Smarter Vectorsearch
Smarter Vectorsearch is a Django application that provides a framework for executing semantic search queries against locally-hosted vector databases managed by Smarter Vectorstore. It is fully integrated into the Smarter Framework, namely the Smarter Application Manifest (SAM) yaml manifest-based API.
A Vectorsearch is a named, ownable configuration describing how to query a single Vectorstore: which retrieval strategy to use (similarity, similarity with a score threshold, or maximal marginal relevance), how many results to return, and any metadata filters to apply. It does not itself hold embeddings or documents, and it does not generate text – it is purely the retrieval half of a retrieval-augmented generation pipeline. Combining a Vectorsearch with a Smarter LLMClient yields a complete RAG solution: the top-k results returned by the Vectorsearch are injected into the LLMClient’s system prompt at inference time, grounding the model’s response in the underlying Vectorstore’s content.
The module follows the same operational conventions as the rest of the Smarter Framework, using feature flags (waffle switches) to control logging behavior, and exposing full manifest-driven lifecycle management – create, apply, describe, and delete – through the SAMVectorsearchBroker. Because a Vectorsearch’s Vectorstore and auth Secret are referenced by name in the manifest spec but held as foreign keys on the underlying ORM model, the broker is also responsible for resolving those references against the owning account before every create, update, or read. Overall, the Smarter Vectorsearch app is architected to be a thin, reusable retrieval layer that any LLMClient in the Smarter Framework can compose into a RAG pipeline without re-implementing search logic on a per-integration basis.