Pre-release

This release is a pre-release and may not be stable for production use.

VaneDB for Python

VaneDB is an embeddable vector database backed by Rust. Store vectors and search for their nearest neighbors inside your Python process, without a database server. Supply your own embeddings; VaneDB does not generate them. It stores only (id, vector) pairs — no metadata or payload storage and no filtered search — so keep your own id-to-document mapping alongside it.

The vanedb package is the shipping Python implementation. C++ bindings are kept in the repository for reference and local testing, and are not published.

Installation

Requires Python 3.11 or newer.

python -m pip install vanedb==0.1.0rc2

The version is pinned because pip does not select a prerelease unless asked; drop the pin once 0.1.0 is out. Building from a checkout is described in the repository.

Python lists work without additional dependencies. NumPy is optional; its arrays can also be used for vector and batch inputs.

Quick start

Use FlatIndex for exact search, or ApproxIndex for approximate search. Both return (id, distance) pairs, with the nearest results first.

from pathlib import Path
from tempfile import TemporaryDirectory

from vanedb import Metric, ApproxIndex, FlatIndex

ids = [101, 202]
vectors = [[1.0, 0.0, 0.0], [0.0, 1.0, 0.0]]
query = [1.0, 0.0, 0.0]

# Exact cosine-distance search.
store = FlatIndex(3, Metric.COSINE)
store.add_batch(ids, vectors)
assert store.search(query, 1) == [(101, 0.0)]

# Approximate search; capacity is a reserve hint and the index can grow.
index = ApproxIndex(3, Metric.COSINE, capacity=100, seed=42)
index.add_batch(ids, vectors)
index.ef_search = 100
hits = index.search(query, 1)
assert hits == [(101, 0.0)]

# Save and reload the index without rebuilding it.
with TemporaryDirectory() as directory:
    path = Path(directory) / "index.bin"
    index.save(path)
    restored = ApproxIndex.load(path)
    assert restored.search(query, 1) == hits

Every path argument — ApproxIndex.save/load, DiskIndexBuilder.save and DiskIndex.open — takes a str or any os.PathLike, so a pathlib.Path needs no str() around it.

Supported metrics are Metric.L2 (squared Euclidean distance), Metric.COSINE, and Metric.DOT (negative dot product). For each metric, smaller distances rank first. Vectors must match the store or index dimension and contain finite values; IDs must be unique unsigned 64-bit integers.

Exceptions are chosen so a caller can branch on the type rather than parse the message. A missing id raises KeyError; validation errors, including negative or out-of-range integer sizes and seeds, raise ValueError; a corrupt file raises ValueError and an absent one FileNotFoundError, so "load it, or build it if absent" needs no message matching. Other I/O failures raise OSError. Arguments of the wrong type can raise TypeError.

ApproxIndex.search accepts a keyword-only ef_search that applies to that query alone, leaving the shared ef_search property untouched — searches release the GIL, so raising recall by assigning to the property is visible to concurrent threads. m, ef_construction and seed are readable on any ApproxIndex, including one from ApproxIndex.load, whose graph the caller did not build. get and get_vector are the same operation on every index type, so swapping FlatIndex for ApproxIndex does not mean renaming call sites.

ApproxIndex writes shared VNDB v2 graph files. Both engines preserve their vectors, links, IDs and deleted slots; further insertions may differ across engines. Legacy Rust graphs remain readable: load and save to a new path to migrate. Older readers cannot open VNDB v2. Keep originals and source vectors while verifying migration. During 0.x, APIs and persistence formats may change in a minor release. Existing format identifiers will not be reinterpreted; new encodings require new identifiers and readers for existing files are retained. Older readers need not accept future formats. Future insertions need not reproduce identical topology. DiskIndex continues to use shared VNDB v1 files. Disk construction buffers vectors in memory before saving, while the opened index uses a read-only memory mapping.

DiskIndexBuilder is the only way to create a file DiskIndex.open accepts:

from vanedb import DiskIndex, DiskIndexBuilder, Metric

builder = DiskIndexBuilder(3, Metric.L2)
builder.add(1, [1.0, 0.0, 0.0])
builder.add(2, [0.0, 1.0, 0.0])
builder.save("corpus.vndb")

index = DiskIndex.open("corpus.vndb")
print(len(index))                        # 2
print(index.search([0.9, 0.1, 0.0], 1))  # [(1, ...)]

DiskIndex has no constructor of its own — DiskIndex(...) raises TypeError.

