IBCSC Institut belge de
cybersécurité citoyenne
EN
Support

Synology, AI and self-hosting 8 min read

AI in Synology MailPlus with Synology AI Console

Linking MailPlus to an AI service, producing conversation summaries, drafting and translating replies, then anonymising the data sent through de-identification.

Synology has started bringing AI into its ecosystem, and the first package served is Synology MailPlus. A new tool appeared for that purpose: Synology AI Console. Here is how it all works in practice.

What Synology AI Console is and which AI services it can work with belongs to the general overview of the package; this article focuses on the concrete use inside the mail client.

To use AI in Synology MailPlus you need version 3.4 or later. At the time of writing, those versions are still in beta. You then need to install Synology AI Console. Everything installs straight from Package Center, without any particular difficulty.

You then have to add, in the settings of Synology AI Console, the API key of the AI service of your choice. Synology publishes a dedicated procedure to obtain an API key or equivalent credentials for integrating a third-party AI. I went with OpenAI and the gpt-4o-mini model, whose requests stay inexpensive.

The edit window of an API integration in Synology AI Console
The API integration gathers the name, the provider, the key and the chosen generative model.

Then you go to the permissions section and grant the Synology MailPlus service access to AI. That is the end of the configuration.

The Permissions tab of Synology AI Console with AI enabled for Synology MailPlus
In the Permissions tab, access to AI is granted package by package — here Synology MailPlus.

The two uses of AI in MailPlus

In Synology MailPlus, the AI can do two things: produce a summary of an exchange, and help answer messages.

Summaries come in three formats:

  • Short summary — as the name says, it summarises very briefly, to the point that there is not much to say about it. Too little information.
  • Detailed summary — this one is excellent. It describes what was exchanged before, recalls what you answered and tells you how to proceed.
  • “Next steps” summary — a middle ground between the short and the detailed version.
A detailed summary of an e-mail exchange in Synology MailPlus
The detailed summary follows the thread message by message: initial request, answer, difficulties, contact details.
A second detailed summary split into sections in Synology MailPlus
On a longer exchange, the summary splits the history into sections and ends with what happens next.
The “Next steps” summary and the menu of the three summary types
The “AI summary” menu offers three formats; “Next steps” lists the actions expected from you.

When replying or writing a new message, the following functions are available:

  • Writing help — you state what you want to talk about and the AI produces a text you can then reuse in the message.
  • Smart draft — the AI composes the reply for you, based on the previous exchange. A real time saver.

Other options: improve quality, check spelling and even translate into another language. Useful functions, all in all.

The AI buttons and the translation menu in the compose window of Synology MailPlus
While composing: writing help, smart draft, proofreading, tone change and translation.

On the whole it all works very fast: once you get the hang of it, you can make life noticeably simpler. You only have to keep in mind that the requests are processed somewhere on the AI provider’s server — OpenAI in my case — and that this is therefore not entirely confidential.

Data de-identification

Synology has clearly measured the risk of leaking confidential data and has therefore added a data de-identification feature. It does come with conditions:

  • Install Container Manager from Package Center on your Synology NAS. That package is required to enable de-identification.
  • Your Synology NAS must have 8 GB of memory or more. The system checks the device’s available memory before enabling the feature.

On my test machine I only had 4 GB of RAM: the option simply did not appear. As soon as I raised the memory to 16 GB, a new section appeared in Synology AI Console, called “De-identification”, meant to secure the transmission of confidential data to third-party AI services.

The screen that enables data de-identification in Synology AI Console
With enough memory, the new “Data de-identification” section becomes available in AI Console.

During the setup you have to choose the languages you want. I picked French and English, which required 2 GB of extra memory. You then choose which data to anonymise. All of this affects how the AI works and how precise it is.

Selecting the languages for data de-identification
Each language you tick installs a language pack of roughly 1 GB of memory.
Checking the available memory before installing the language packs
Before installing, the system compares the available memory with the needs of the selected models.
Selecting the entities to de-identify in Synology AI Console
The masked entities are chosen one by one: the longer the list, the less precise the AI answer.
Downloading the de-identification language packs at 13 per cent
AI Console then downloads the language models; here the progress is at 13 %.
De-identification enabled with the English and French languages
Once enabled, the page lists the selected languages and entities separately.

After these operations, a new project and a container appear in Container Manager. The resource usage matches the values requested earlier.

The synology-ai-console project in Container Manager with the presidio-analyzer container
De-identification creates a project and a container in Container Manager, whose usage matches the requested resources.

De-identification did not slow the AI in Synology MailPlus at all: everything kept working just as well and just as fast. So if you have enough memory, you might as well enable it in Synology AI Console.

References

About the author

Jeremy Kraft runs the infrastructure and the self-hosted services of the IBCSC. These guides describe configurations that were actually deployed and re-checked before publication.

Ask a question All guides

Discussion

Comments

0

No comment yet. Feel free to open the discussion with a question or your own experience.