Swarmauri Query Image Vector Store Tool

swarmauri_tool_queryimagevectorstore queries image-vector indexes with either an image input or a precomputed numeric embedding and returns structured ranked matches.

Features

• Accepts exactly one image or embedding per query.
• Converts images through an explicit infer_vector(image) adapter.
• Supports retrieve_by_vector and similarity_search_by_vector stores.
• Preserves identifiers, scores, content, URIs, and metadata.
• Applies exact-match metadata filters and bounded result counts.

Installation

uv add swarmauri_tool_queryimagevectorstore

pip install swarmauri_tool_queryimagevectorstore

Usage

from swarmauri_tool_queryimagevectorstore import QueryImageVectorStoreTool

tool = QueryImageVectorStoreTool(
    vector_store=my_image_index,
    image_embedder=my_vision_embedder,
)

matches = tool(
    image=image_payload,
    top_k=5,
    metadata_filter={"category": "diagram"},
)

For precomputed embeddings, call tool(embedding=[...], top_k=5). The vector store must expose retrieve_by_vector(embedding, top_k=...) or similarity_search_by_vector(embedding, top_k=...).

Project Links

• Swarmauri SDK
• Contributing guide

License

Apache-2.0.