For DiskIndex.open(path), keep the underlying file immutable from before the call until the last reference to the index is released. Prevent every process from writing or truncating that file, including through other paths or open handles. A read-only mapping does not prevent those changes; they can corrupt results or crash Python. To update the data, write a separate file and atomically replace the path where supported. Existing indexes keep the original mapping; open a new index to read the replacement.

The C++ Python package has different constructor keywords and returns separate id and distance arrays. Changing engines requires adapting those calls and checking feature availability, in addition to changing the import.

Project

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

vanedb-0.1.0rc2.tar.gz (165.1 kB view details)

Uploaded Sep 10, 2026 Source

Built Distributions

If you're not sure about the file name format, learn more about wheel file names.

The dropdown lists show the available interpreters, ABIs, and platforms. Enable javascript to be able to filter the list of wheel files.

vanedb-0.1.0rc2-cp314-cp314-win_amd64.whl (267.4 kB view details)

Uploaded Sep 10, 2026 CPython 3.14Windows x86-64

vanedb-0.1.0rc2-cp314-cp314-musllinux_1_2_x86_64.whl (619.8 kB view details)

Uploaded Sep 10, 2026 CPython 3.14musllinux: musl 1.2+ x86-64

vanedb-0.1.0rc2-cp314-cp314-musllinux_1_2_aarch64.whl (584.3 kB view details)

Uploaded Sep 10, 2026 CPython 3.14musllinux: musl 1.2+ ARM64

vanedb-0.1.0rc2-cp314-cp314-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (413.7 kB view details)

Uploaded Sep 10, 2026 CPython 3.14manylinux: glibc 2.17+ x86-64

vanedb-0.1.0rc2-cp314-cp314-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (405.0 kB view details)

Uploaded Sep 10, 2026 CPython 3.14manylinux: glibc 2.17+ ARM64

vanedb-0.1.0rc2-cp314-cp314-macosx_11_0_arm64.whl (366.6 kB view details)

Uploaded Sep 10, 2026 CPython 3.14macOS 11.0+ ARM64

vanedb-0.1.0rc2-cp314-cp314-macosx_10_12_x86_64.whl (375.6 kB view details)

Uploaded Sep 10, 2026 CPython 3.14macOS 10.12+ x86-64

vanedb-0.1.0rc2-cp313-cp313-win_amd64.whl (266.2 kB view details)

Uploaded Sep 10, 2026 CPython 3.13Windows x86-64

vanedb-0.1.0rc2-cp313-cp313-musllinux_1_2_x86_64.whl (619.5 kB view details)

Uploaded Sep 10, 2026 CPython 3.13musllinux: musl 1.2+ x86-64

vanedb-0.1.0rc2-cp313-cp313-musllinux_1_2_aarch64.whl (583.6 kB view details)

Uploaded Sep 10, 2026 CPython 3.13musllinux: musl 1.2+ ARM64

vanedb-0.1.0rc2-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (413.1 kB view details)

Uploaded Sep 10, 2026 CPython 3.13manylinux: glibc 2.17+ x86-64

vanedb-0.1.0rc2-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (404.2 kB view details)

Uploaded Sep 10, 2026 CPython 3.13manylinux: glibc 2.17+ ARM64

vanedb-0.1.0rc2-cp313-cp313-macosx_11_0_arm64.whl (365.6 kB view details)

Uploaded Sep 10, 2026 CPython 3.13macOS 11.0+ ARM64

vanedb-0.1.0rc2-cp313-cp313-macosx_10_12_x86_64.whl (374.3 kB view details)

Uploaded Sep 10, 2026 CPython 3.13macOS 10.12+ x86-64

vanedb-0.1.0rc2-cp312-cp312-win_amd64.whl (267.3 kB view details)

Uploaded Sep 10, 2026 CPython 3.12Windows x86-64

vanedb-0.1.0rc2-cp312-cp312-musllinux_1_2_x86_64.whl (619.0 kB view details)

Uploaded Sep 10, 2026 CPython 3.12musllinux: musl 1.2+ x86-64

vanedb-0.1.0rc2-cp312-cp312-musllinux_1_2_aarch64.whl (583.1 kB view details)

Uploaded Sep 10, 2026 CPython 3.12musllinux: musl 1.2+ ARM64

vanedb-0.1.0rc2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (413.0 kB view details)

