Both sit on top of Gmail and Outlook. They are answering different questions: Fyxer asks how do I get through this faster, Klorn asks should this have interrupted me at all.
| Axis | Klorn | Fyxer |
|---|---|---|
| What it produces | One classification per email, into five lanes — with the reason attached | A drafted reply in your writing style, plus inbox categories and meeting notes |
| Why it decided that | Shown on every row. Move the row and the correction becomes training signal | Not surfaced |
| Acting unattended | Approval-gated by default. Unattended replies only in AUTO mode, under guidelines you write, and every send writes a signed receipt | Drafts are prepared for your review |
| Source & hosting | AGPL-3.0, self-hostable end to end | Closed, cloud only |
| Your own model key | Yes — bring your own key, or run a local model | Not offered |
| AI spend | Hard daily cap. It stops rather than overspend | Included in the subscription |
| Published price | $8.99/mo web · $9.99 native | $30/user/mo Starter · $50 Professional |
Fyxer's pitch is an executive assistant, and it is priced and built like one — reply drafting in your voice, and a notetaker that joins Google Meet and Teams calls. Klorn does not join your calls. If what you want is someone to write the reply, that is the product built for it.
A drafting assistant is judged on the reply. An attention firewall is judged on the mail you never saw. That second one only works if you can audit it — which is why every Klorn decision carries the reason it was made, and why moving a message to a different lane feeds back as a correction rather than a one-off override.
The same logic runs the outbound side. Send, permanent delete and forward-external cannot ride on classifier confidence: each one requires a receipt minted at approval time that pins a hash of the exact payload, verified again at execution. If anything drifted in between, the action is refused.
What we are not claiming. We have not benchmarked Fyxer's drafting quality against Klorn's, and this page does not pretend otherwise. Everything above is each product's own public description.
Klorn's accuracy numbers live in committed, PII-scrubbed evaluation sets in the repository and are re-runnable with pnpm eval:judge — measured on one inbox's labelled mail so far, which is stated in the repo alongside the number.