Features
Multi-agent simulation with distinct personality traits
LightRAG knowledge graph for conversation memory
Resonance forecasting to predict message longevity
Drift detection and semantic shift alerts
Local LLM support via Ollama (Mistral, Llama 3, Qwen)
Web-based reactive UI dashboard
Animated node-link diagram visualization
Multilingual simulation in 15+ languages
Long-running 24/7 simulation endurance
Export simulations as JSON, CSV, animated GIF
Replay scenarios with modified parameters
Seed message parsing for entities and valence
Population generation with synthetic personas
Intervention sandbox for testing corrections
Offline operation without cloud dependencies
Knowledge graph extraction from raw text
Hierarchical community clustering (Leiden)
Community summary generation (bottom-up)
Global search using community summaries
Local search by entity neighborhood fan-out
DRIFT search blending local and community context
Basic search (vector similarity fallback)
Prompt auto-tuning for better out-of-box results
Indexing pipeline with TextUnits
Entity, relationship, and claim extraction
Configuration file management (YAML)
Versioning and migration support
Initial public release under MIT License