Uploaded Sep 10, 2026 CPython 3.12manylinux: glibc 2.17+ x86-64

vanedb-0.1.0rc2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (404.2 kB view details)

Uploaded Sep 10, 2026 CPython 3.12manylinux: glibc 2.17+ ARM64

vanedb-0.1.0rc2-cp312-cp312-macosx_11_0_arm64.whl (365.9 kB view details)

Uploaded Sep 10, 2026 CPython 3.12macOS 11.0+ ARM64

vanedb-0.1.0rc2-cp312-cp312-macosx_10_12_x86_64.whl (374.5 kB view details)

Uploaded Sep 10, 2026 CPython 3.12macOS 10.12+ x86-64

vanedb-0.1.0rc2-cp311-cp311-win_amd64.whl (269.6 kB view details)

Uploaded Sep 10, 2026 CPython 3.11Windows x86-64

vanedb-0.1.0rc2-cp311-cp311-musllinux_1_2_x86_64.whl (622.7 kB view details)

Uploaded Sep 10, 2026 CPython 3.11musllinux: musl 1.2+ x86-64

vanedb-0.1.0rc2-cp311-cp311-musllinux_1_2_aarch64.whl (588.1 kB view details)

Uploaded Sep 10, 2026 CPython 3.11musllinux: musl 1.2+ ARM64

vanedb-0.1.0rc2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (416.9 kB view details)

Uploaded Sep 10, 2026 CPython 3.11manylinux: glibc 2.17+ x86-64

vanedb-0.1.0rc2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl (408.8 kB view details)

Uploaded Sep 10, 2026 CPython 3.11manylinux: glibc 2.17+ ARM64

vanedb-0.1.0rc2-cp311-cp311-macosx_11_0_arm64.whl (370.8 kB view details)

Uploaded Sep 10, 2026 CPython 3.11macOS 11.0+ ARM64

vanedb-0.1.0rc2-cp311-cp311-macosx_10_12_x86_64.whl (373.8 kB view details)

Uploaded Sep 10, 2026 CPython 3.11macOS 10.12+ x86-64

File details

Details for the file vanedb-0.1.0rc2.tar.gz.

File metadata

Download URL: vanedb-0.1.0rc2.tar.gz
Upload date: Sep 10, 2026
Size: 165.1 kB
Tags: Source
Uploaded using Trusted Publishing? Yes
Uploaded via: twine/7.0.0 CPython/3.13.14

File hashes

Hashes for vanedb-0.1.0rc2.tar.gz
Algorithm	Hash digest
SHA256	`6b9d8565ecade0d543fef02a8400657082b9bcd5ef8605cdbbdface249581c54`
MD5	`1e30558b1fd5d7ac122959c25775dd0c`
BLAKE2b-256	`a5a5d91025f2bddab5939ce4f5476a198d9fff54c7c819d5b810e5d452f5c763`

Algorithm	Hash digest
SHA256	`140baa8feb11c06aa2a0dd3074b75b0528b6147a346288c77d9433ae7def85df`
MD5	`5a7ffbda9a22cc1971bda458633350e0`
BLAKE2b-256	`a8ff3a569a1b0cc0acf2ddf9eba038a71d79e170eb3bc1a9c2738af0faad2ff8`

Algorithm	Hash digest
SHA256	`08743abb8166c9e9202f510b60b880a2057b303f80878af84e32de5bb4d7975a`
MD5	`62abd327cb01987c4ac3e6d387087285`
BLAKE2b-256	`9a0479cdf356a3b4f77f8b7395c6bf57620c164368c743b12ac41d328c377626`

Algorithm	Hash digest
SHA256	`33a8871264f78cb5043cff6c1d21eb04c55d7f9e8d801c59fe4c520a388343b0`
MD5	`4643d5f103efb4d190615d4e39bf46d6`
BLAKE2b-256	`335fae8227ac67a445f909de8827116cf6572ae385f827b371d15d9c730380d3`

Algorithm	Hash digest
SHA256	`8d8aa4219bc4e6405256a80aba0bd6dada6f706d9108a2e53c91502c05af04c5`
MD5	`feab6fecda6fc46687afd0605bda5bae`
BLAKE2b-256	`926603d321d68bd200788f4957be9014c48d62e2d717cb9df2a679c9eae7c7f3`

