Inspiration

As video creators and digital artists, we found ourselves drowning in local media storage. We were spending hours hunting for specific drone clips or B-roll footage by arbitrary file names, and constantly losing track of which client contracts or licensing terms applied to which assets. Cloud-based AI tools offered solutions, but they required uploading massive raw video files and sacrificing privacy. We realized we needed a tool that combined semantic media search with document intelligence, running entirely on our local hardware. The UGREEN AI NAS provided the perfect environment to build this securely.

What it does

Agentia is a UGREEN AI NAS Skill that acts as a private, intelligent archivist for creators. When a user types a plain-text request (e.g., "Get my best city drone clips ready for stock sites"), Agentia uses the NAS's local AI to: Find matching video clips and images by semantic meaning. Read and extract licensing terms, client data, and expiry dates from associated contracts or invoices. Package the selected assets and a generated licensing text summary into a dedicated Agentia_Exports folder. Agentia enforces strict data safety: it never deletes, moves, overwrites, or shares files, and all processing remains strictly on the local NAS.

How we built it

We architected Agentia using the UGOS Skill Creation Tool, leveraging the native capabilities of the UGREEN AI NAS. We designed the workflow to chain multiple core CLI capabilities: using Video and Files for semantic media retrieval, and Documents for contract intelligence. We established a strict orchestration flow that requires manual user selection for contract matching and enforces a hard confirmation gate before any file copying occurs. Large Language Models (Gemini and Claude) were utilized during the Hackathon to structure the SKILL.md syntax and audit our workflow logic against the Uliya Skill Development Standard.

Challenges we ran into

Our biggest challenge was aligning our initial vision with the actual capabilities of the UGOS Skill Creation Tool. Initially, we planned to automate contract matching and generate dynamic video previews. However, after auditing the API capabilities against the hackathon rules, we had to strip out unsupported features (like preview generation and automated contract matching) and redesign the workflow to rely on manual user selection for contracts. Ensuring the workflow handled edge cases safely—such as missing contracts or partial user confirmations—required significant iterative logic refinement.

Accomplishments that we're proud of

We successfully designed a cohesive workflow that bridges media retrieval and document intelligence without relying on external APIs. We are proud of our strict adherence to user privacy and data safety, specifically our robust confirmation gates and immutable file handling rules that guarantee the user's core library is never altered or exposed.

What we learned

We learned the critical importance of defensive system design when building automated agents for file management. Moving from a theoretical workflow to a strictly bounded orchestration plan taught us how to handle failure states gracefully (e.g., handling unreadable contracts or missing expiry dates) without breaking the core functionality.

What's next for AGENTIA

If selected for Round 2, we plan to test Agentia on the UGREEN AI NAS iDX6011 Pro hardware to optimize the semantic search accuracy on large, real-world media libraries and refine the contract extraction models for complex legal documents.

Built With

  • document-extraction
  • local-ai
  • markdown
  • python
  • semantic-search
  • ugos-skill-creation-tool
  • ugreen-ai-nas
Share this project:

Updates

Submission history