Vector Stores¶
lionagi does not include built-in vector store integration. However, its data primitives support embedding vectors, and you can connect any vector store through tools.
Node Embeddings¶
The Node class (used throughout lionagi's graph system) has an optional embedding field:
from lionagi.protocols.graph.node import Node
node = Node(
content="Some text to embed",
embedding=[0.1, 0.2, 0.3, ...], # Your embedding vector
)
This lets you store embeddings alongside content in lionagi's graph structures, but querying and indexing are left to your vector store of choice.
Connecting a Vector Store as a Tool¶
The recommended approach is to wrap your vector store operations as tool functions:
from lionagi import Branch
async def vector_search(query: str, top_k: int = 5) -> str:
"""Search the vector database for relevant documents.
Args:
query: The search query.
top_k: Number of results to return.
"""
# Your vector store client (Qdrant, Pinecone, Chroma, etc.)
results = await your_vector_client.search(query, limit=top_k)
return "\n".join(r.text for r in results)
branch = Branch(
tools=[vector_search],
system="Search the knowledge base before answering questions."
)
result = await branch.ReAct(
instruct={"instruction": "What are the latest findings on climate change?"},
max_extensions=2,
)
This approach works with any vector store and keeps lionagi decoupled from specific database choices.