Video Summary - Zettlab AI NAS | Almost the PERFECT Media Server & Local AI NAS!.. or is it?

Product reviewed

Zettlab (Zetlab) D6 AI NAS (prototype) — a compact NAS with 6× 3.5” SATA bays plus 1× M.2 NVMe, built around an 8-core Rockchip RK3588 ARM chipset. The title frames it as potentially “almost the perfect media server & local AI NAS,” but the review questions whether Zetlab delivers on that promise.

Key features mentioned

Hardware / design

6-bay 3.5” SATA NAS with high-quality removable trays (slide-in tray mechanism)
1× M.2 NVMe slot
Aluminum housing, with front LED accents and an LCD status screen
Front controls/slots
- Magnetic removable front grill
- SD card slots: pair of SD + microSD
- USB 3 Type-C + USB 3 Type-A plus a copy button
- The copy feature can automatically copy files from external media to the NAS
Rear I/O
- Dual Ethernet: 1GbE + 2.5GbE
- Additional USB-A 2.0 ports
- HDMI port
- DC power input
Includes magnetic mesh for the rear for ventilation/clean look

Compute platform

Uses Rockchip RK3588 (8-core ARM) (not the typical x86/Intel-based approach)
Expected tradeoff: lower power/heat, potentially slightly less performance

On-device “AI” / ZOS features (important caveat)

ZOS analyzes files/metadata to build an internal database, including:
- photo descriptions
- video transcripts
- EXIF data
- other extracted snippets
Learned/processed files show an AI status such as “learned”
Zet AI app enables natural-language search over scanned content
Example tested: upload local YouTube videos → ask the NAS to find videos by topic using generated transcripts

AI compute location (major limitation by model)

For D4 and D6: AI chat/document scanning does NOT run on the NAS itself
- Instead, you’re told to use desktop software on your PC
- That desktop workflow uses NAS data plus an LLM running locally on your machine
To run LLMs on the NAS itself, you must upgrade to D6 Ultra or D8 Ultra
- Those use Intel Core 5 Ultra 125H, enabling on-NAS LLM execution

Media / media-server performance claims

Runs Jellyfin, Home Assistant, Transmission, etc.
Streaming: no issues streaming 4K movies while other tasks were running
Performance comparison (generalization given by reviewer):
- RK3588 is ~20% slower on average than Intel N100 in single-thread and multi-thread workloads
RAM: 16GB soldered LPDDR4X
Docker note: Jellyfin runs in a Docker container, implying Docker runtime exists, but the full “container UI” experience is limited

Software / OS

ZOS based on Debian Linux (reviewer likes Debian for stability and package quality)
Praised:
- Nice UI
- Particularly easy pool setup and Jellyfin setup
Promised-but-not-yet-available features in this prototype/software stage:
- Docker front-end
- virtual machines
- SSH access
- additional media organization features (e.g., dedicated photo gallery / movie front end)
Reviewer-confirmed limitations:
- Can run published third-party apps
- Cannot SSH in, so deeper inspection/kernel/service tweaks weren’t possible
- Zetlab says SSH and third-party OS install support are planned for next year

Pros (as stated)

Strong hardware build quality/design
Excellent front LCD status display (shows IP address + CPU/memory/storage usage)
Easy setup/UI, including Jellyfin configuration
AI-based natural language search over NAS-scanned metadata is “pretty cool”
Real-world media performance is sufficient
- 4K streaming works even with other tasks
Debian-based OS is a positive (stability, community support)

Cons / downsides mentioned

Prototype/software is alpha-like; many features are still “promises”
AI limitation on D4/D6
- AI chat/scanning integration requires desktop software and a local LLM on your PC, not purely on-NAS
Missing user experience for Docker
- No full “download and run arbitrary containers” UI outside the app store
No SSH access currently
- Reviewer calls this “disappointing” and a major point of contention
Crowdfunding risk
- Kickstarter pledges are donations; no guarantee of product delivery (even though the reviewer believes the team is capable)

