Henji

Most AI writing tools hand you back something that technically answers the message and reads like it was drafted by nobody in particular. Henji, built by Appelier for macOS, takes a different approach: you type a rough note or a few keywords, select the text, press ⌘↩, and get a reply drafted in your voice — pulled from patterns in your past messages and sharpened by every edit you make afterward. The workflow sits on top of any Mac app, so it works in Gmail, Slack, LINE, Messages, and what the vendor describes as the vast majority of the over two million apps on macOS, without requiring per-app integrations or browser extensions.

The differentiating feature is the style memory layer. During setup, Henji reads your existing message history to establish a baseline voice. After that, each edit you accept or modify is stored — preferred phrases, tone tendencies, how you open and close a message — so the gap between ‘AI draft’ and ‘what I would actually send’ narrows over time. The recipient-switching feature extends this: you set contexts like Customer, Colleague, or Bestie, and the same underlying message adapts in register automatically, from formal acknowledgment to casual shorthand, without you rewriting.

The fit is narrow and deliberate. Henji works well for Mac users who send high volumes of short replies — team threads, client nudges, polite deadline negotiations — and who are willing to edit early drafts to train the model. The vendor states that data handling and style learning are described in the FAQ, and the privacy policy addresses server storage questions. Where it breaks: the tool generates one reply at a time from a fragment you provide; it does not read thread context autonomously, draft first messages, or handle anything requiring document attachments or calendar awareness. Teams that need those capabilities will find the tool’s scope a stopping point, not a workaround.