Media Vault Pipeline

Inspiration

What it does

How we built it

Inspiration

Independent creators, videographers, and commercial studios generate hundreds of gigabytes of raw footage per shoot[cite: 2]. Uploading terabytes of unreleased footage to public cloud platforms introduces severe client confidentiality risks, prohibitive subscription fees, and multi-hour bandwidth bottlenecks[cite: 2]. We asked ourselves: why rely on expensive cloud SaaS when modern NAS hardware like UGREEN AI NAS features desktop-grade NPUs, massive local RAID storage, and 10GbE local network speeds[cite: 1, 2]? This inspired Media Vault Pipeline—a zero-touch, privacy-first ingestion and storage tiering pipeline that runs entirely on-premise[cite: 2, 5].

What it does

Media Vault Pipeline turns UGREEN AI NAS into an autonomous media ingestion and archiving powerhouse[cite: 1, 2]:

• Zero-Touch Intake: Ingests media dumps from external drives or staging directories into NVMe storage[cite: 2, 9].
• Local AI Understanding: Utilizes on-device NPU models to perform Whisper speech transcription and CLIP visual semantic tagging without any internet connection[cite: 2, 5].
• Portable Metadata: Automatically generates co-located JSON sidecars containing technical EXIF/codec data and AI tags alongside raw files[cite: 5, 8].
• Automated Storage Tiering: Intelligently moves heavy raw footage from high-speed NVMe staging to high-capacity mechanical RAID cold storage while retaining searchable pointers in Uliya[cite: 2, 5].

How we built it

We engineered the solution strictly following the official Uliya Skill Standard[cite: 1, 5]:

• Orchestration Architecture: Designed two coordinated expert agents (ingest-agent for file discovery and validation; tiering-agent for cold archival lifecycle management)[cite: 2, 5].
• UGOS Integration: Routed file system operations, metadata indexing, and asynchronous job queues through native UGOS CLI capabilities (ugos-fs-cli, ugos-media-cli, ugos-ai-cli)[cite: 5, 9].
• Sidecar Architecture: Avoided brittle proprietary databases by writing self-contained {file_id}.json descriptors next to each asset[cite: 8].

Challenges we ran into

• Ensuring Zero Unauthorized Executables: Adhering strictly to Uliya guidelines meant avoiding external binary wrappers or custom background daemon scripts[cite: 1]. We solved this by designing a declarative state machine that delegates execution entirely to native platform CLI capabilities[cite: 1, 5].
• Balancing Ingestion vs. Heavy Inference: High-bitrate ProRes/RAW video requires massive I/O ## Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Media Vault Pipeline

Built With

• ai
• clip
• exiftool
• ffmpeg
• json
• linux
• npu
• storage
• ugos
• whisper