Need help with annoy?
Click the “chat” button below for chat support from the developer who created it, or find similar developers for support.

About the developer

9.0K Stars 963 Forks Apache License 2.0 830 Commits 40 Opened issues


Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk

Services available


Need anything else?

Contributors list


.. figure:: :alt: Annoy example :align: center

.. image:: :target:

.. image:: :target:

.. image:: :target:

Annoy (

Approximate Nearest Neighbors 
__ Oh Yeah) is a C++ library with Python bindings to search for points in space that are close to a given query point. It also creates large read-only file-based data structures that are
__ into memory so that many processes may share the same data.


To install, simply do

pip install --user annoy
to pull down the latest version from

For the C++ version, just clone the repo and

#include "annoylib.h"


There are some other libraries to do nearest neighbor search. Annoy is almost as fast as the fastest libraries, (see below), but there is actually another feature that really sets Annoy apart: it has the ability to use static files as indexes. In particular, this means you can share index across processes. Annoy also decouples creating indexes from loading them, so you can pass around indexes as files and map them into memory quickly. Another nice thing of Annoy is that it tries to minimize memory footprint so the indexes are quite small.

Why is this useful? If you want to find nearest neighbors and you have many CPU's, you only need to build the index once. You can also pass around and distribute static files to use in production environment, in Hadoop jobs, etc. Any process will be able to load (mmap) the index into memory and will be able to do lookups immediately.

We use it at

__ for music recommendations. After running matrix factorization algorithms, every user/item can be represented as a vector in f-dimensional space. This library helps us search for similar users/items. We have many millions of tracks in a high-dimensional space, so memory usage is a prime concern.

Annoy was built by

Erik Bernhardsson 
__ in a couple of afternoons during
Hack Week 

Summary of features

  • Euclidean distance 
    Manhattan distance 
    cosine distance 
    Hamming distance 
    , or
    Dot (Inner) Product distance 
  • Cosine distance is equivalent to Euclidean distance of normalized vectors = sqrt(2-2*cos(u, v))
  • Works better if you don't have too many dimensions (like <100) but seems to perform surprisingly well even up to 1,000 dimensions
  • Small memory usage
  • Lets you share memory between multiple processes
  • Index creation is separate from lookup (in particular you can not add more items once the tree has been created)
  • Native Python support, tested with 2.7, 3.6, and 3.7.
  • Build index on disk to enable indexing big datasets that won't fit into memory (contributed by
    Rene Hollander 

Python code example

.. code-block:: python

from annoy import AnnoyIndex import random

f = 40 t = AnnoyIndex(f, 'angular') # Length of item vector that will be indexed for i in range(1000): v = [random.gauss(0, 1) for z in range(f)] t.add_item(i, v) # 10 trees'test.ann')

# ...

u = AnnoyIndex(f, 'angular') u.load('test.ann') # super fast, will just mmap the file print(u.getnnsby_item(0, 1000)) # will find the 1000 nearest neighbors

Right now it only accepts integers as identifiers for items. Note that it will allocate memory for max(id)+1 items because it assumes your items are numbered 0 … n-1. If you need other id's, you will have to keep track of a map yourself.

Full Python API

  • AnnoyIndex(f, metric)
    returns a new index that's read-write and stores vector of
    dimensions. Metric can be
    , or
  • a.add_item(i, v)
    adds item
    (any nonnegative integer) with vector
    . Note that it will allocate memory for
  •, n_jobs=-1)
    builds a forest of
    trees. More trees gives higher precision when querying. After calling
    , no more items can be added.
    specifies the number of threads used to build the trees.
    uses all available CPU cores.
  •, prefault=False)
    saves the index to disk and loads it (see next function). After saving, no more items can be added.
  • a.load(fn, prefault=False)
    loads (mmaps) an index from disk. If
    is set to
    , it will pre-read the entire file into memory (using mmap with
    ). Default is
  • a.unload()
  • a.get_nns_by_item(i, n, search_k=-1, include_distances=False)
    returns the
    closest items. During the query it will inspect up to
    nodes which defaults to
    n_trees * n
    if not provided.
    gives you a run-time tradeoff between better accuracy and speed. If you set
    , it will return a 2 element tuple with two lists in it: the second one containing all corresponding distances.
  • a.get_nns_by_vector(v, n, search_k=-1, include_distances=False)
    same but query by vector
  • a.get_item_vector(i)
    returns the vector for item
    that was previously added.
  • a.get_distance(i, j)
    returns the distance between items
    . NOTE: this used to return the squared distance, but has been changed as of Aug 2016.
  • a.get_n_items()
    returns the number of items in the index.
  • a.get_n_trees()
    returns the number of trees in the index.
  • a.on_disk_build(fn)
    prepares annoy to build the index in the specified file instead of RAM (execute before adding items, no need to save after build)
  • a.set_seed(seed)
    will initialize the random number generator with the given seed. Only used for building up the tree, i. e. only necessary to pass this before adding the items. Will have no effect after calling


