Premail v1.0.5: Less Setup, More Trust
Prefer the short version? View the v1.0.5 release notes.
Premail v1.0.5 is out, and this release is mostly about removing friction. Premail watches your inbox and uses AI to help classify what belongs there and what does not, but for that to feel useful, the surrounding experience has to be boring in the best possible way.
It should launch when your computer starts. It should know which models you can actually use. It should be easy to move to another machine, and its classifications should be consistent enough that you can trust the rules you put in place. v1.0.5 makes meaningful progress on all of those fronts.
Your Premail setup is now portable
The biggest new feature is Backup & Restore. Once you have Premail configured the way you like it, rebuilding that setup on another machine should not become a project. Whether you bought a new computer, are replacing an old one, or simply want a backup before experimenting with your configuration, Premail can now export your settings and restore them elsewhere.
That can sound like a small feature until you need it. Then it becomes one of those details that makes the difference between software feeling disposable and software feeling like something you actually own. Our complete Backup & Restore guide explains exactly what travels, what stays safely behind, and how API keys are handled.
Premail now knows which models you can use
AI providers move quickly. Models appear, models disappear, access varies by account, and a model name that works for one person may not work for another. Premail previously relied more heavily on you knowing what to enter, which made setup more uncertain than it needed to be.
In v1.0.5, Premail can scan the models available to your configured OpenAI or Anthropic API key and show you what your account can actually use. Configure your provider, let Premail inspect what is available, and choose from real options instead of copying model names out of documentation and hoping your key has access.
Better classification, especially with GPT-5 mini
We have also spent time refining the classification prompts used by Premail's rule packs, with particular attention to how the GPT-5 family interprets instructions, context, and rule criteria. These are not generic prompt tweaks: Premail's classification behavior has been deliberately optimized for GPT-5-family models so existing rule packs can make more consistent decisions without asking you to rewrite them.
If you use OpenAI with Premail, we now recommend GPT-5 mini as the default model. That recommendation is not just about cost or speed. Email classification is a high-volume workload, and GPT-5 mini offers the balance we want across classification quality, latency, and cost while taking advantage of the rule-pack improvements in this release.
You can still use another supported OpenAI or Anthropic model if you prefer. Premail remains intentionally flexible, but GPT-5 mini is the starting point we currently recommend for OpenAI users.
Premail should be running before you think about it
We also fixed a bug that could sometimes prevent Premail from launching when your machine boots. This matters more than it sounds because Premail is background software: you should not have to remember to launch your email protection layer every morning. If you told Premail to start automatically, it should simply be there.
The boring work matters
v1.0.5 includes additional bug fixes and smaller improvements throughout the app. Not every release needs a giant new feature; a lot of good software development is steadily eliminating the moments where the user has to stop and think about the software itself.
That is what this release is about. Premail is easier to move, easier to configure, smarter about the AI models available to you, more reliable at startup, and more consistent when applying your rules. If you are already using Premail, update to v1.0.5. If you have been waiting for Premail to feel a little more mature before giving it a try, this is a good release to jump in.