I built a local AI writing assistant for Mac text fields

Most AI writing tools follow the same pattern: you write, copy the text, open a chat window, paste it, explain your request, copy the answer, then paste it back into the original app. This works fine for lengthy drafts but becomes irritating for the numerous small moments that occur daily, such as making an email less awkward, finishing a sentence, shortening a Slack reply, or cleaning up a paragraph.

The reporter wanted that assistance directly within the Mac text field they were currently using. That desire led to the creation of DraftKey, a native Mac application designed for private, on-device AI autocomplete and rewriting within macOS text fields.

The core issue, according to the reporter, is not just writing but the constant context switching. When writing an email, for example, the reporter does not want to switch to another app unless absolutely necessary. Most of the time, they already know what they want to say and merely need assistance with refining the wording, making it shorter, warmer, or more direct.

Autocomplete tools can be similarly disruptive, as opening a separate AI chat while halfway through a sentence breaks the flow more than it helps. The ideal system, the reporter believes, should feel more like spellcheck than a chatbot, appearing where you type, offering help when needed, and receding into the background when you continue writing.

Privacy is an essential consideration for everyday writing, which often involves sensitive content like client emails, internal notes, half-formed thoughts, personal messages, support replies, strategy notes, and rough drafts. While cloud AI offers powerful capabilities, the reporter argues that writing assistance carries a different expectation regarding privacy when it actively observes the text field you are editing.

DraftKey operates locally on the Mac, performing autocomplete and rewriting tasks without the need for constant internet access. Initial setup does require internet for tasks such as license activation, app updates, and the initial model download, but the actual writing assistance is designed for local inference.

The limitations of macOS text fields pose a challenge for this type of application. Some text fields provide sufficient information for a helper app to read the relevant text, display a suggestion, and apply the rewrite. However, others do not. Secure fields and custom editors may behave differently, so the app cannot guarantee compatibility everywhere. The reporter acknowledges this limitation, emphasizing that the app works within compatible macOS text fields with clear limitations while maintaining user control.

For DraftKey, the reporter envisions a UX that keeps users within the current app, allows them to select text or continue typing, prompt for a rewrite or autocomplete, review the suggestion, and accept it only if it improves the text. The crucial aspect is the review process, ensuring the user remains in control of their writing. The reporter believes that AI writing tools should provide a fast second pass for assistance instead of silently replacing the user's work.

An optional feature in DraftKey includes mechanical keyboard typing sounds, which are not integral to the AI functionality but serve as a fun bonus. Users can choose to enable or disable these sounds independently of the AI features. DraftKey caters to Mac users who frequently write and do not wish to navigate through browser tabs for minor edits.

Ideal candidates include developers working on issues, PRs, docs, and support replies, founders drafting emails and product updates, freelancers composing client messages, privacy-conscious individuals avoiding cloud writing tools for sensitive content, and users who prefer native Mac utilities over extensive web applications. The app requires macOS 14+, Apple Silicon, and sufficient memory for local AI features.

For users primarily seeking a chat assistant, the reporter recommends utilizing a full chat assistant instead. DraftKey focuses on the smaller, more frequent task of enhancing the text already present, aligning with the reporter's desired workflow on their Mac.