Connecting to Libra Backend...

👋 Welcome to Libra!

This is your personal AI Assistant and educational Large Language Model laboratory, built from first principles.

Core System Capabilities:

  • First-Principles Engine: Modern decoder-only transformer with RoPE, RMSNorm, SwiGLU, and weight tying.
  • Multi-Turn Persistent Memory: Powered by SQLite WAL with dynamic sliding-window context management.
  • Advanced Hybrid RAG: Okapi BM25 sparse search + Dense vector embeddings fused via Reciprocal Rank Fusion (RRF $k=60$), multi-factor re-ranking, and chunk deduplication.
  • Real-Time Streaming UX: Live token-by-token Markdown parsing, code syntax highlighting, one-click code copy, and abortable generation.
python
1# Example: Grounded RAG Query in Libra
2response = await libra.chat(
3    query="Explain Rotary Position Embeddings",
4    mode="hybrid",
5    use_rrf=True
6)
7print(response.content)

Ask any question or test your knowledge base to begin!

Ready