Algorithm	Hash digest
SHA256	`bb831a5c07d3da4bf06e1855e282b84250eae513d191e39aa0b851027748c8f6`
MD5	`ece67c358b199479a7fa4f662f0948fa`
BLAKE2b-256	`60c836d0fd03e140f2c3dce15ee562bf3b5c79642b1b016717b598b72dca42be`

Algorithm	Hash digest
SHA256	`ea4399fcfe3170f21f4159ba5f926ccc67132d86ee607747241654b817bb7840`
MD5	`a8b6f61db90bb63033b3f56112ac1f98`
BLAKE2b-256	`c1e6264aed83fdf351ecae58c34eda600e6fca87baa07f59c57713b533545873`

Algorithm	Hash digest
SHA256	`6b5f587117e343f5131b1cbeb0de62154df218b1bb3f0e5fa9f2730f116ee0d7`
MD5	`f02891135969fc135496ac8bd3dab420`
BLAKE2b-256	`a32250b50fe98839299556a190a4db423485ba66fe20f56251b81ccb266ae3c4`

Algorithm	Hash digest
SHA256	`f1142a40d19493bd4c267a26c47425d14414f5184e1e2d8d2c29e6cc5ff03a90`
MD5	`f0e1a15972cf9c2fe37b3f0a281154a1`
BLAKE2b-256	`ab7acddc505d641313cc3de4df6231f100258eb2177c758aafdd76d0dd38edb4`

Algorithm	Hash digest
SHA256	`f5c48d672f3aab6c0320257797ae21ff2fae1b708aa6058659b5217ac9cfe413`
MD5	`6137d4f148a365b5cac557d2130f2036`
BLAKE2b-256	`60eb3e8aeac8591be9023c68497a85ed27074933559afb7f6fa1ec0a58c7b020`

Algorithm	Hash digest
SHA256	`488cae2c5f35216d763d8e577f20710a286bad0c9a22eeb291c5c753d58c7ff4`
MD5	`c5cc498ab46b4de1dc2d3083de4182d6`
BLAKE2b-256	`22d0d41989bd5607700ad669a78abaf12e59903be29e969ae10dbcb5a1a38a42`

Algorithm	Hash digest
SHA256	`997ff4304ac0386ad8b1d280e7dfcb789377da842692a63138024bf944dd9feb`
MD5	`2b1beb646f16162a35bf80682ce9ca30`
BLAKE2b-256	`aae38824401980f8f3fcfde0f67e1980b18c3bb590dbadca62caaf050212fec8`

Algorithm	Hash digest
SHA256	`31debed36c9e766787cc167e2321bc909735a08a73f1424e58ecbed23e23b6fe`
MD5	`604256010bbe8c5b60b2683e4c1a15a8`
BLAKE2b-256	`8340d9c43b56e5b4896184a0501d6b908e422b048b48c26e41b24a089a828c00`

Algorithm	Hash digest
SHA256	`192fc2550e47865ed5937dfbd2be9114584775a233986777475d2191b1c96025`
MD5	`99f531024dbe64e750ff9487ec3f34e7`
BLAKE2b-256	`1ce013a4f14268b662631f29b333a3ad6f0d8fc44db6b35ac66d9291cea7f606`

Algorithm	Hash digest
SHA256	`840325a7e71308d555a97b2f2feaa9093ab1a962a10e3ba8aa5849c17946f620`
MD5	`00e1e5a190f8d714fa819807b98964c7`
BLAKE2b-256	`765bda95814a28816c6f254ea28062290feeca5280df1620a13db278c4963b1e`

Algorithm	Hash digest
SHA256	`3b1a6625e74211138a10f1cdd58934f39b6c9af1447241124508a5ff943585e2`
MD5	`edf1125ca147d1ff7ef86ddf70a481e3`
BLAKE2b-256	`59fda92ae94a5b0234e075b9275da9f27893cc35b2b2dc5ad576ff8af75317b5`

Algorithm	Hash digest
SHA256	`df7057488c9c1509a86eab0843fcc8d39a5661aa8e74314e51d847b7c2d3cc23`
MD5	`593e3c0d71ce4bdcae6d84a7fb0e34ee`
BLAKE2b-256	`a23a3e6ecb76ac2c471d4ca85c1c0b10ebb539994f94a95899fadb9f93b1c2bb`

Algorithm	Hash digest
SHA256	`05e4aa47a4500db94eb15703a915a7d4f51132f25a0ad979efa387df4515ca99`
MD5	`3311103b02daeb561e535c8e7f098a5d`
BLAKE2b-256	`0b914b449c50eff631dc5b5303c88d516641fef9b0b5b113c6b1256597b212bc`

