Jockbench - Revolutionary Agentic Video (AKA: Saddle Stack)

Edit your videos using natural language. Jockbench isn't just a demo, it is an open source server that allows you to deploy your own agentic video production service.

Comment

Your Complete AI Video Development Platform

A Brain Wave Collective Project

What's This All About?

Imagine going from zero to having a fully functional AI video processing platform in an hour. Not just a fragile demo, but a complete, extensible framework that you can actually build upon. We've combined NVIDIA AI Workbench's powerful infrastructure management with an enhanced version of Jockey's video framework to create something unique - a platform that's both incredibly easy to start with and robust enough for serious development.

The Magic (In Plain English)

Tell our system what kind of video you want, and it'll create it from your existing footage. For example, upload a few hours of Tour de France footage and ask for "a 4-scene video of Jonas Vingegaard climbing, challenging Pogacar, crossing the finish line, and celebrating on the podium" - a minute later, you've got your custom highlight reel.

What Makes This Special

  • Complete Package: Everything is covered - from remote server setup to video processing to web interface
  • Incredibly Simple Setup: Deploy and configure a GPU server with just a few clicks - no DevOps expertise needed
  • An Eye on Production: Built with real-world usage in mind, ready for you to build upon
  • Power Under the Hood: Leverages TwelveLabs' state-of-the-art video understanding models
  • Developer Friendly: Extensible architecture with clean separation between UI, backend, and processing
  • Open Source: We intend to make this entire project repository available for community use and contribution

Getting Started Is Easy

  1. Complete basic prerequisites (~30 minutes)
  2. Create/configure instance and deploy the Jockey server (~20 minutes)
  3. Launch the frontend and create your first video (~10 minutes)

That's it - no previous experience with Docker, Python, or GPU setup required. This could literally be your first introduction to all of the related technologies and products and we will walk you through it all.

Built For Growth

This isn't just another quick-start solution - it's a comprehensive platform designed to grow with your needs. Whether you're making your first AI-powered video or building a sophisticated video processing application, you'll find a solid foundation here. The architecture provides clear patterns for adding services, extending functionality, and scaling your application, making it an ideal starting point for exploring AI-powered video processing, GPU computing, or modern web architecture.

What's Next?

While our current implementation already demonstrates the power of bringing together NVIDIA AI Workbench with enhanced video processing capabilities, we're just getting started. We're continuing to develop this into an even more robust platform while maintaining its approachability. Future plans include exploring advanced AI Workbench capabilities, expanding the front-end features, and building out additional reference implementations. We are also working with related teams to discuss how we can best give back to the open source projects that led us here.

Technical Foundation

Built on the incredible novelty of Jockey (TwelveLabs' agentic video framework), TwelveLabs' unmatched video understanding capabilities, and NVIDIA AI Workbench's ability to abstract away complex environment configurations, we've created something that just works. Our journey involved extending Jockey for practical remote development, streamlining Workbench deployments, and creating a cohesive platform that lets you focus on building rather than configuring.

Challenges We Overcame

Building a seamless experience meant tackling some significant technical hurdles:

  • Getting Jockey (a "pre-alpha" framework) to run smoothly within containers - something it wasn't originally expected to do
  • Learning to work within NVIDIA AI Workbench's opinionated architecture while maintaining our vision for a user-friendly platform
  • Creating a robust interface despite neither of us being proper front-end developers
  • Navigating the complexities of a novel real-time video processing solution

Proud Accomplishments

Beyond being a team of two who managed to bring this all together, we're particularly proud of:

  • Creating a truly approachable platform for advanced AI video processing - something that typically requires significant technical expertise
  • Successfully extending Jockey's powerful functions to a robust remote server
  • Building a complete front-end application that showcases what's possible with this technology
  • Wrapping complex infrastructure management into simple, automated processes
  • Achieving our goal of making advanced AI technology accessible to more developers

The real achievement isn't just in the technical solutions we've created, but in how they come together to provide a foundation that others can build upon. We've transformed what would typically be days of painful setup and configuration into a process that can be completed in less than an hour, without sacrificing the power and flexibility that developers need for serious projects.

Socials

Thienthanh Trinh
LinkedIn: https://www.linkedin.com/in/thienthanh-trinh-71478bb0/
Twitter: @thienthanhtrinh
Discord: ttrinh2306 (1295925057285263461)

Daniel Ritchie
LinkedIn: https://www.linkedin.com/in/danielritchie123/
Twitter: @deploydan
Discord: quantumpoet (769583125579169812)

https://brainwavecollective.ai/

Built With

Share this project:

Updates

posted an update —

We've spent the last few days down rabbit holes to resolve what we thought were code issues, but turns out we found a bug with the service provider! They're investigating the issue, and we continue to improve the examples. Code will be public in a couple of weeks.

Log in or sign up for Devpost to join the conversation.