seshat-embedding

Embedding provider abstraction with a built-in local provider for Seshat.

When the [embedding] section is present in seshat.toml and the crate is compiled with the builtin-embeddings feature (enabled by default), embeddings are generated locally using fastembed-rs (all-MiniLM-L6-v2 model, 384 dimensions) — no external services needed.

When the section is absent or the feature is disabled, all embedding code is compiled away with zero overhead.

Configuration

# seshat.toml — uncomment to enable vector search
# [embedding]
# model = ""          # empty → provider default (all-MiniLM-L6-v2)
# dimension = 0       # 0     → provider default (384)
# batch_size = 32