Algorithm	Hash digest
SHA256	`a3175245b902167f3cfdc74aa6f279fa4597bb73f489cd7310362e977eac63b4`
MD5	`9cc5b9b4a8e1e69b9e39142e9931f29d`
BLAKE2b-256	`e2b8cc1034a979833bdd4c632c2dda5e6e8da36930e0fa04511baebd1b6dcf37`

Algorithm	Hash digest
SHA256	`3d704227ea3b5bee328445c9de27241943bbb88c17b38893ba1eac110e9e6202`
MD5	`9c8a392e9609dd2aff0e191cd97f6fc6`
BLAKE2b-256	`c4ee4ee9c30e20c5d3212527dac4f9e5d0c4fdeff9a4a344f210a7f24e8b197f`

Algorithm	Hash digest
SHA256	`422d3bc8adeba8a9ef816ddbebb2f907b71e7a13e497fc1da9c3b420bbc260d1`
MD5	`581d382de02ddf72a25d9fb35238b380`
BLAKE2b-256	`a4c6ffc55519a24b46d1869ba8433f94ba1a3ea694601cefdc21f225c6d01c85`

vanedb 0.1.0rc2

VaneDB for Python

Installation

Quick start

Project

Key dates

1 maintainer

License expression

Requires

Tags

Classifiers

Download files

Source Distribution

Built Distributions

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

File hashes

Provenance

File details

File metadata

Algorithm	Hash digest
SHA256	`f277e38081dea8622d128e9f7c482ed2d41fe91e96d44d8802c9c9d2504ddb06`
MD5	`142215627679ddee21a9b035aeb7457b`
BLAKE2b-256	`18ba5fd32ccf478b81f08b845209fd60fa5677199c3d5ecb4a8a16a130e9c07f`

Algorithm	Hash digest
SHA256	`ed2aa4a9a5563e30f03f20e0602ad8ae147e4e0524c166fa3752217797e77844`
MD5	`cb3819703dd646844f91f9e6fa55f0d2`
BLAKE2b-256	`d00b7bc386f1027c694c9eab49a69a9752c9fcf1e48013a8d7c214cf9eba419b`

Algorithm	Hash digest
SHA256	`b38745471b5f13b43a54553506c0069891b6b7a6669e225d21dc9cfd395de8bb`
MD5	`3c99556fce005fa6b528825cafcfec6b`
BLAKE2b-256	`3fd6c162cd54d0ca6112202a4f884b8f310dc1a462611b1d09b448a3975ad70c`

Algorithm	Hash digest
SHA256	`67b9734e3737cca8d0fee41cf1f46d565b209462cc0ca4f1278632531e2d91b4`
MD5	`977a65837cbcf1259a8c3075e6edf658`
BLAKE2b-256	`44ca638b9a3809a611584453a0a6b166cc90e1d8435fc17937926cc6c0d68971`

Algorithm	Hash digest
SHA256	`6a4156380edcc5854d37d0f4c8f87578fed424cdca7ed72a5205ec5f8c60a3a6`
MD5	`650c72cd98c91c910c2ceef1b4ad82f1`
BLAKE2b-256	`a3cb6a5d2cfdbc3ceedc8eaeffb34d027766676dc1e5b0041a86f91da6d5271a`

Algorithm	Hash digest
SHA256	`73c41b85e87506e1e1d68210bfefe93719d62ca3402d2e8355069ae1cc86521a`
MD5	`61f8b52352f6f0cc15688717d8b1b905`
BLAKE2b-256	`1f4848051e3ec5dd8561f03d3993c8b1e5b52d2a3519ff9e1398ff5206cfae3d`

Algorithm	Hash digest
SHA256	`d2aab3b006385e6814ffe504b1df4c0519130afb55ad98af844f87a4ec333e02`
MD5	`b8b257b5d518465ba7a7db88835059f1`
BLAKE2b-256	`74688eaa21f8b79c6204e810fa810cd0b3b50e1b363ff638f874e32a42607415`

Algorithm	Hash digest
SHA256	`c1a4550ee0be0ef756f2aad8916378b92d1d359278533aa66c8d7edc60bb2165`
MD5	`5ded0aad85cbe33a9408836ac1d82dcb`
BLAKE2b-256	`752f781146d3b30c402bd677dd1b650a9987a281861df084c0f9653c7c1da5bd`