# FERAL

> Supervised raw-video-to-frame-label toolkit for animal behavior. Runs on user-provided NVIDIA GPUs; annotations define the behavior vocabulary.

The website is the documentation source. Inspect the installed CLI help before using workflow commands.

## Documentation

- [Released CLI and Python API](https://www.getferal.ai/docs/reference.md): CLI commands and packaged configuration without browser execution.

- [Run FERAL](https://www.getferal.ai/docs/workflow.md): labels schema, validation, local training, smoke checks and checkpoint-only inference.
- [Hardware and recipes](https://www.getferal.ai/docs/hardware.md): measured GPU requirements, runtimes and recipe trade-offs from bioRxiv v2.
- [Agent skill](https://www.getferal.ai/skills/feral-run/SKILL.md): execution guidance for an agent with videos, labels and authorized compute.
- [Full website](https://www.getferal.ai/): installation, CLI and Python API reference.
- [Source](https://github.com/Skovorp/feral): implementation, tests and releases.
- [Starter checkpoints](https://www.getferal.ai/pretrained_checkpoints/): CalMS21, Adult-larva and PanAf500 checkpoints.
- [Paper v2](https://www.biorxiv.org/content/10.1101/2025.11.16.688666v2.full.pdf): methods, evaluation and hardware comparisons.