Wurian — Search With Context, Not Just Keywords
Inspiration
Imagine searching your NAS for a beautiful sunset photograph to use as your wallpaper.
Your AI finds a family vacation photo, a university assignment, a work presentation, and a screenshot containing sensitive information.
All of them contain a sunset. But are they equally relevant?
Recognizing what appears in a file is not the same as understanding why that file matters.
That simple observation inspired Wurian.
What It Does
Wurian is a privacy-first, context-aware AI Skill designed for UGREEN NAS.
It introduces a contextual decision layer between what the NAS can find and what the user actually needs.
Instead of relying only on keywords or visual similarities, Wurian considers three questions:
- Intent: What is the user trying to accomplish?
- Context: What do we know about the file's origin and purpose?
- Relevance: Does the available evidence justify including this result?
The Skill is designed to distinguish personal memories, professional documents, academic material, and sensitive information without unnecessarily mixing them.
When evidence is insufficient, Wurian identifies uncertainty rather than inventing a classification.
Files remain in their original locations. No moving, renaming, or deleting is required.
How We Built It
Wurian was designed using the official UGREEN Uliya Skill Development Standard.
Its orchestration workflow coordinates existing UGREEN AI capabilities:
- file-agent: authorized file discovery and metadata.
- photo-agent: photograph and screenshot retrieval.
- document-agent: additional evidence from authorized documents when needed.
Wurian uses explicit search boundaries, contextual eligibility rules, and explainable decisions to evaluate retrieved results.
The initial Skill package includes its instructions and supporting Markdown references. It successfully passed the official Uliya v0.7.7 static validation with zero errors and zero warnings.
We also explored local AI inference in separate Windows and Android development environments to better understand the opportunities and constraints of local processing.
Challenges We Faced
The biggest challenge was realizing that AI can recognize content without reliably understanding its context.
A visually identical image may belong to a private memory, a business document, or an academic project.
File names may be misleading. Metadata may be incomplete. User intent may be ambiguous.
We learned that effective contextual search requires more than a powerful model. It needs permission boundaries, reliable evidence, transparent reasoning, and the ability to say, "I don't know."
These principles shaped Wurian's conservative, read-only design.
Current Status
The Skill specification, orchestration workflow, and English submission package are complete and statically validated.
End-to-end execution on UGREEN NAS and real-world contextual classification accuracy still require testing.
What's Next
If selected as a finalist, our next milestones are:
- Classify 20 screenshots into appropriate contexts.
- Retrieve personal sunset photographs while excluding unrelated or sensitive material.
- Evaluate ambiguous files, missing metadata, and uncertain classifications.
We aim to demonstrate that context-aware retrieval can make AI NAS storage more useful without compromising the integrity of original files.
Our Vision
Most AI search systems ask:
"What does this file contain?"
Wurian adds another question:
"Does this file actually belong to what you are trying to do?"
Because the future of intelligent storage is not simply finding more files.
It is finding the right files, for the right purpose, in the right context.
Built With
- ai-agents
- markdown
- semantic
- ugos
- ugreen-nas
- uliya
- yaml