AI Disclosure

AI inside Premail

Email classification

The core AI feature in Premail is email classification. When a new message arrives, Premail sends the message, including the entire content of the email, to a language model and asks it to sort the message into a category with a confidence score. Premail uses that classification to apply the rules you have configured, such as archive, trash, label, or move to a folder.

Classification runs only on the accounts you connect and only to provide the filtering features you turn on. You stay in control of which rules act on the result.

Which AI models Premail uses

Premail supports three kinds of AI providers, and you choose which one to use:

• Anthropic (Claude) using your own API key. The default model is a small, fast Claude model.
• OpenAI using your own API key.
• Local (Ollama), which runs a model entirely on your own computer so no email content leaves your machine.

Where your email content goes

Premail is a local-first desktop app. When you use a bring-your-own-key provider such as Anthropic or OpenAI, the relevant email content is sent directly from your computer to that provider so the classification can run. Those requests do not pass through Premail servers, and the provider's own privacy and API terms apply to that processing.

If you use the local Ollama option, email content is processed on your computer and is not sent to Premail or to any third-party AI provider.

What AI in Premail never does

• Premail does not use your email content to train AI models.
• Premail does not send your email content to Premail-operated servers.
• Premail does not sell, rent, or broker your email content.
• Premail does not use AI to take actions in your mailbox beyond the rules you configure or the actions you initiate.

AI in how I build Premail

Premail is built by a single developer, and I use AI development tools as part of that work. This includes AI coding assistants that help write, review, and refactor code, as well as AI tools that help with documentation, support drafts, and marketing content. I review the work before it ships.

I also use AI to generate visual work, including images, illustrations, icons, designs, and page layouts you see across the app and this website. I treat these as starting points and refine them by hand, but I want to be upfront that AI is part of how the look of Premail comes together.

I mention this not because it is unusual, but because transparency about AI should include the parts most companies leave out. AI helps me move faster and ship more as a single developer. It does not replace human judgment about what Premail should do or how it should treat your data.

I do not feed your private data, email content, or account credentials into the AI tools I use to build the product. Those tools work on Premail's own source code and content, not on your inbox.

How I optimize my sites

A lot of how I build and optimize Premail's website, and my other sites, runs through a set of reusable AI skills I maintain in the open. These cover the work that goes into a fast, well-structured marketing site: SEO and meta tags, structured data, share images, sitemaps, prerendering, performance, accessibility, and more.

You can see exactly how this works, and the prompts and conventions behind it, at github.com/MattSenter/senternet-site-skills.

Human oversight

AI classification is a suggestion that drives the rules you set up, not an unsupervised authority over your inbox. You decide which rules act on a classification, you can review what Premail did in the Activity log, and you can adjust or disable any rule at any time.

Where AI gets something wrong, the design goal is that you can see it, understand why, and correct it.

Changes to this disclosure

As Premail's use of AI evolves, including the models I support and how I use AI to build the product, I will update this page and revise the “Last updated” date above.

Contact

If you have questions about how Premail uses AI, contact me at ai@premail.pro.