Melvin on macOS

Stop doing the work. Start directing it.

Melvin handles the tasks so you do not have to. Speak or type what should happen next: it reads the window you have open, drafts and moves across Mail, Calendar, the browser, and files. Mail stays in drafts until you send. That is the difference between another chat tab and a Mac assistant that actually runs the play.

1. The capability gap

Most chatbots only talk. Melvin is a director for your Mac. It reads what you allow on screen, plans, then acts in Mail, the browser, Calendar, and files with clear checkpoints before anything irreversible.

Feature

Standard chatbot

Melvin

Context

Blind. It only knows what you type unless you paste a long explanation every time.

Grounded. Melvin reads the active window, titles, and the screen you allow so "this email" resolves.

Execution

Talk only. It can describe steps but cannot operate Mail, Calendar, or your browser.

Director mode. Melvin runs multi step flows across macOS apps and the web—you direct each run and review output before it leaves your Mac.

Memory of you

Forgetful. Each chat starts cold unless you repeat your tools and people.

Local first. Melvin keeps your tools and people on this Mac so follow ups stay short.

Voice and pace

Typing heavy. Fine for essays, slow for quick corrections while you work.

Voice or type. Speak while Melvin floats at the edge of the screen on macOS.

2. The "this thread" test

User asks: "Summarize this thread and draft a polite no for me."

Standard chatbot

Reply: "I cannot see your inbox. Please paste the full thread here and I will summarize it."

Result: You do the copy paste work first.

Melvin

Melvin: "I see the thread in Mail. Three action items, two deadlines. Here is a two sentence no that matches your usual tone. Say send when you want it filed."

Result: Works from the window you already have open.

3. The "make tickets" test

User asks: "Turn this spec into tickets for engineering."

Standard chatbot

Reply: "Paste the spec text. Also tell me which system you use for tickets."

Result: You retype context it cannot see.

Melvin

Melvin: "Spec is open in the browser. I drafted six tickets with owners and estimates. Two items need your pick on priority. Ready to create them in your tracker when you say go."

Result: Plans from the doc on screen, not from memory guesses.

4. The "book it" test

User asks: "Find a thirty minute slot with Alex next week."

Standard chatbot

Reply: "I cannot access calendars. Try comparing availability manually."

Result: Stops at advice.

Melvin

Melvin: "Calendar is open. Three mutual gaps on Tuesday and Thursday. I can send an invite that names the project and attaches the doc you have focused."

Result: Moves from suggestion to a concrete next step on macOS.

5. The "do not ship that" test

User asks: "Double check this message before I send it."

Standard chatbot

Reply: "Paste the message. I will review tone and clarity in general terms."

Result: Generic editing without the surrounding thread.

Melvin

Melvin: "I see the draft and the last four replies. Two facts drift from the thread. Here is a tightened version that matches what you already promised."

Result: Grounded review tied to the conversation you are in.

6. What Melvin is made of

Melvin is not a generic web crawl. It is a macOS assistant that pairs observation with execution so your intent turns into finished work on this machine.

Observation: active app window title focused field allowed screen capture

People and tools: who is the team which inbox which tracker which browser profile

Ask pipeline: short answers long answers markdown citations when you ask

Execute pipeline: plan tool pick verify outcomes retry and verify

Recipes: Mail Calendar Messages Notes Safari Chrome Slack desktop

Browser: persistent profile sites you already signed into careful automation

Files: folder and file moves exports naming cleanup

Voice: dictation push to talk quick corrections

Privacy stance: local observation settings no screenshot hoarding you stay in control

Public beta: 30-day free trial then $26 per month managed model access developer key override