Unify LLM icon

Unify LLM

LangChain-style orchestration and multi-provider intelligence via Unify LLM

Actions10

Vector Store → Similarity Search

AI-generated

Summary

Search stored vectors for similarity to the given query prompt using a deterministic embedding.

Inputs

  • prompt (required) — Prompt or query text used to generate a deterministic embedding for the search.
  • vectorBackend — Choose where vectors are persisted between executions: localCache, Qdrant, or Pinecone.
  • vectorDimension — Deterministic embedding dimension used for generating the query vector; must match the dimension used during upsert.
  • topK — Maximum number of similar results to return.
  • includeScores — Whether to include cosine similarity scores in each result item.
  • cacheNamespace — Namespace key for the local persistent cache. Only used when vectorBackend is localCache.
  • qdrantUrl — Qdrant server URL. Only used when vectorBackend is qdrant.
  • qdrantCollection — Qdrant collection name. Only used when vectorBackend is qdrant.
  • qdrantApiKey — Optional API key for Qdrant. Only used when vectorBackend is qdrant; not required for local/self-hosted instances.
  • pineconeHost — Pinecone index host. Only used when vectorBackend is pinecone.
  • pineconeApiKey — Optional API key for Pinecone. Only used when vectorBackend is pinecone.
  • pineconeNamespace — Pinecone namespace. Only used when vectorBackend is pinecone.

Output shape

{ resource, operation, backend, query, topK, results }

The exact return JSON varies by backend: localCache adds vectorStoreSize at the top level; qdrant and pinecone return the base shape without vectorStoreSize. results array items contain id, content, metadata, and optionally score (present only if includeScores is true).

Examples

Example 1: Similarity search against local cache

vectorBackend=localCache, cacheNamespace=myData, topK=5, includeScores=true, prompt='find similar documents'

Links

Discussion