Problem
The system creates an embedding provider and a VectorIndex, but only SearchIndex (BM25) is populated during the compress pipeline. There is no code path that automatically generates embeddings and adds them to the vector index.
The HybridSearch can search both BM25 and vector indices, but the vector index will always be empty unless manually populated.
Locations
src/functions/compress.ts — only adds to BM25 search index, never to vector indexsrc/state/vector-index.ts — exists but never populatedsrc/state/hybrid-search.ts — falls back to BM25-only when vector index is empty
Suggested Fix
Add embedding generation in the compress pipeline when an embedding provider is available. After compressing an observation, generate an embedding from its title + facts + narrative, then add to the vector index.
Problem
The system creates an embedding provider and a
VectorIndex, but onlySearchIndex(BM25) is populated during thecompresspipeline. There is no code path that automatically generates embeddings and adds them to the vector index.The
HybridSearchcan search both BM25 and vector indices, but the vector index will always be empty unless manually populated.Locations
src/functions/compress.ts— only adds to BM25 search index, never to vector indexsrc/state/vector-index.ts— exists but never populatedsrc/state/hybrid-search.ts— falls back to BM25-only when vector index is emptySuggested Fix
Add embedding generation in the compress pipeline when an embedding provider is available. After compressing an observation, generate an embedding from its title + facts + narrative, then add to the vector index.