vectorless-sdk 1.0.0

Official Python SDK for Vectorless — structure-preserving document retrieval without embeddings

Vectorless Python SDK

Official Python client for Vectorless — structure-preserving document retrieval without embeddings.

Install

pip install vectorless-sdk

Quick Start

from vectorless import VectorlessClient

# Connect to a deployed instance
client = VectorlessClient(
    base_url="https://api.vectorless.dev",
    api_key="vl_live_...",
)

# Or self-hosted (no API key needed)
client = VectorlessClient(base_url="http://localhost:8080")

# Ingest a document
result = client.ingest_document(
    "./report.pdf",
    filename="report.pdf",
    metadata={"team": "engineering"},
)

# Wait for processing
doc = client.wait_for_ready(result.document_id)

# Get the document tree
tree = client.get_document_tree(doc.id)
for section in tree.sections:
    print("  " * section.depth + section.title)

# Query — LLM navigates the tree to find relevant sections
response = client.query(doc.id, "How does authentication work?")
for section in response.sections:
    print(f"[{section.title}]")
    print(section.content)
print(f"Strategy: {response.strategy}, Time: {response.elapsed_ms}ms")

Async Support

from vectorless import AsyncVectorlessClient

async with AsyncVectorlessClient(
    base_url="https://api.vectorless.dev",
    api_key="vl_live_...",
) as client:
    result = await client.ingest_document(b"# Hello\n\nWorld")
    doc = await client.wait_for_ready(result.document_id)
    response = await client.query(doc.id, "What is this about?")

Transport Protocols

The SDK supports two wire protocols:

# HTTP/REST (default) — standard REST endpoints
client = VectorlessClient(base_url="http://localhost:8080")

# ConnectRPC — protobuf JSON encoding, native streaming
client = VectorlessClient(base_url="http://localhost:8080", transport="connect")

Streaming Queries

for event in client.query_stream(doc_id, "Explain clustering"):
    if event.type == "section_selected" and event.section:
        print(f"Found: {event.section.title}")
    if event.type == "completed":
        print(f"Done in {event.elapsed_ms}ms")

API Reference

Method	Description
ingest_document()	Upload and ingest a document
get_document()	Get document metadata and status
list_documents()	List documents with pagination
delete_document()	Delete a document
wait_for_ready()	Poll until document is processed
get_document_tree()	Get hierarchical document outline
get_section()	Fetch a section with full content
get_sections()	Fetch multiple sections
query()	Query a document
query_stream()	Stream query results
health()	Server health check
version()	Server version

License

MIT