Open-source vector similarity search for Postgres
Store your vectors with the rest of your data. Supports:
- exact and approximate nearest neighbor search
- L2 distance, inner product, and cosine distance
- any language with a Postgres client
Plus ACID compliance, point-in-time recovery, JOINs, and all of the other great features of Postgres
Compile and install the extension (supports Postgres 11+)
cd /tmp
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
cd pgvector
make
make install # may need sudo
See the installation notes if you run into issues
You can also install it with Docker, Homebrew, PGXN, APT, Yum, or conda-forge, and it comes preinstalled with Postgres.app and many hosted providers. There are also instructions for GitHub Actions.
Enable the extension (do this once in each database where you want to use it)
CREATE EXTENSION vector;
Create a vector column with 3 dimensions
CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
Insert vectors
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
Get the nearest neighbors by L2 distance
SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
Also supports inner product (<#>
) and cosine distance (<=>
)
Note: <#>
returns the negative inner product since Postgres only supports ASC
order index scans on operators
Create a new table with a vector column
CREATE TABLE items (id bigserial PRIMARY KEY, embedding vector(3));
Or add a vector column to an existing table
ALTER TABLE items ADD COLUMN embedding vector(3);
Insert vectors
INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
Upsert vectors
INSERT INTO items (id, embedding) VALUES (1, '[1,2,3]'), (2, '[4,5,6]')
ON CONFLICT (id) DO UPDATE SET embedding = EXCLUDED.embedding;
Update vectors
UPDATE items SET embedding = '[1,2,3]' WHERE id = 1;
Delete vectors
DELETE FROM items WHERE id = 1;
Get the nearest neighbors to a vector
SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
Get the nearest neighbors to a row
SELECT * FROM items WHERE id != 1 ORDER BY embedding <-> (SELECT embedding FROM items WHERE id = 1) LIMIT 5;
Get rows within a certain distance
SELECT * FROM items WHERE embedding <-> '[3,1,2]' < 5;
Note: Combine with ORDER BY
and LIMIT
to use an index
Get the distance
SELECT embedding <-> '[3,1,2]' AS distance FROM items;
For inner product, multiply by -1 (since <#>
returns the negative inner product)
SELECT (embedding <#> '[3,1,2]') * -1 AS inner_product FROM items;
For cosine similarity, use 1 - cosine distance
SELECT 1 - (embedding <=> '[3,1,2]') AS cosine_similarity FROM items;
Average vectors
SELECT AVG(embedding) FROM items;
Average groups of vectors
SELECT category_id, AVG(embedding) FROM items GROUP BY category_id;
By default, pgvector performs exact nearest neighbor search, which provides perfect recall.
You can add an index to use approximate nearest neighbor search, which trades some recall for speed. Unlike typical indexes, you will see different results for queries after adding an approximate index.
Supported index types are:
An HNSW index creates a multilayer graph. It has better query performance than IVFFlat (in terms of speed-recall tradeoff), but has slower build times and uses more memory. Also, an index can be created without any data in the table since there isn’t a training step like IVFFlat.
Add an index for each distance function you want to use.
L2 distance
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops);
Inner product
CREATE INDEX ON items USING hnsw (embedding vector_ip_ops);
Cosine distance
CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
Vectors with up to 2,000 dimensions can be indexed.
Specify HNSW parameters
m
- the max number of connections per layer (16 by default)ef_construction
- the size of the dynamic candidate list for constructing the graph (64 by default)
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WITH (m = 16, ef_construction = 64);
A higher value of ef_construction
provides better recall at the cost of index build time / insert speed.
Specify the size of the dynamic candidate list for search (40 by default)
SET hnsw.ef_search = 100;
A higher value provides better recall at the cost of speed.
Use SET LOCAL
inside a transaction to set it for a single query
BEGIN;
SET LOCAL hnsw.ef_search = 100;
SELECT ...
COMMIT;
Indexes build significantly faster when the graph fits into maintenance_work_mem
SET maintenance_work_mem = '8GB';
A notice is shown when the graph no longer fits
NOTICE: hnsw graph no longer fits into maintenance_work_mem after 100000 tuples
DETAIL: Building will take significantly more time.
HINT: Increase maintenance_work_mem to speed up builds.
Note: Do not set maintenance_work_mem
so high that it exhausts the memory on the server
Check indexing progress with Postgres 12+
SELECT phase, round(100.0 * blocks_done / nullif(blocks_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
The phases for HNSW are:
initializing
loading tuples
An IVFFlat index divides vectors into lists, and then searches a subset of those lists that are closest to the query vector. It has faster build times and uses less memory than HNSW, but has lower query performance (in terms of speed-recall tradeoff).
Three keys to achieving good recall are:
- Create the index after the table has some data
- Choose an appropriate number of lists - a good place to start is
rows / 1000
for up to 1M rows andsqrt(rows)
for over 1M rows - When querying, specify an appropriate number of probes (higher is better for recall, lower is better for speed) - a good place to start is
sqrt(lists)
Add an index for each distance function you want to use.
