Featured · 02 · Native macOS AI agent
Movi app icon

Movi

Talk to your Mac. Movi does the work.

An autonomous AI agent that lives in the MacBook notch. Say what you need — Movi sees your screen, understands the app you're in, then clicks, types, codes, researches and writes across your apps until the whole job is done. It runs normal work on its own and stops for a single "yes" only before anything irreversible.

Role
Solo designer & engineer
Platform
macOS · native
Built with
Swift · SwiftUI · AppKit
Codebase
~25k lines · no Electron
Movi — Talk to your Mac. Movi does the work. With its six agent characters.
6specialist agents
24MCP connectors
4,157skills in the library
63dev task recipes
~25klines of Swift
138source files
375phrases routed instantly
~975automated checks

How it works

Every request runs the same loop — and Movi doesn't call a job done until it has checked the result.

01

Speak

Hold ⌃⌥, say "Hey Movi", or tap ⌃⌃ to type. Just ask, in plain words.

02

Look

Reads the screen through the Accessibility tree plus on-device Vision OCR, so it knows exactly where you are.

03

Act

Clicks, types, runs code and drives apps in the background — handing work to the right specialist agent.

04

Verify

Confirms the result really happened. It never says "sent" or "saved" unless a tool actually did it.

→ Autonomy, with judgment Movi finishes normal work end to end on its own. Before anything you can't take back — sending, deleting, paying — it asks once. That's what makes it safe to hand the whole job to.

What it does

Hands-free control

Hold ⌃⌥ or fn⌃ to talk, tap ⌃⌃ to type, or say "Hey Movi" (on-device wake word). Hold silently and it reads your screen and asks what you need.

Computer use

Clicks, types, scrolls and presses shortcuts in any app — in the background, without taking your cursor. Drives browsers, the iOS Simulator and Terminal.

Six agents in parallel

Specialists for browsing, research, writing, code, planning and Mac apps — each in its own chat. Create your own by voice: "make me an agent that…"

Thinks ahead

A daily Suggestions engine studies your projects and recent work and proposes jobs. Say yes and an agent starts.

Developer task library

63 runnable recipes — new Next.js/Expo apps, GitHub, Vercel, Supabase, Stripe, fix a slow site. Installs what's missing, explains failures, verifies the result. Works inside Lovable, Bolt, v0, Replit and Rork.

Learn mode

Ask "what is this?" on any screen — a database table, a dashboard, an error — and Movi numbers the parts on screen and explains each one.

Prompt Coach

Rewrites your prompt in place for 17 AI tools — ChatGPT, Claude, Cursor, Lovable, Midjourney — with one-word undo.

Documents

PDF, Word and Markdown from a sentence. "Plan 3 days in Goa under ₹20,000 and save it as a PDF" — done in about 6 seconds.

Routines

Recurring background jobs — "check this site every 5 hours" — with results delivered to the agent's chat.

4,157 skills

A searchable skill library. Movi picks and loads the right skill for each task automatically.

24 connectors

Gmail, Drive, GitHub, Notion, Linear, Figma and more over MCP — full OAuth lifecycle with refresh, revoke and an audit log.

Memory + on-device AI

Remembers facts, preferences and what worked (imports Claude memory). Greetings, OCR, pointing and voice-to-text run on the Mac at zero cloud cost.

Meet the agents

Six specialists that work in parallel, each in its own chat — Movi hands each piece of a job to whoever does it best.

Scout

Scout

Websites & browsing

Operates real websites — finds, fills and navigates for you.

Sage

Sage

Research

Finds, reads and summarizes sources into a clear answer.

Quill

Quill

Writing

Drafts emails, posts and docs in your voice.

Bolt

Bolt

Code

Explains, writes and debugs code — and runs it.

Atlas

Atlas

Planning

Plans trips, projects and days, down to the schedule.

Pilot

Pilot

Mac apps

Opens apps and works across them like you would.

Need something else? Say "make me an agent that…" and Movi builds a custom one on the spot.

Real screens

// home — your agent team, suggestions and tools
// connectors — give agents access to the apps you use
// product website — HTML/CSS/JS, GSAP scroll animation, Three.js 3D mascot

Built-in intelligence & safety

Smart routingAn instant intent router (375 tested phrases) handles common commands with no AI call; an agent loop takes the complex ones.

Honesty checksMovi never claims "sent" or "saved" unless a tool actually did it.

GuardrailsNever types passwords or card numbers, never pays, never solves CAPTCHAs. Always asks before sending, deleting or connector writes — even in Auto mode.

Privacy modeLooks at the screen only during a request, then forgets. No telemetry.

Tested22 automated test suites (~975 checks), including an end-to-end OAuth lifecycle test.

On-device firstWake word, OCR and voice-to-text run locally with Apple Foundation Models. Cloud models are bring-your-own (Groq, OpenAI, NVIDIA NIM).

Connectors

24 over MCP, plus any custom MCP server — with sign-in, automatic token refresh, reconnect, revoke, a per-connector AI-access switch, a permissions view and an audit log.

GmailGoogle DriveGoogle CalendarGitHubNotionLinearJira / ConfluenceFigmaCanvaAirtableClickUpMondayIntercomWebflowWixVercelSupabaseCloudflareFirecrawlSentry Apple CalendarRemindersNotesMail+ custom MCP

Tech stack

App
SwiftSwiftUIAppKitSparkleCode signing
Perception
Accessibility APIsScreenCaptureKitApple VisionSpeech framework
On-device AI
Foundation ModelsEventKitKeychain
Agents
Tool callingMulti-agent delegationIntent routingPrompt engineering
Integrations
Model Context ProtocolOAuth 2.1 + PKCEDynamic client registration
Website
HTML / CSS / JSGSAPThree.js

My role

Solo designer & engineer

Designed and built Movi end to end from scratch — product, character design, agent architecture, the native Swift app (~25,000 lines across 138 files, no Electron), the connector and OAuth system, the test suites and the marketing website.