PaveDB Python SDK

Python SDK package for the PaveDB /v1 API.

Use pavedb-sdk when your code should talk to PaveDB from Python.

There are three runtime paths:

• Connect to a PaveDB server over HTTP.
• Install pavedb alongside the SDK and use the same Client / Collection handle API with a local embedded engine.
• Use ephemeral local mode for temporary in-process stores during tests, notebooks, and short-lived experiments.

To run your own server instance, use the PaveDB core repository: GitLab / GitHub. The core repository remains the source of truth for the OpenAPI contract.

SDK source lives on GitLab and GitHub.

Install

pip install pavedb-sdk

SDK 0.1.x targets the PaveDB /v1 API. SDK package versions are independent from PaveDB core release versions; use pavesdk.__version__ for the SDK release and pavesdk.PAVEDB_API_PREFIX for the wire API.

For local embedded/persisted mode:

pip install pavedb-sdk pavedb

Local Package Build

Build the local PyPI package artifacts with GNU Make:

gmake package

That creates the source distribution (.tar.gz) and wheel in dist/, checks them with Twine, and copies them to artifacts/.

Upload targets are explicit and do not infer release channels from the version:

gmake pypitest-push
gmake pypi-push

Runnable HTTP examples are installed with the package:

python -m pavesdk.examples.http_search
python -m pavesdk.examples.observability

The source distribution also includes the book companion programs under examples/. They are intentionally not installed in the wheel; start with examples/1-intuition/README.md after installing pavedb alongside the SDK.

The SDK checkout also includes demo/20k_leagues.txt, which the examples use via a hardcoded relative path.

Generated API Reference

The source distribution includes its generated API reference, examples index, and generator. From a checkout or unpacked SDK sdist, regenerate them with:

python docs/generate_reference.py

gmake docs-check verifies that the checked-in Markdown is current, every indexed example imports, and each one keeps its python -m pavesdk.examples... command.

HTTP Client

from pavesdk.client import connect

db = connect(
    "http://localhost:8086",
    api_key="super-sekret",
    tenant="demo",
)
books = db.collection("books")

hits = books.search("captain nemo", k=3)
hits

[
    {
        "id": "note-1:0000",
        "score": 0.86,
        "text": "Captain Nemo commands the Nautilus.",
        "meta": {"docid": "note-1", "kind": "note"},
    },
    {
        "id": "note-2:0000",
        "score": 0.73,
        "text": "The Nautilus dives beneath the ice.",
        "meta": {"docid": "note-2", "kind": "note"},
    },
]

for hit in hits:
    print(hit["score"], hit["meta"]["docid"], hit["text"][:80])

0.86 note-1 Captain Nemo commands the Nautilus.
0.73 note-2 The Nautilus dives beneath the ice.

connect("http://...") and connect("https://...") create an HttpClient. Bare paths are local targets and require pavedb to be installed.

Collections

The API is handle-based: pick a collection once, then call methods on it.

from pavesdk.client import connect

db = connect("http://localhost:8086", api_key="super-sekret")
books = db.create_collection("books", tenant="demo")

books.add(
    "Captain Nemo commands the Nautilus.",
    docid="note-1",
    metadata={"kind": "note"},
)
books.add_many([
    ("The Nautilus dives beneath the ice.", "note-2", None),
    {
        "text": "Nemo studies ocean currents.",
        "docid": "note-3",
        "metadata": {"kind": "note"},
    },
])

matches = books.search(
    "submarine captain",
    k=5,
    filters={"kind": "note"},
)
matches

[
    {
        "id": "note-1:0000",
        "score": 0.81,
        "text": "Captain Nemo commands the Nautilus.",
        "meta": {"docid": "note-1", "kind": "note"},
    },
    {
        "id": "note-3:0000",
        "score": 0.69,
        "text": "Nemo studies ocean currents.",
        "meta": {"docid": "note-3", "kind": "note"},
    },
]

Observability

Searches are logged by PaveDB. Use query inspection to see what ran, replay it against current data, and inspect the source chunks behind a document.

from pavesdk.client import connect

db = connect("http://localhost:8086", api_key="super-sekret")
books = db.collection("books", tenant="demo")

books.search("captain nemo", k=3)

latest = books.queries(limit=1)[0]
latest

{
    "query_id": "0d4f5a1b-9e4b-41c7-8b3f-8f6b5de3e74a",
    "tenant": "demo",
    "collection": "books",
    "query_text": "captain nemo",
    "k": 3,
    "filters": None,
    "result_count": 2,
    "latency_ms": 12.4,
    "created_at": "2026-06-20T18:42:16.153201Z",
}

query = books.get_query(latest["query_id"])
query

{
    "query_id": "0d4f5a1b-9e4b-41c7-8b3f-8f6b5de3e74a",
    "tenant": "demo",
    "collection": "books",
    "query_text": "captain nemo",
    "k": 3,
    "filters": None,
    "result_ids": ["note-1:0000", "note-2:0000"],
    "result_count": 2,
    "latency_ms": 12.4,
}

replayed = books.replay(query["query_id"])
replayed

[
    {
        "id": "note-1:0000",
        "score": 0.86,
        "text": "Captain Nemo commands the Nautilus.",
        "meta": {"docid": "note-1", "kind": "note"},
    },
    {
        "id": "note-2:0000",
        "score": 0.73,
        "text": "The Nautilus dives beneath the ice.",
        "meta": {"docid": "note-2", "kind": "note"},
    },
]

docid = replayed[0]["meta"]["docid"]
chunks = books.list_chunks(docid)
chunks

[
    {
        "rid": "note-1:0000",
        "docid": "note-1",
        "chunk": 0,
        "text": "Captain Nemo commands the Nautilus.",
        "metadata": {"kind": "note"},
    }
]

chunk = books.get_chunk(chunks[0]["rid"])
chunk

{
    "rid": "note-1:0000",
    "docid": "note-1",
    "chunk": 0,
    "text": "Captain Nemo commands the Nautilus.",
    "metadata": {"kind": "note"},
}

content = books.get_chunk_content(chunk["rid"])
content

{
    "content": b"Captain Nemo commands the Nautilus.",
    "content_type": "text/plain; charset=utf-8",
}

Local Mode

With pavedb installed, the same API can use a local persisted store:

from pavesdk.client import connect

with connect("./data", tenant="demo") as db:
    books = db.create_collection("books")
    books.add("Captain Nemo commands the Nautilus.", docid="note-1")
    print(books.search("captain", k=3))

If pavedb is not installed, local targets raise LocalClientUnavailable.

Archives

from pathlib import Path
from pavesdk.client import connect

with connect("http://localhost:8086", api_key="super-sekret") as db:
    archive_bytes = db.dump_archive()
    Path("pavedb-data.zip").write_bytes(archive_bytes)

    saved_path = db.dump_archive("pavedb-data.zip")
    db.restore_archive(Path(saved_path).read_bytes())

One collection moves on its own. restore_archive on a new name creates the collection; replace=True rolls an existing one back to the snapshot:

with connect("http://localhost:8086", api_key="tenant-key", tenant="demo") as db:
    books = db.collection("books")
    snapshot = books.dump_archive()
    db.collection("books-copy").restore_archive(snapshot)
    books.restore_archive(snapshot, replace=True)

    job = books.reindex(embed_model="sentence-transformers/all-MiniLM-L6-v2")
    print(books.reindex_job(job["job_id"])["status"])