Competing Philosophies About Where AI Should Live - Truly Local vs Cloud VM
Local Doesn't Mean What You Think
"Local-first" has become a marketing term. Some tools use it to mean "we process some things on your device." Others mean "we spin up a cloud VM that acts like a local machine." These are very different things.
True local means native code running directly on your operating system, with direct access to your file system, your apps, and your OS APIs. No VM. No container. No remote desktop. Just your machine doing the work.
The Cloud VM Illusion
Cloud VM agents give you a virtual desktop in the cloud. The agent controls that virtual machine - clicking, typing, installing software. It looks like a local experience when you watch the demo video, but your data is on someone else's server, your latency depends on your internet connection, and the agent can't interact with the apps actually running on your Mac.
Want the agent to drag a file from Finder to a Slack message? It can't - those apps aren't in the VM. The agent lives in an isolated sandbox that looks like a computer but isn't your computer.
Why the Distinction Matters
A truly local agent can use macOS accessibility APIs to interact with any app you have open. It can read the state of your applications, click buttons, fill forms, and navigate menus - all through the same interfaces that assistive technologies use. This works with every app, not just web apps, and it works without sending any data off your machine.
The performance difference is also significant. A local agent responds in milliseconds. A cloud VM agent adds at minimum 50-100ms of network latency to every interaction, and often much more. For an agent that's taking dozens of actions per task, that adds up fast.
Choose Real Local
When evaluating AI agents, ask a simple question: does this run on my actual machine, or on a machine somewhere else? The answer determines your privacy, your speed, and what the agent can actually do for you.
Fazm is an open source macOS AI agent. Open source on GitHub.