tortoise-embeddings 1.0.0

Adds pgvector support to TortoiseORM.

Tortoise Embeddings

Tortoise Embeddings adds pgvector support to TortoiseORM, enabling efficient vector storage and similarity search in PostgreSQL.

Features

• Four Vector Field Types:
  • VectorField: Supports vector type.
  • HalfVectorField: Supports halfvec type.
  • BinaryVector: Supports bit type.
  • SparseVector: Supports sparsevec type.
• Aerich Migration Support: Automatically includes vector-based types from your models into the migration scripts.
• Similarity Operations:
  • Custom Functions: L2Distance, CosineDistance, InnerProduct, L1Distance, HammingDistance, JaccardDistance.
  • Custom Filters: Convenient filter suffixes - __l2, __cosine, __inner, __l1, __hamming, and __jaccard.
• Works with numpy arrays and standard Python types.

Installation

pip install tortoise-embeddings

Make sure you have pgvector installed in your PostgreSQL database.

Usage

Define Your Models

from tortoise.models import Model
from tortoise.fields import IntField
from tortoise_embeddings import VectorField, HalfVectorField, BinaryVectorField, SparseVectorField

class Item(Model):
    id = IntField(primary_key=True)
    # Define a vector field with 3 dimensions
    embedding = VectorField(dimensions=3)
    # Support for other pgvector types
    half_embedding = HalfVectorField(dimensions=3)
    binary_embedding = BinaryVectorField(dimensions=4)
    sparse_embedding = SparseVectorField(dimensions=3)

Initialize with Custom Client

To enable binary codecs and migration support, use the provided AsyncpgDBClient:

from tortoise import Tortoise
import tortoise.backends.asyncpg
from tortoise_embeddings import VectorAsyncpgDBClient

# Override the client class for the asyncpg backend
tortoise.backends.asyncpg.client_class = VectorAsyncpgDBClient

async def init():
    await Tortoise.init(
        db_url='postgres://user:pass@host:port/db',
        modules={'models': ['your_models_module']}
    )
    await Tortoise.generate_schemas()

Similarity Search

Using Custom Filters (Recommended)

import numpy

target_vector = numpy.array([1.0, 0.0, 0.0])

# Find items where L2 distance is less than 0.5
items = await Item.filter(embedding__l2=(target_vector, 0.5))

# Find items where Cosine distance is less than 0.1
items = await Item.filter(embedding__cosine=(target_vector, 0.1))

Using Custom Functions

from tortoise_embeddings.similarity import L2Distance

target_vector = [1.0, 0.0, 0.0]

# Annotate results with distance and order by it
items = await Item.all().annotate(
    dist=L2Distance(Item._meta.fields_map['embedding'], target_vector, alias="dist")
).order_by("dist")

for item in items:
    print(f"Item {item.id} has distance {item.dist}")

Aerich Migrations

If you use aerich, simply use the AerichVectorPostgresDDL to ensure the vector extension is created:

# In your project configuration
from tortoise_embeddings import AerichVectorPostgresDDL
# This is usually handled automatically if you use the VectorAsyncpgDBClient

Development setup

Set the following environment variables:

Variable	Description	How to get?
PSQL_CONNECTION_STRING	A connection string to a PostgreSQL database.	postgres://user:password@host:port/database
GEMINI_API_KEY	An API key to Google GenAI. (optional)	Go to aistudio.google.com and create an API key.

Dependencies

• tortoise-orm>=0.25.0,<1.0.0
• aerich>=0.9.0
• asyncpg>=0.31.0
• pgvector>=0.4.0
• numpy>=2.4.0

Running Tests

Tests require a PostgreSQL database with the pgvector extension.

export PSQL_CONNECTION_STRING="postgres://user:password@host:port/database"
export GEMINI_API_KEY="your_api_key" # Optional for Gemini integration tests
pytest tests/

Publishing

Execute the following command to generate the distributable files.

python -m build

License

Copyright 2026 Kotz

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.