Ollama on a Synology NAS: an LLM on the server familial

#Which Synology NAS devices are compatible with Ollama

Memory is the first thing to check before any purchase or installation attempt. Ollama publishes Linux versions for both x86-64 and ARM64, so the ARM processor is not the real blocker. However, entry-level Synology NAS devices built around a Realtek ARM processor have only 1 GB (DS124, DS223j) or 2 GB (DS223) of memory, according to Synology's specifications, while a 7B model quantized in Q4 weighs about 5 GB by itself. x86 models (Intel Celeron or AMD Ryzen, generally the “Plus” series) are the realistic target because their memory can be expanded.

In practice, the “Plus” and higher-tier series, equipped with an Intel Celeron or an AMD Ryzen, are the ones that work. Before installing anything, check your exact model's specifications: the series name (Plus) is a clue, not a guarantee by itself.

#How much RAM to plan for realistic use

RAM requirements vary based on the loaded model, quantization, context length, and number of concurrent users; you also need to leave headroom for DSM and other packages running in parallel. A 7B model in Q4 uses about 5 GB for its weights, plus context and processes: plan for 6 to 8 GB of free RAM. On an 8 GB NAS, headroom becomes very limited once DSM and Container Manager are accounted for; 16 GB provides comfortable operation.

The DS225+ and DS425+ ship with only 2 GB of RAM by default: that is far from enough for a usable LLM, even a small model. A memory upgrade is essential on these models before installing Ollama.

#Which Synology models to prioritize

The Ryzen-equipped models (DS725+, DS925+, DS1525+, DS1825+) offer expandable RAM up to 32 GB and a CPU better suited to computation than the Celerons in the DS225+/DS425+ lines. For serious LLM use, these are the ranges to target, with a memory upgrade to the maximum supported amount rather than the original configuration.

#Install Ollama via Container Manager

Synology does not offer Ollama as an official package in Package Center: you have to use Container Manager, which replaced the former Docker package starting with DSM 7.2 (on DSM 7.0 and 7.1, the former Docker package is still available).

1. 01
   Check the DSM version

   Container Manager requires DSM 7.2 or later. On an earlier version, you need to use the old Docker package, which has a different interface.
2. 02
   Open Container Manager and search for the official image

   From the DSM main menu, open Container Manager, go to Registry, search for ollama/ollama, select the official image, and click Download.
3. 03
   Configure the port and start the container

   Map the container port (11434 by default for Ollama) to a high-numbered range on the host, along the lines of the 8000-9000 range recommended for this type of service, rather than changing DSM's default ports.
4. 04
   Download a First Model Suited to the Available RAM

   Once the container is running, connect via the command line or the interface and retrieve a model sized for the RAM actually available on the NAS, not the total installed RAM.

• Reminder: install Ollama with Docker
• Source: detailed Container Manager steps

#Add an interface: Open WebUI in Docker Compose

Ollama alone provides only an API and a command line, not a chat interface: to get an experience close to ChatGPT, you need a second piece of software connected to it. The recommended approach is to deploy Ollama and Open WebUI as two separate Docker containers on the same NAS, then obtain a model sized for the RAM actually available rather than the most capable model found online—the model choice remains governed by the same memory constraints as Ollama alone, described above. Since DSM 7.2, Container Manager has offered a “Project” tab that runs a docker-compose.yml file directly, without using SSH or typing commands in a terminal, bringing the Synology experience closer to that of a standard Linux server while remaining controlled from the DSM graphical interface.

1. 01
   Create the project folder

   In File Station, create a dedicated folder, such as /volume1/docker/openwebui, to hold the project's docker-compose.yml file; keeping this folder separate from the one used to install Ollama alone prevents the two containers' configurations from getting mixed together.
2. 02
   Write the docker-compose.yml file

   Declare two services: ollama (the API already installed in the previous step, or reintroduced here) and open-webui, connected to Ollama through its internal URL, with port 3000 exposed on the host for the web interface.
3. 03
   Launch the project from Container Manager

   Create a new Project in Container Manager, point it to this folder, then deploy: the underlying command is equivalent to docker compose up -d and starts both containers together, in the correct dependency order.
4. 04
   Open the interface and create the admin account

   Once the containers are running, open the NAS IP address followed by port 3000 in a browser on the local network, create the administrator account on first access, then retrieve a model from Ollama directly in the interface.

• Complete guide Open WebUI with Ollama
• Source: Open WebUI on NAS using Docker Compose

#Realistic expectations with CPU only

A Synology NAS runs Ollama on the CPU only: there is no dedicated GPU to use on these machines. Actual throughput must be measured on the target model and workload, because it varies with context length, the Ollama version, and concurrent NAS activity (backups, indexing, and other packages). A qualitative ballpark observed for long generation is around 5 tokens per second—more than sufficient for asynchronous use (summarization, background processing), but more limiting for a demanding live conversation.

#Network security: do not expose Ollama to the Internet

A NAS generally contains all the personal or business files of a household or small organization: it’s a machine to protect as a priority, not expose for convenience. Forwarding a DSM port or container port directly to the Internet places a device containing all your files directly in the path of automated scans, and Synology NAS devices have already been targeted by ransomware campaigns for precisely this reason.

#Compared with a production Docker deployment

This guide targets a simple, single-container installation on a home or small-business NAS. For a production stack with an HTTPS reverse proxy, dedicated web interface, vector RAG, and monitoring, the approach is different and more complex to operate: it is covered by this site's production Docker Compose guide, designed for a dedicated Linux server rather than a NAS.

• Production LLM deployment with Docker Compose
• Alternative: Ollama on Proxmox (LXC/VM)
• Source: CPU and RAM compatibility by model
• Source: Container Manager and network security

#Who this setup makes sense for

A Ollama on a Synology NAS is suitable for anyone who already has an x86 NAS running for storage and wants to add a light, asynchronous AI workload accessible across the local network: summarizing documents placed in a shared folder, extracting information from archived emails, or serving as a component in a home automation system. It is not the right target if the goal is a smooth, fast everyday conversation, or if the NAS already serves as the primary backup and has no memory headroom for an LLM without risking slower performance for its usual tasks.

You already have an x86 Plus NAS with free RAM

This is the most cost-effective scenario: no additional hardware to buy, just a possible memory upgrade and a container to configure.

You want an instant conversation like ChatGPT

A NAS CPU alone remains slow compared with a dedicated GPU or a recent unified-memory chip; for this use case, a mini PC or a machine with a GPU is a better choice.

The NAS already hosts critical backups

Check the available RAM and CPU load before adding Ollama: an undersized service can slow existing backup tasks.