Synology NAS and AI — what's official, what's DIY
What works
- AI photo library (faces, albums, search) on Synology Photos — official, free
- 3B-class local LLMs via Container Manager + Ollama on 8 GB x86 models
- 7-8B models with a RAM ceiling you accept (slow, CPU-only)
What breaks / isn't offered
- No official Synology LLM/assistant package (feature request still open)
- No NPU in any current Synology consumer/plus model
- ARM models cannot run mainstream Ollama x86 images
Key facts (as of 2026-11)
- Synology Photos provides on-device face/object/location recognition — the official AI feature Synology ships on every photos-capable model.
- No official LLM integration exists as of 2026-11; Synology's community forum has an open feature request thread for local LLM support.
- The working route on x86 DiskStations is Container Manager (Docker) + Ollama; ARM models (J-series) cannot run the common x86 images.
- The 2025/2026 Ryzen V1500B generation (DS425+/DS925+/DS1825+) has no integrated GPU — every LLM token is CPU-generated, which caps usable model size.
- RAM ceilings: DS925+ ships 4 GB ECC, DS1825+ 8 GB ECC — 7-8B Q4 models are the practical ceiling after DSM takes its share.
Sources
- Synology Photos product page (official)
- Synology community: feature request for local LLM integration
- Community guide: running Ollama on Synology NAS
FAQ
Can a Synology NAS run ChatGPT-like AI locally?
Not as an official feature. As of 2026-11 Synology ships no LLM package; the route is Container Manager (Docker) running Ollama with a small open model (3-8B). It works on x86 models with enough RAM, but it is community-supported, not Synology-supported.
Does Synology Photos AI run on the NAS or in the cloud?
On the NAS. Face, object and location recognition in Synology Photos runs on-device for supported models — no photo data needs to leave your DiskStation for indexing.
Which Synology is best for local AI?
An x86 model with the highest RAM ceiling you can get — the more RAM, the larger the model. The 2025 Ryzen generation (DS925+/DS1825+) works but has no iGPU, so expect CPU-speed inference. See our local LLM hardware table for model-size fit.
For model-size fit on specific boxes see the local LLM hardware table and the Ollama requirements page. Then come back to what this site does best: every model's official drive compatibility list.