L2 distance
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);
Inner product
CREATE INDEX ON items USING ivfflat (embedding vector_ip_ops) WITH (lists = 100);
Cosine distance
CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
Vectors with up to 2,000 dimensions can be indexed.
Specify the number of probes (1 by default)
SET ivfflat.probes = 10;
A higher value provides better recall at the cost of speed, and it can be set to the number of lists for exact nearest neighbor search (at which point the planner won’t use the index)
Use SET LOCAL
inside a transaction to set it for a single query
BEGIN;
SET LOCAL ivfflat.probes = 10;
SELECT ...
COMMIT;
Speed up index creation on large tables by increasing the number of parallel workers (2 by default)
SET max_parallel_maintenance_workers = 7; -- plus leader
For a large number of workers, you may also need to increase max_parallel_workers
(8 by default)
Check indexing progress with Postgres 12+
SELECT phase, round(100.0 * tuples_done / nullif(tuples_total, 0), 1) AS "%" FROM pg_stat_progress_create_index;
The phases for IVFFlat are:
initializing
performing k-means
assigning tuples
loading tuples
Note: %
is only populated during the loading tuples
phase
There are a few ways to index nearest neighbor queries with a WHERE
clause
SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
Create an index on one or more of the WHERE
columns for exact search
CREATE INDEX ON items (category_id);
Or a partial index on the vector column for approximate search
CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WHERE (category_id = 123);
Use partitioning for approximate search on many different values of the WHERE
columns
CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
Use together with Postgres full-text search for hybrid search.
SELECT id, content FROM items, plainto_tsquery('hello search') query
WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
You can use Reciprocal Rank Fusion or a cross-encoder to combine results.
Use EXPLAIN ANALYZE
to debug performance.
EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
To speed up queries without an index, increase max_parallel_workers_per_gather
.
SET max_parallel_workers_per_gather = 4;
If vectors are normalized to length 1 (like OpenAI embeddings), use inner product for best performance.
SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
Language | Libraries / Examples |
---|---|
C | pgvector-c |
C++ | pgvector-cpp |
C#, F#, Visual Basic | pgvector-dotnet |
Crystal | pgvector-crystal |
Dart | pgvector-dart |
Elixir | pgvector-elixir |
Go | pgvector-go |
Haskell | pgvector-haskell |
Java, Kotlin, Groovy, Scala | pgvector-java |
JavaScript, TypeScript | pgvector-node |
Julia | pgvector-julia |
Lisp | pgvector-lisp |
Lua | pgvector-lua |
Nim | pgvector-nim |
OCaml | pgvector-ocaml |
Perl | pgvector-perl |
PHP | pgvector-php |
Python | pgvector-python |
R | pgvector-r |
Ruby | pgvector-ruby, Neighbor |
Rust | pgvector-rust |
Swift | pgvector-swift |
Zig | pgvector-zig |
A non-partitioned table has a limit of 32 TB by default in Postgres. A partitioned table can have thousands of partitions of that size.
Yes, pgvector uses the write-ahead log (WAL), which allows for replication and point-in-time recovery.
You’ll need to use dimensionality reduction at the moment.
You can use vector
as the type (instead of vector(3)
).
CREATE TABLE embeddings (model_id bigint, item_id bigint, embedding vector, PRIMARY KEY (model_id, item_id));
However, you can only create indexes on rows with the same number of dimensions (using expression and partial indexing):
CREATE INDEX ON embeddings USING hnsw ((embedding::vector(3)) vector_l2_ops) WHERE (model_id = 123);
and query with:
SELECT * FROM embeddings WHERE model_id = 123 ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
You can use the double precision[]
or numeric[]
type to store vectors with more precision.
CREATE TABLE items (id bigserial PRIMARY KEY, embedding double precision[]);
-- use {} instead of [] for Postgres arrays
INSERT INTO items (embedding) VALUES ('{1,2,3}'), ('{4,5,6}');
Optionally, add a check constraint to ensure data can be converted to the vector
type and has the expected dimensions.
ALTER TABLE items ADD CHECK (vector_dims(embedding::vector) = 3);
Use expression indexing to index (at a lower precision):
CREATE INDEX ON items USING hnsw ((embedding::vector(3)) vector_l2_ops);
and query with:
SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
SELECT pg_size_pretty(pg_relation_size('index_name'));
The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
BEGIN;
SET LOCAL enable_seqscan = off;
SELECT ...
COMMIT;
Also, if the table is small, a table scan may be faster.
The planner doesn’t consider out-of-line storage in cost estimates, which can make a serial scan look cheaper. You can reduce the cost of a parallel scan for a query with:
BEGIN;
SET LOCAL min_parallel_table_scan_size = 1;
SET LOCAL parallel_setup_cost = 1;
SELECT ...
COMMIT;
or choose to store vectors inline:
ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
DROP INDEX index_name;
Each vector takes 4 * dimensions + 8
bytes of storage. Each element is a single precision floating-point number (like the real
type in Postgres), and all elements must be finite (no NaN
, Infinity
or -Infinity
). Vectors can have up to 16,000 dimensions.
