SDKs
Python
The vectorless-sdk package, sync and async.
Updated 2026
Install
pip install vectorless-sdkThe import name is vectorless.
Initialize
from vectorless import VectorlessClient
# A local engine needs no API key.
client = VectorlessClient(base_url="http://localhost:7654")
# A server with API-key auth:
# VectorlessClient(base_url="https://your-host", api_key="<your key>")transport="connect" switches from HTTP/REST to ConnectRPC.
AsyncVectorlessClient has the same methods, awaitable.
Ingest and query
result = client.ingest_document("./10-K.pdf")
doc = client.wait_for_ready(result.document_id, on_progress=print)
response = client.query(doc.id, "What were FY2018 capital expenditures?")
for unit in response.sections:
print(unit.id, unit.page, unit.title)
print(response.cited_pages)With Jev navigation, the leading units are evidence pages: id is page_<n>
and page is set. cited_pages lists the page ranges.
The generative page walk
answer = client.answer_treewalk(
doc.id,
"What were FY2018 capital expenditures?",
llm_key="<your model key>", # optional: your own key for this request
)
print(answer.answer)This calls treewalk, not the Jev pipeline, and
returns cited page ranges rather than pincites.
Other calls
get_document, list_documents, delete_document, get_document_tree,
get_llms_txt, get_section, get_sections, query_stream, health,
version.