We're all-in on Open Source

Production-ready models and datasets for brand safety and content moderation.

Models

vision-1

Our flagship model for brand safety classification, built on Meta Llama 3.1 (15GB). Full precision model optimized for high accuracy.

  • Meta Llama 3.1 architecture
  • Trained on BrandSafe-16k
  • High accuracy for production
  • Advanced GPU optimization
Hugging Face

vision-1-mini

Lightweight version of VISION-1 optimized for Apple Silicon and smaller compute resources. Perfect for local deployment.

  • Llama 3.1 8B architecture
  • GGUF V3 quantization
  • Fast 3.27s load time
  • Efficient 4.58 GiB size
Hugging Face

Datasets

BrandSafe-16k

Comprehensive dataset for training brand safety classification models, featuring 16 distinct risk categories.

  • 16 risk categories
  • Brand safety focused
  • High-quality annotations
  • Production-ready data
Hugging Face

Developer Tools

Overseer SDK

Official TypeScript/JavaScript SDK for integrating Overseer AI validation into your applications.

  • Simple API integration
  • Type-safe interfaces
  • Comprehensive documentation
GitHub npm

Python Client Coming Soon

Official Python client library for Overseer AI, enabling seamless integration with Python applications.

  • Async support
  • Pythonic interface
  • Extensive examples
GitHub
PyPI

Join Our Community

Help us build the future of AI safety. Contribute to our open-source projects or start a discussion.

View on Hugging Face View on GitHub