Expand description
Connect document and vector frameworks through the public API or CLI.
§Framework integrations
The core integration boundary is intentionally small: frameworks can store
Document values, run TextQuery values, or use vectors directly through
CollectionHandle. An adapter only needs to map its document ID, text, metadata,
and embedding provider onto those types.
For LangChain or LlamaIndex today, use their Rust document/chunking layer with a
small application adapter that calls upsert_documents and query_batch. Keep
the provider’s model and revision in Embedder::model_id; git-vdb persists and
checks that identity on every reopen.
For language-agnostic pipelines, use the JSON CLI:
producer | git-vdb --db vectors.git upsert docs - --batch-size 1000
git-vdb --db vectors.git search docs --vector '[0.1,0.2]' --format jsonThe CLI writes result data only to stdout and progress or errors to stderr, so it is safe to compose with subprocess-based tools.