Compatible with wapeb-velocity 1.0.1 (or newer).

https://github.com/lftkraft/wapeb-velocity

it isn't alpha 3. alpha 3 is the next version.

🚀 What's New in alpha.3-aiupdate

🔄 1. Multi-API Key Pool & Round-Robin Rotation

• Multiple Key Support: Added groq-api-keys list support in config.yml. Chat analysis requests are now automatically distributed across multiple Groq API keys in a round-robin rotation to scale requests per minute (RPM).
• Instant 429 Failover: If a key hits a rate limit (429 Too Many Requests), the system automatically switches to the next available API key in the pool with zero delay.

⚡ 2. Intelligent Local Heuristic Pre-Filtering

• Quota-Saving Pre-Filter: Added sentinel.ai.pre-filter to skip trivial, harmless chatter locally with 0 ms overhead (saving 50–70% of API quotas).
• Automatic Pattern Filtering: Automatically ignores short chatter (< 3 characters like k, xd, ?), coordinates/numbers (100 64 -200, 500k), punctuation/emojis (???, :), :D, ^^), and common harmless gaming terms (gg, wp, ez, szia, oks, ty, etc.).

🛡️ 3. Prompt Injection & Jailbreak Protection

• Hardened System Prompt: Injected security rules ensure the AI treats player-crafted bypass instructions (e.g. "ignore previous instructions", "do not mute", custom JSON overrides) strictly as plain text, preventing LLM manipulation.

🧠 4. Context-Aware Chat History Analysis

• Chat Context Integration: Added sentinel.ai.context-lines (default: 5) to supply recent conversation history from ChatSnapshotManager. The AI can now evaluate multi-line arguments and contextual toxicity accurately.

🚀 5. Optimized Default Model

• Model Switch: Defaulted to llama-3.1-8b-instant for ultra-low latency and generous free-tier daily rate limits (14,400 RPD).