Inspiration

What it does

How we built it

Challenges we ran into## Inspiration

Software developers often spend more time rediscovering old problems than solving new ones. Build logs, release notes, bug reports, and technical decisions accumulate across folders, making it difficult to understand why a failure happened or what changed between releases.

I wanted to turn the UGREEN AI NAS into a private project memory that helps developers recover technical context from files they already own, without routinely uploading proprietary project data to external services.

What it does

ProjectVault is a read-only AI Skill designed for UGREEN AI NAS and its Uliya assistant.

Users select an authorized project folder and ask questions such as:

  • Why did the last three builds fail?
  • What changed between release 1.5 and 1.6?
  • Which performance issues keep recurring?
  • Where was a particular bug first documented?

ProjectVault is designed to discover relevant project files, extract technical facts, compare evidence across versions, and produce concise answers with source references, confidence levels, and unresolved questions.

It distinguishes documented facts from possible explanations rather than inventing a root cause when evidence is incomplete.

How I built it

For Round 1, I developed a Markdown-based SKILL.md package following the UGREEN Uliya Skill framework.

The workflow uses registered expert-agent delegation:

  1. File discovery: Delegate to file-agent using supported ug-filesystem capabilities to locate authorized project files.
  2. Document analysis: Delegate to document-agent using ug-document capabilities to extract relevant information from logs, release notes, and technical documentation.
  3. Evidence synthesis: Compare retrieved facts, identify recurring patterns, and produce source-grounded explanations.
  4. Error handling: Handle missing documents, contradictory records, permission restrictions, and incomplete agent results.

The initial package defines the orchestration instructions, permission boundaries, evidence policies, and validation scenarios. It does not require custom executables, external APIs, or modifications to the user's source files.

Challenges

The main challenge was designing a useful multi-agent workflow while staying within the officially documented UGREEN capabilities.

Another challenge was ensuring that answers remain reliable when logs are incomplete or multiple documents disagree. ProjectVault therefore treats evidence attribution, uncertainty, and failure handling as essential parts of the workflow.

Accomplishments

  • Created the initial ProjectVault Skill specification and pre-test package.
  • Designed a scoped, read-only retrieval and document-analysis workflow.
  • Defined structured outputs with supporting file evidence and confidence levels.
  • Added explicit handling for incomplete, conflicting, or inaccessible sources.
  • Prepared three validation scenarios covering recurring failures, release comparisons, and contradictory records.

This is a Round 1 pre-test submission. Official Creation Tool validation and testing on physical UGREEN AI NAS hardware remain pending.

What I learned

Reliable AI workflows require more than generating plausible answers. They need clear authorization boundaries, verifiable sources, controlled delegation, and honest handling of uncertainty.

What's next

If selected as a finalist, I plan to validate ProjectVault using the official UGREEN Skill Creation Tool and the provided AI NAS hardware, evaluate it against synthetic engineering project logs, improve document compatibility, and refine retrieval accuracy and response quality.

The long-term goal is to make a developer's NAS a practical, trustworthy memory of their technical projects.

Accomplishments that we're proud of

What we learned

What's next for ProjectVault — Local-first Project Memory for Developers

Built With

  • ai-agents
  • developer-tools
  • document
  • markdown
  • skill.md
  • ug-document
  • ug-filesystem
  • ugos
  • ugreen-ai-nas
  • uliya
Share this project:

Updates

Submission history