Operator | Description | Added |
---|---|---|
+ | element-wise addition | |
- | element-wise subtraction | |
* | element-wise multiplication | 0.5.0 |
<-> | Euclidean distance | |
<#> | negative inner product | |
<=> | cosine distance |
Function | Description | Added |
---|---|---|
cosine_distance(vector, vector) → double precision | cosine distance | |
inner_product(vector, vector) → double precision | inner product | |
l2_distance(vector, vector) → double precision | Euclidean distance | |
l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0 |
vector_dims(vector) → integer | number of dimensions | |
vector_norm(vector) → double precision | Euclidean norm |
Function | Description | Added |
---|---|---|
avg(vector) → vector | average | |
sum(vector) → vector | sum | 0.5.0 |
If your machine has multiple Postgres installations, specify the path to pg_config with:
export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
Then re-run the installation instructions (run make clean
before make
if needed). If sudo
is needed for make install
, use:
sudo --preserve-env=PG_CONFIG make install
If compilation fails with fatal error: postgres.h: No such file or directory
, make sure Postgres development files are installed on the server.
For Ubuntu and Debian, use:
sudo apt install postgresql-server-dev-16
Note: Replace 16
with your Postgres server version
Support for Windows is currently experimental. Ensure C++ support in Visual Studio is installed, and run:
call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
Note: The exact path will vary depending on your Visual Studio version and edition
Then use nmake
to build:
set "PGROOT=C:\Program Files\PostgreSQL\16"
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
cd pgvector
nmake /F Makefile.win
nmake /F Makefile.win install
Get the Docker image with:
docker pull ankane/pgvector
This adds pgvector to the Postgres image (run it the same way).
You can also build the image manually:
git clone --branch v0.5.1 https://github.com/pgvector/pgvector.git
cd pgvector
docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
With Homebrew Postgres, you can use:
brew install pgvector
Note: This only adds it to the postgresql@14
formula
Install from the PostgreSQL Extension Network with:
pgxn install vector
Debian and Ubuntu packages are available from the PostgreSQL APT Repository. Follow the setup instructions and run:
sudo apt install postgresql-16-pgvector
Note: Replace 16
with your Postgres server version
RPM packages are available from the PostgreSQL Yum Repository. Follow the setup instructions for your distribution and run:
sudo yum install pgvector_16
# or
sudo dnf install pgvector_16
Note: Replace 16
with your Postgres server version
With Conda Postgres, install from conda-forge with:
conda install -c conda-forge pgvector
This method is community-maintained by @mmcauliffe
Download the latest release with Postgres 15+.
pgvector is available on these providers.
Install the latest version (use the same method as the original installation). Then in each database you want to upgrade, run:
ALTER EXTENSION vector UPDATE;
You can check the version in the current database with:
SELECT extversion FROM pg_extension WHERE extname = 'vector';
If upgrading with Postgres < 13, remove this line from sql/vector--0.3.2--0.4.0.sql
:
ALTER TYPE vector SET (STORAGE = extended);
Then run make install
and ALTER EXTENSION vector UPDATE;
.
If upgrading from 0.2.7 or 0.3.0, recreate all ivfflat
indexes after upgrading to ensure all data is indexed.
-- Postgres 12+
REINDEX INDEX CONCURRENTLY index_name;
-- Postgres < 12
CREATE INDEX CONCURRENTLY temp_name ON table USING ivfflat (column opclass);
DROP INDEX CONCURRENTLY index_name;
ALTER INDEX temp_name RENAME TO index_name;
Thanks to:
- PASE: PostgreSQL Ultra-High-Dimensional Approximate Nearest Neighbor Search Extension
- Faiss: A Library for Efficient Similarity Search and Clustering of Dense Vectors
- Using the Triangle Inequality to Accelerate k-means
- k-means++: The Advantage of Careful Seeding
- Concept Decompositions for Large Sparse Text Data using Clustering
- Efficient and Robust Approximate Nearest Neighbor Search using Hierarchical Navigable Small World Graphs
View the changelog
Everyone is encouraged to help improve this project. Here are a few ways you can help:
- Report bugs
- Fix bugs and submit pull requests
- Write, clarify, or fix documentation
- Suggest or add new features
To get started with development:
git clone https://github.com/pgvector/pgvector.git
cd pgvector
make
make install
To run all tests:
make installcheck # regression tests
make prove_installcheck # TAP tests
To run single tests:
make installcheck REGRESS=functions # regression test
make prove_installcheck PROVE_TESTS=test/t/001_wal.pl # TAP test
To enable benchmarking:
make clean && PG_CFLAGS="-DIVFFLAT_BENCH" make && make install
To show memory usage:
make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
To enable assertions:
make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
To get k-means metrics:
make clean && PG_CFLAGS="-DIVFFLAT_KMEANS_DEBUG" make && make install
Resources for contributors