Retrieval & embeddings

AthenaDB unifies retrieval modes behind one hybrid query surface:

• Keyword / full-text — classic inverted-index search
• Vector — semantic similarity over embeddings (HNSW / IVF, quantization)
• Graph — entity relationships
• Timeline — time-ordered retrieval

A single query can blend these rather than forcing one index type.

Embeddings

A provider abstraction sits over external embedding services (for example Text-Embeddings-Inference / BGE-M3) and native Athena embeddings, with re-embedding pipelines. This is available for Early Access evaluation while platform parity evidence completes.

Native generative AI

In-database LLM generation is Planned (separate from embeddings). See Native AI runtime.

Related: Architecture · Lite