Knowledge Base (RAG)

Built-in Retrieval-Augmented Generation. Upload markdown files and Axora automatically enhances chat responses with relevant context.

How It Works

knowledge/*.md→chunking + embeddings
↓
knowledge_chunks→PostgreSQL (embeddings as float arrays)
↓
Hybrid Search→70% semantic + 30% keyword
↓
Re-ranking→relevance scoring
↓
System Prompt→injected as context

Adding Documents

# Add markdown files to the knowledge directory
cp my-docs/api-guide.md knowledge/

# Files are auto-detected and ingested
# No restart required — file watcher handles changes

Embeddings

Embeddings come from OpenAI, Gemini, or HuggingFace (default: huggingface/sentence-transformers/all-MiniLM-L6-v2, 384 dimensions). If no provider key is configured, Axora falls back to hash-based pseudo-embeddings.