- Extreme holder concentration (>90% in top 10)
- Very low liquidity relative to market cap
- Wash trading patterns (high volume, few unique traders)
- Known scam patterns from historical data
Green Flags (Pass Indicators)
- Professional metadata (complete info, proper branding)
- Healthy holder distribution (<50% in top 10)
- Adequate liquidity (good liquidity/mcap ratio)
- Organic volume patterns (diverse traders)
- Valid social links and community presence
- Positive indicators from RugCheck integration
Real-World Example
Scenario: New Token Discovered
Token: "MOONSHOT 🚀"
Liquidity: $50,000 USD
Volume (24h): $200,000 USD
Holders: 500
Top 10 Holders: 85% of supply
Passes Numerical Filters
Liquidity > $10k ✓, Volume > $50k ✓, Holders > 100 ✓
AI Analysis Triggered
Token sent to AI for quality check. AI notices suspicious name ("MOONSHOT"), high holder concentration (85% in top 10), and disproportionate volume to liquidity ratio.
AI Recommendation: FAIL
Confidence: 0.92. Token is likely a pump & dump scheme. With auto-blacklist enabled, token is added to blacklist and rejected from trading consideration.
Result: Potential loss avoided. Despite passing numerical filters, AI caught the scam pattern and prevented a risky trade.
Best Practices
- Start in Advisory Mode:Use
filtering_enforce = falseinitially to observe AI recommendations without blocking tokens. Adjust after you trust the results. - Use Fast Providers:For filtering, speed matters. Use Groq or DeepSeek for sub-second analysis. Premium providers like GPT-4o-mini offer better accuracy but slower responses.
- Configure Fallbacks:Set up 2-3 providers in
filtering_providers. If your primary hits rate limits, the bot automatically tries the next provider. - Review Blacklist Periodically:Auto-blacklisted tokens are saved to your blacklist file. Review periodically to ensure no false positives. You can manually remove tokens if needed.
- Combine with RugCheck:AI filtering works best alongside RugCheck integration. AI catches social/metadata scams, RugCheck catches technical risks (mintable, freezable, top holder concentration).
Limitations & Considerations
AI Can Make Mistakes
AI is not perfect. It can flag legitimate tokens (false positive) or miss sophisticated scams (false negative). Always combine AI with your own judgment and risk management.
API Costs Apply
Every AI filtering call counts toward your API quota. With high token volumes, costs can add up. Use free providers like Groq (30 RPM) or DeepSeek (500K/day) to minimize costs.
Slows Down Filtering
AI analysis adds latency (100ms-2s per token depending on provider). In fast-moving markets, this delay could impact entry timing. Balance thoroughness with speed.
Model Quality Matters
Smaller/cheaper models (llama-3.2, mistral-small) are faster but less accurate. Larger models (GPT-4o-mini, Claude-3.5-Sonnet) are slower but catch more subtle scam patterns.