Comparisons made

vs Intel N100 (performance): RK3588 is described as ~20% slower on average (single and multi-thread).
vs x86 NAS norm: RK3588 is the first NAS in the reviewer’s testing using a high-end ARM SOC rather than embedded x86.
Model lineup comparison for AI capability
- D4/D6: AI workloads require desktop/local LLM on your machine
- D6 Ultra / D8 Ultra: AI workloads can run on the NAS due to stronger CPU (Core 5 Ultra 125H)

User experience notes

Physical experience

Clean, premium enclosure
Easy tray drive insertion
Magnetic front grill

Operational experience

LCD makes it easy to identify the NAS without hunting in software
File import/copy automation via USB/SD and copy button

Media server experience

“Perfectly acceptable” performance
4K streaming works alongside other services

AI experience

Scanning/learning appears in UI
Natural language search works using transcript-derived metadata (tested with local video uploads)

Overall verdict / recommendation

The reviewer is excited about the Zetlab D6’s hardware quality, LCD/UX polish, and the concept of AI-assisted local search, and says it can serve as a solid media NAS (including 4K streaming).

However, it’s not yet a complete “perfect local AI NAS” because:

the software is still early/alpha
key features (Docker UI, SSH, extra organization tools) are promised but missing
true on-NAS LLM functionality is not available on D6/D4 (requires Ultra models)

Recommendation implied: Consider it promising (especially for media + metadata AI search), but only as a good bet if you’re comfortable with early-stage software limitations and the AI compute caveats—or plan to move up to D6 Ultra / D8 Ultra if you need AI running directly on the NAS.

Unique points mentioned about the product (deduplicated list)

Prototype reviewed; hardware/software may change by retail.
6× 3.5” SATA bays + 1× M.2 NVMe.
Aluminum housing, stylish LEDs, magnetic removable front grill.
High-quality drive tray mechanism.
Front: dual SD slots + microSD, USB-C, USB-A, and copy button for automated imports.
Rear: 1GbE + 2.5GbE, USB-A 2.0, HDMI, DC power.
Cooling/vent design; optional magnetic rear mesh.
Rockchip RK3588 ARM platform; ARM approach vs typical x86.
Expected lower power/heat; potential ~20% slower than Intel N100 (general benchmark note).
16GB soldered LPDDR4X.
LCD shows IP + CPU/memory/storage usage.
ZOS UI is well-designed; storage pool and Jellyfin setup are easier than other NAS tested.
ZOS AI scanning creates a database of metadata (photo descriptions, transcripts, EXIF, etc.).
Files show “AI status/learned” after scanning.
Zet AI app supports natural-language queries over scanned content.
Tested example: searching local YouTube uploads using transcript relevance.
Critical caveat: AI chat/document scanning doesn’t run on D6/D4 NAS itself; uses desktop software + local LLM.
On-NAS LLM requires D6 Ultra or D8 Ultra with Core 5 Ultra 125H.
Performance: acceptable for Jellyfin, Home Assistant, Transmission.
4K streaming works while running other services.
Promised-but-missing: full Docker front-end, VMs, SSH, photo/movie gallery UI, etc.
Third-party apps via app store are sufficient for many tasks (but arbitrary container UI is missing).
Jellyfin confirmed runs in Docker container (OS has Docker runtime).
OS is Debian-based (positive stability/package quality).
SSH and third-party OS install planned next year; reviewer couldn’t evaluate deeply due to no SSH.
Kickstarter/crowdfunding disclaimer (risk of non-delivery).
Community openness: Discord engineers responding.

Speakers’ contributions

Logan (main reviewer/narrator): Provided product details, testing notes and caveats (AI compute limitations, Docker/SSH gaps), performance comparison, and framed the final recommendation.

Zettlab AI NAS | Almost the PERFECT Media Server & Local AI NAS!.. or is it?

Key takeaways