A Mac app for individuals and teams.
SightArc watches the way you actually work across apps, files, tabs, docs, and messages, and learns it. Repetitive work becomes an automation you can read, edit, and run with one click. Every model stays on your machine.
Observed 4 weeks in a row · ~92 minutes saved
Runs locally on your Mac. Reviewed by you before anything happens.
For the individual
SightArc watches what you do across your Mac, in Canvas, Drive, Notion, Slack, Mail, Notes, and Terminal, and surfaces the small workflows you didn't realize you had. Sunday course prep. Weekly reporting. The same five-tab research dive every Tuesday. It proposes them as automations you can read in plain script, edit, and run.
“Ninety minutes every Sunday, just to get to the starting line. SightArc gave me my mornings back.”
For the company
Across a team, the same patterns repeat invisibly. Manual reporting in three roles, the same context being rebuilt before every Monday review, the same answers being typed for the fifth time. SightArc builds a live model of how the work actually happens, one person at a time. What's repetitive becomes proposed automation. What's redundant becomes obvious. Onboarding writes itself.
Why it works
Most tools see the org chart. SightArc sees the work. From the bottom up: what each person actually does, which apps, which sequences, how often. Patterns surface. Repetitive work in one role. Redundant work across three. SightArc proposes the automations and shows you which subtrees of the company are quietly carrying the load.
Director
Review team output · Approve plan
Manager X
Repetitive load
IC
Repetitive
IC
Repetitive
IC
Repetitive
Redundancy
Repeated reporting across 3 roles
Suggested automation
Auto-generate weekly report + sync CRM
Manager Z
Steady cadence
IC
Repetitive
IC
Efficient
Data flows up. Insights flow everywhere. Automation scales across the organization.
Local-first
Every model SightArc trains is trained on your machine, from your data, owned by you. Nothing leaves your Mac unless you explicitly send it. There is no SaaS account holding a copy of your work, because there is no SaaS holding your work. SightArc is yours, the way your Mac is yours.
Granular Control
Choose what SightArc watches. Everything else, it never sees.
iMessage. 1Password. Your banking app. Anything you toggle off becomes invisible to SightArc, not redacted, not summarized, not there at all.
SightArc respects the line you draw. Always.
Allowed
Private
Why now
Messages - Anna
Today, 2:47
Anna · 2:47 PM
You · 2:47 PM
Type a message…
SightArc
WatchingPattern detected
You replied to 3 sync confirmations this week.
Tue 9:14·Anna Kim
RepliedWed 11:02·Mike Chen
RepliedToday 2:47·Anna Kim
RepliedProposed automation
Zapier and n8n require you to define the workflow upfront. Real work is messy, personal, and changes weekly. Static rules break the moment context shifts.
With on-device multimodal models now strong enough to run locally on a Mac, the right starting point isn’t a builder UI. It’s observing what the user already does and proposing the automation back to them.
SightArc starts as personal workflow automation. The data it builds, a live, local model of how work happens, becomes the context layer companies have been missing.
Built by
Adam and Jessie built and ran the multimodal LLM platform behind Meta AI on Ray-Ban Meta smart glasses, training, deployment, and inference at consumer scale across 20+ countries and 12+ languages. SightArc points that same operational discipline at the work you do every day on your Mac.
Cofounder
Previously: Tech Lead, Meta AI (Reality Labs Wearables) · Dual MS/MBA, Harvard
Four-plus years as an ML engineer at Meta, most recently leading GenAI Inference for Ray-Ban Meta Smart Glasses. Drove Contextual AI and deployed multimodal foundation LLMs that shipped to the entire user base, millions of devices in active use, across zero-to-one launches and end-to-end systems.
AI lives or dies by the platform, model, and context underneath it. Building the model is one problem; making it run reliably on a device someone wears on their face is another. SightArc is what happens when that platform discipline is aimed at a Mac.
Cofounder
Previously: Engineering Manager, Meta AI (Reality Labs Wearables) · Lecturer, UC Berkeley · ex-Apple
Led the multimodal AI photo inference platform for Meta AI wearables at Reality Labs, the team that owns how multimodal LLMs train, deploy, and serve traffic for Ray-Ban Meta and the rest of the AR wearables lineup. The infra behind Meta AI with Vision, used by millions daily.
Taught design thinking and entrepreneurship at UC Berkeley's College of Engineering (SCET and the Jacobs Institute), and cofounded Close the Gap Foundation. Alumni of his programs, classes, and teams now at Harvard ( JD, GSEd, BA, etc.), Y Combinator, Stanford, UC Berkeley, UCLA, and more. Recent work on robust checkpoint selection for multimodal LLMs, in submission to NeurIPS 2026.
ex-Meta AI · Multimodal LLMs at consumer scale · Inference platform owners · 20+ countries · 12+ languages
The hard part of consumer AI is not the model. It's everything around the model. SightArc is building AI infra for the next decade of growth.
Early access for macOS is opening in waves. Join the list.
We'll only email you when your invite is ready.