  • There's no bounds checking performed on the values so be careful.
  • Annoy uses Euclidean distance of normalized vectors for its angular distance, which for two vectors u,v is equal to

The C++ API is very similar: just

#include "annoylib.h"
to get access to it.


There are just two main parameters needed to tune Annoy: the number of trees

and the number of nodes to inspect during searching
  • n_trees
    is provided during build time and affects the build time and the index size. A larger value will give more accurate results, but larger indexes.
  • search_k
    is provided in runtime and affects the search performance. A larger value will give more accurate results, but will take longer time to return.


is not provided, it will default to
n * n_trees
is the number of approximate nearest neighbors. Otherwise,
are roughly independent, i.e. the value of
will not affect search time if
is held constant and vice versa. Basically it's recommended to set
as large as possible given the amount of memory you can afford, and it's recommended to set
as large as possible given the time constraints you have for the queries.

You can also accept slower search times in favour of reduced loading times, memory usage, and disk IO. On supported platforms the index is prefaulted during

, causing the file to be pre-emptively read from disk into memory. If you set
, pages of the mmapped index are instead read from disk and cached in memory on-demand, as necessary for a search to complete. This can significantly increase early search times but may be better suited for systems with low memory compared to index size, when few queries are executed against a loaded index, and/or when large areas of the index are unlikely to be relevant to search queries.

How does it work


random projections 
__ and by building up a tree. At every intermediate node in the tree, a random hyperplane is chosen, which divides the space into two subspaces. This hyperplane is chosen by sampling two points from the subset and taking the hyperplane equidistant from them.

We do this k times so that we get a forest of trees. k has to be tuned to your need, by looking at what tradeoff you have between precision and performance.

Hamming distance (contributed by

Martin Aumüller 
__) packs the data into 64-bit integers under the hood and uses built-in bit count primitives so it could be quite fast. All splits are axis-aligned.

Dot Product distance (contributed by

Peter Sobot 
) reduces the provided vectors from dot (or "inner-product") space to a more query-friendly cosine space using
a method by Bachrach et al., at Microsoft Research, published in 2014 

More info

  • Dirk Eddelbuettel 
    __ provides an
    R version of Annoy 
  • Andy Sloane 
    __ provides a
    Java version of Annoy 
    __ although currently limited to cosine and read-only.
  • Pishen Tsai 
    __ provides a
    Scala wrapper of Annoy 
    __ which uses JNA to call the C++ library of Annoy.
  • Atsushi Tatsuma 
    __ provides
    Ruby bindings for Annoy 
  • There is
    experimental support for Go 
    __ provided by
    Taneli Leppä 
  • Boris Nagaev 
    __ wrote
    Lua bindings 
  • During part of Spotify Hack Week 2016 (and a bit afterward),
    Jim Kang 
    __ wrote
    Node bindings 
    __ for Annoy.
  • Min-Seok Kim 
    __ built a
    Scala version 
    __ of Annoy.
  • hanabi1224 
    __ built a read-only
    Rust version 
    __ of Annoy, together with dotnet, jvm and dart read-only bindings.
  • Presentation from New York Machine Learning meetup 
    __ about Annoy
  • Annoy is available as a
    conda package 
    __ on Linux, OS X, and Windows.
  • ann-benchmarks 
    __ is a benchmark for several approximate nearest neighbor libraries. Annoy seems to be fairly competitive, especially at higher precisions:

.. figure:: :alt: ANN benchmarks :align: center :target:

Source code

It's all written in C++ with a handful of ugly optimizations for performance and memory usage. You have been warned :)

The code should support Windows, thanks to

Qiang Kou 
__ and
Timothy Riley 

To run the tests, execute

python nosetests
. The test suite includes a big real world dataset that is downloaded from the internet, so it will take a few minutes to execute.


Feel free to post any questions or comments to the

__ group. I'm
__ on Twitter.

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.