Repository navigation
Releases: rillToMe/Changli
Releases · rillToMe/Changli
Release list
NewFeatures & Add New Bug
🚀 v1.1.0
Major feature update introducing multi-provider AI, image analysis, and smarter utilities.
✨ Added
- Image analysis via Mistral Vision
- YouTube transcript summarizer (
.ai yt-rs) - Cerebras LLM integration (
gpt-oss-120b) - Multi-key rotation for Mistral & Cerebras
- Quoted message context support
- Reminder system (
.ai remind) - Command descriptions & usage in
.ai help
🔧 Improved
- Automatic AI routing (image → Mistral, text → Groq)
- Baileys v7 ESM compatibility
- Auto-reconnect handling
- Linux CRLF fix for
.env
🐛 Fixed
- Baileys v7 connection issues (CJS vs ESM)
- Linux command parsing bug
- Image messages not processed
- Missing destructure error in AI fallback
Changli
🚀 Changli AI Bot v1.0.0
Initial public release of Changli - a modular AI-powered WhatsApp bot built on Baileys v7.
This version marks the first stable milestone with a complete AI architecture, middleware system, and multi-key LLM integration.
✨ Highlights
🧠 AI Core
- Groq LLM integration
- Multi API key rotation with automatic cooldown
- Smart context builder
- Per-user memory system
- Explicit
.aimode - Command isolation (no memory pollution)
- Token usage tracking
⚙️ Modular Architecture
- Auto command loader
- Middleware chain:
- Permission
- Cooldown
- AI fallback
- Clean handler separation
- Scalable structure
🔐 Access Control
- Admin whitelist
- Group restriction support
- Role-based command permissions
- Anti-spam cooldown system
🛠 Utilities
.ai ping.ai help.ai clear.ai memory
📁 Stability
- Safe JSON memory persistence
- Auto-create logs directory
- Linux + Windows compatibility
- Clean socket lifecycle handling
🏗 Technical Stack
- Node.js 20+
- Baileys v7.x
- Groq LLM API
- Express (Admin Dashboard)
- Pino Logger
🎯 Project Vision
Changli is designed as a lightweight but extensible AI assistant for small WhatsApp groups, focusing on:
- Intelligent conversation flow
- Controlled AI token usage
- Modular and maintainable architecture
- Scalable multi-key LLM strategy
⚠ Notes
- Requires active WhatsApp account
- Requires Groq API keys
- Node 20+ recommended
This release establishes the foundation for future expansions including:
- Advanced mini utilities
- Tone classifier
- Spam detection
- Plugin system
- Real-time dashboard updates
- Advanced RAG system
First stable milestone. More refinement ahead.