Roboflow Python Package
Roboflow provides everything you need to build and deploy computer vision models. roboflow-python is the official Roboflow Python package. roboflow-python enables you to interact with models, datasets, and projects hosted on Roboflow.
With this Python package, you can:
- Create and manage projects;
- Upload images, annotations, and datasets to manage in Roboflow;
- Start training vision models on Roboflow;
- Run inference on models hosted on Roboflow, or Roboflow models self-hosted via Roboflow Inference, and more.
The Python package is documented on the official Roboflow documentation site. If you are developing a feature for this Python package, or need a full Python library reference, refer to the package developer documentation.
💻 Installation
You will need to have Python 3.10 or higher set up to use the Roboflow Python package.
Run the following command to install the Roboflow Python package:
pip install roboflow
For desktop features, use:
pip install "roboflow[desktop]"
To decode HEIC/HEIF images locally (for example model.predict("photo.heic")), use:
pip install "roboflow[heic]"
This installs pillow-heif, whose binary wheels are GPL-2.0 licensed because they bundle the x265 encoder. Uploading HEIC images works without it, since the server decodes them.
Lightweight install (roboflow-slim)
If you only need vision events, workspace management, and the CLI (no image processing, inference, or training), install the lightweight package:
pip install roboflow-slim
This skips heavy dependencies like OpenCV, NumPy, Matplotlib, and Pillow, reducing install size from ~400MB to ~50MB. Useful for embedded devices, CI pipelines, and serverless environments.
Both packages share the same codebase and version. pip install roboflow includes everything.
Install from source
You can also install the Roboflow Python package from source using the following commands:
git clone https://github.com/roboflow-ai/roboflow-python.git
cd roboflow-python
python3 -m venv env
source env/bin/activate
pip install .
Command line tool
By installing roboflow python package you can use some of its functionality in the command line (without having to write python code). See CLI-COMMANDS.md
🚀 Getting Started
To use the Roboflow Python package, you first need to authenticate with your Roboflow account. You can do this by running the following command:
import roboflow
roboflow.login()
Using Roboflow EU
The same package supports Roboflow's EU data-residency platform. Select it when logging in with the CLI:
roboflow auth login --region eu
For environment-based configuration and CI, set ROBOFLOW_REGION=eu before
running Python or CLI commands. EU and US use separate authentication
backends, so use an EU API key obtained from
app.roboflow.eu.
Authenticate with an API key
You can also authenticate with an API key by using the following code:
import roboflow
rf = roboflow.Roboflow(api_key="")
Quickstart
Below are some common methods used with the Roboflow Python package, presented concisely for reference. For a full library reference, refer to the Roboflow API reference documentation.
import roboflow
# Pass API key or use roboflow.login()
rf = roboflow.Roboflow(api_key="MY_API_KEY")
workspace = rf.workspace()
# creating object detection model that will detect flowers
project = workspace.create_project(
project_name="Flower detector",
project_type="object-detection", # Or "classification", "instance-segmentation", "semantic-segmentation"
project_license="MIT", # "private" for private projects, only available for paid customers
annotation="flowers" # If you plan to annotate lillys, sunflowers, etc.
)
# upload a dataset
workspace.upload_dataset(
dataset_path="./dataset/",
num_workers=10,
dataset_format="yolov8", # supports yolov8, yolov5, and Pascal VOC
project_license="MIT",
project_type="object-detection"
)
version = project.version("VERSION_NUMBER")
# upload model weights - yolov10
version.deploy(model_type="yolov10", model_path=f”{HOME}/runs/detect/train/”, filename="weights.pt")
# run inference (a version may own several trained models; models() returns all of them)
model = version.models()[0]
img_url = "https://media.roboflow.com/quickstart/aerial_drone.jpeg"
predictions = model.predict(img_url, hosted=True).json()
print(predictions)
Search and Export
Search for images across your workspace and export matching results as a ready-to-use dataset:
workspace = rf.workspace()
# Export images matching a search query
workspace.search_export(
query="class:person", # search query (e.g. "tag:review", "class:dog", "*")
format="coco", # annotation format: coco, yolov8, yolov5, voc, etc.
dataset="my-project", # optional: limit to a specific project
location="./my-export", # optional: output directory
)
Or from the CLI:
roboflow search-export "class:person" -f coco -d my-project -l ./my-export
Library Structure
The Roboflow Python library is structured using the same Workspace, Project, and Version ontology that you will see in the Roboflow application.
import roboflow
roboflow.login()
rf = roboflow.Roboflow()
workspace = rf.workspace("WORKSPACE_URL")
project = workspace.project("PROJECT_URL")
version = project.version("VERSION_NUMBER")
The workspace, project, and version parameters are the same as those you will find in the URL addresses at app.roboflow.com and universe.roboflow.com.
Within the workspace object you can perform actions like making a new project, listing your projects, or performing active learning where you are using predictions from one project's model to upload images to a new project.
Within the project object, you can retrieve metadata about the project, list versions, generate a new dataset version with preprocessing and augmentation settings, train a model in your project, and upload images and annotations to your project.
Within the version object, you can download the dataset version in any model format, train the version on Roboflow, and deploy your own external model to Roboflow.
🏆 Contributing
We would love your input on how we can improve the Roboflow Python package! Please see our contributing guide to get started. Thank you 🙏 to all our contributors!
4 maintainers
Metadata
4 maintainers
Release files for roboflow 1.6.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| roboflow-1.6.1.tar.gz | 314.5 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| roboflow-1.6.1-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 651.4 kB
Release files / roboflow-1.6.1.tar.gz
| Download URL | roboflow-1.6.1.tar.gz |
|---|---|
| Size | 314.5 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
3a57f2cbe8f0b34c663e0eda6df064eb87660d46ea750bf08ffd7f56f98f626a
|
|
BLAKE2b-256 checksum How to use checksums |
ace952647b86558b50b245ddbb61d806b6c5db3e4292c2d2abcc7d43f952cdf0
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 2, 2026.
Transparency logRelease files / roboflow-1.6.1-py3-none-any.whl
| Download URL | roboflow-1.6.1-py3-none-any.whl |
|---|---|
| Size | 336.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
43b85fb6fb24d9a7d64e2bf0c762ed1fdebd67a5d74d47149168396ea5a3f5d4
|
|
BLAKE2b-256 checksum How to use checksums |
ea8bd3827d7f227cd0d9c58495aa64a3f805b2294ceb0b67ef94b55a5c73048f
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 2, 2026.
Transparency log
