Quickstart

This page walks through the shortest path from a fresh Python environment to a working inference call against a public Roboflow Universe model. Once you have this working you can swap in your own workspace and project.

1. Install

You need Python >=3.10.

pip install roboflow

If you only need lightweight CLI / Vision Events functionality without OpenCV / NumPy / Pillow, install the lite variant instead:

pip install roboflow-slim

2. Authenticate

Grab your API key from https://app.roboflow.com/settings/api, then set it as an environment variable:

export ROBOFLOW_API_KEY=rf_xxxxx

Verify:

roboflow auth status

The CLI prints your active workspace and a masked API key.

3. Run inference

The fastest way to see a result is the CLI against a public Universe model:

roboflow infer ~/Downloads/my-image.jpg -m poker-cards-cxcvz/1 -c 0.5

Or with the Python SDK:

Or with curl against the REST API:

4. Pick a tool to go deeper

The three developer tools serve different needs - see Choosing the Right Tool for guidance on when to use each.

• CLI - best for scripts, ad-hoc operations, and AI coding agents. See Using the CLI.

• Python SDK - best for scheduled jobs, notebooks, and integrations inside Python applications. See Using the Python SDK.

• REST API - best for non-Python applications and webhooks. See Using the REST API.

5. Try a workflow recipe

The Recipes section has end-to-end task walkthroughs that combine the CLI, SDK, and REST API:

• Upload, Train, and Deploy a Model

• Process a Video Stream

• Watch a Folder for New Images