HomeWeave

Inspiration

A home needs a shared place for its records, a way to understand them, and control over who can see them. Manuals, invoices, photos and project decisions often sit across phones, inboxes and one person's laptop. UGREEN AI NAS offers a foundation for keeping those records together and available to authorized family members over the household network.

The challenge also reaches beyond the family. A repair technician may need an appliance's model and service history; a contractor may need project photos, plans and specifications. Preparing that information should let the owner choose exactly what to share while keeping unrelated household records private.

Those records also need to remain useful years later. Kitchen renovations, solar installations and appliance replacements can happen far apart. Inspired by real household energy and networking workflows, I designed HomeWeave to connect shared storage, household understanding, controlled disclosure and a history that grows with the home.

What it does

HomeWeave is a Uliya Skill for UGREEN AI NAS: a second brain for the household's energy, solar, EV charging, networking, appliances, maintenance, repairs and renovations. It turns selected documents, photographs, voice notes and saved knowledge into source-linked answers, explanations and practical project briefings.

Family members can use the NAS as a shared home for records, with access governed by native user and folder permissions. For a contractor or repair technician, HomeWeave prepares a recipient-specific draft and proposed supporting files, flags conflicts and missing evidence, and leaves the owner to review both the content and attachments before disclosure.

The owner can then share approved files or a dedicated project folder through UGREEN's native sharing controls, including a password, expiry date and download settings. HomeWeave prepares the material; creating the share and granting access remain owner-controlled actions in the NAS interface.

Ask: "Who installed our dishwasher, when, and which model?" Revisit a recorded decision or compare historical models and prices with their dates and scope visible. Inspect an interpretation, correct it, and optionally save a reviewed knowledge record with its sources and dates. The next question or project can build on that history.

How I built it

The first-stage package is a Markdown Uliya Skill with SKILL.md and three directly linked references. It uses registered file, document, photo and knowledge experts for bounded discovery, document questions, photo OCR, audio transcription and approved text saving. Each requested saved result must be reopened and verified.

There is no custom application runtime, database or device-control integration. The product is intended for macOS, Linux and Windows clients; platform behavior remains to be validated on the organizer's NAS.

Challenges and safeguards

A newer file does not automatically resolve conflicting evidence. A quote does not prove completed work. An unreadable photo label remains uncertain. Connections between records require supporting evidence, and source access does not authorize disclosure.

NAS storage does not prove fully local AI processing. Actual model routing, permissions and persistence await native validation. Only fictional records are used for contest preparation; no real household records have been uploaded.

Current progress

The household Skill, fictional household scenarios and a detailed multimodal gym fixture are authored. Local static checks, packaging checks and authored-contract reviews are complete. These are local preparation evidence, not proof of execution on UGREEN hardware.

What's next

If selected as a finalist, validate the Skill on the assigned UGREEN AI NAS. Test household questions and saving/reusing reviewed knowledge; demonstrate a fictional home-gym project combining documents, photos and a voice note; and check missing or contradictory evidence. Compare against ordinary native search/document chat using equivalent inputs before claiming an advantage.

AI tools and third-party resources

OpenAI Codex assisted research, planning, original Skill and proposal authoring, reviews and checks. OpenAI image generation created fictional fixture photographs and the original project thumbnail. ElevenLabs eleven_v4/River generated the fictional voice note; eleven_scribe_v1 checked its transcription. No real household records were supplied to these media services.

The UGREEN starter kit 0.7.3/catalog 2.0.0 supplied the authoring standard and checker; its code is not redistributed. ReportLab, Pillow, FFmpeg, Poppler, pypdf and Tesseract were development tools, not bundled runtime dependencies. Sources, licenses and data-flow details are recorded in the proposal and Skill disclosures.