Ambient

I'm building Ambient: a way to tell your Mac what you want in plain English and have it actually do the multi-step work — across your apps and files. The catch that makes it different: it runs entirely on-device. Nothing leaves your Mac. No cloud, no account required to run it. Shortcuts/Automator make you pre-define every step. Cloud AI agents make you ship your data off the machine. Ambient is intent-driven AND local. It's pre-MVP — I'm building it in the open. Follow along, and if "nothing leaves my Mac" is a thing you care about, the waitlist is here: ambient.aihum4n.com

Milestone: clipboard + app-switch observation pipeline is live end to end. Pattern detection next. Then the one-click deploy. Building toward a demo where an automation gets discovered and deployed in under 10 seconds — target: ProductHunt/Show HN.

4/ First integration ships before the macOS MVP demo. More soon.

3/ Big tech won't build this — it's too niche for them, and it needs local-first trust to work at all.

2/ Not another "connect your API key and hope" setup. Ambient watches you use these tools, then offers to automate what it sees.

Native integrations coming for Notion, Linear and Raycast — the tools indie hackers actually live in. 🧵1/4

Unpopular opinion: your workflow data shouldn't need a server to be useful. Ambient's observation engine runs entirely on-device. If a company can't tell you where your data lives, that's the product, not a bug.

Exactly — and it flips the usual tradeoff. Normally "more capable agent" means "more of your data in someone's cloud." On-device forces us to make the model do more with less context leaving the machine. Harder to build, but the privacy isn't a feature bolted on — it's structural.

@ambient_agent On-device is the interesting constraint here. Privacy becomes part of the product.

I'm starting to build in public. The project: Ambient — an AI that automates your Mac from plain-English intent, and runs 100% on-device. Nothing leaves your machine. Here's why I think this needs to exist now 🧵

If a private, intent-driven Mac assistant is something you'd use, two asks: Follow along — the real demos are coming Join the waitlist: ambient.aihum4n.com I'll reply to everyone. What would you automate first?

Honest status: this is pre-MVP. I'm building the core loop now. I'll post the messy parts too — architecture decisions, what local models can/can't do yet, where it breaks. Building in public means the failures are public.