SaneBox has been doing folder-based triage since 2010 and does it without an app of its own. The trade it makes is deliberate: it looks at sender, subject and date, never the message body — so it can be private in a way Klorn is not, and it can never explain a decision.
| Axis | Klorn | SaneBox |
|---|---|---|
| How it decides | Reads the message, scores four features, then a deterministic rule maps them to a lane | Behavioural learning on header and metadata signals — sender, subject, date |
| Why it decided that | Shown on every row | Not shown — you infer it from where mail landed |
| Reads the body | Yes. That is what buys the explanation | No — headers and metadata only |
| Correcting it | Move the row; two matching corrections make that sender deterministic | Drag between folders; it learns from the move |
| Needs its own app | No, if you use the label mode; there are also web, macOS and Windows clients | No — it works inside whatever client you already use |
| Source & hosting | AGPL-3.0, self-hostable end to end | Closed, cloud only |
| Published price | $8.99/mo web · $9.99 native | $3.49 Snack · $5.99 Lunch · $16.99 Dinner |
“We never read your mail” is one sentence, it needs no qualifier, and Klorn cannot say it. Sorting mail into a lane with a stated reason requires reading the message. If header-only filtering is enough for you, that is a real advantage and we are not going to argue it away.
What Klorn offers instead is structural rather than rhetorical: the whole stack is AGPL-3.0 and self-hostable, you can point it at your own model key or a local model so message text never leaves your machine, and every classification is bound to a hash of the exact bytes that produced it. The short version is that you can own the server that reads your mail — a longer sentence than SaneBox's, and a checkable one.
Header signals cannot tell a scheduling request from a newsletter that mentions a date. Klorn routes scheduling mail into its own lane and cross-checks it against your calendar — the proposed slot, whether it clashes, and the first slots verified free for both sides. That is not a difference of quality; it is a difference of what information the system is allowed to see.
What we are not claiming. No head-to-head accuracy test between Klorn and SaneBox has been run. SaneBox's approach is described here as the vendor describes it publicly.
SaneBox is also the counter-example worth respecting: bootstrapped since 2010, profitable, and running on a fraction of the surface area. Being open source does not make Klorn better at what SaneBox is good at.