Skip to content

AGENTS.md

Latest commit

 

History

History
89 lines (70 loc) · 2.96 KB

AGENTS.md

File metadata and controls

89 lines (70 loc) · 2.96 KB

AGENTS.md

This guide governs the entire repository. If a subfolder provides its own AGENTS.md, instructions there override this file for that subtree.

Overview

Roboflow Inference is a set of Python packages that run computer vision models locally and expose them via an HTTP API and command line interface. The repo contains the core library, CLI, SDK, and Dockerfiles for building CPU or GPU images. Target Python version is 3.10 (minimum 3.8).

Project Structure

  • inference/ – core library with model loading and streaming utilities.
  • inference_cli/ – command line tools and server entry points.
  • inference_sdk/ – Python SDK for interacting with a running inference server.
  • docker/ – Dockerfiles used to build CPU and GPU images.
  • tests/ – unit and integration tests for all packages.
  • docs/ – mkdocs documentation source.

Setup / Environment

Create a Python environment and install the repo in editable mode:

conda create -n inference-development python=3.10
conda activate inference-development
pip install -e .
# optional models
pip install -e ".[sam]"

Important environment variables (see inference/core/env.py for all):

Variable Default Purpose
PROJECT roboflow-platform Selects prod or staging behavior
ROBOFLOW_API_KEY "" Enables authenticated requests
MODEL_CACHE_DIR /tmp/cache Stores downloaded models
PORT 9001 API port when running locally
NUM_WORKERS 1 Number of server worker threads

Defaults above mirror the Dockerfiles in docker/dockerfiles/.

Build & Running

Build a development image and start the server from the repository root:

docker build -t roboflow/roboflow-inference-server-cpu:dev \
    -f docker/dockerfiles/Dockerfile.onnx.cpu.dev .
docker run -p 9001:9001 \
    -v ./inference:/app/inference \
    roboflow/roboflow-inference-server-cpu:dev

Testing

Unit tests live in package specific folders. Run them individually with:

pytest tests/inference/unit_tests/
pytest tests/inference_cli/unit_tests/
pytest tests/inference_sdk/unit_tests/
pytest tests/workflows/unit_tests/

To run the entire suite while skipping slow tests:

pytest -m "not slow" tests/

Code Style

Format code with:

make style

Check linting and formatting with:

make check_code_quality

The repository follows PEP 8 and uses Black (88 characters), isort and flake8.

Contribution / PR Guidelines

  • Ensure all relevant tests pass before opening a pull request.
  • Keep commit messages concise and in the present tense, e.g. "Add model loader".
  • PR descriptions should explain what changed and why, list test commands run, and follow the templates in .github.
  • Update documentation when applicable.