A Nokia phone receiving responses from the EXO AI assistant
So far in this series, we've explored how EXO enables running and training LLMs on your own hardware — no data centers required. One of the most powerful aspects is getting cloud-like AI personalization while keeping your private data local.
Over the holidays, I've been trying to unplug more often. I swapped my iPhone for an old Nokia 3310, but I still wanted a way to do some basic things like asking for my schedule and checking messages.
Let's break down the three key components that make this possible:
Think of this as your local "command center" for AI. It aggregates all the compute you've got on-hand—your CPU, GPU(s), even multiple Macs on the same network—and stitches them together (via pipeline parallel inference) to serve as a powerful LLM backend. You can download the latest version here.
One of the main draws of cloud AI assistants is their ability to integrate with your personal data—emails, calendar, or streaming history—for enhanced personalization. With EXO, you get that same rich personalization without sending your private data to external servers.
We've begun rolling out integrations for Messages, emails, calendar data, and watch histories from YouTube and Netflix. All of this context stays on your system—no third-party cloud services required.
The magic happens through two key mechanisms:
Let's break down exactly what happens when you text your local LLM:
When you send a text from your phone (even from that Nokia 3310!), it travels through your cellular network to your phone. On macOS, this incoming SMS automatically syncs to the Messages app and gets recorded in the SQLite database.
EXO AI agent actively monitors the Messages database. When a new text arrives, it extracts the message content (for example, "Hey EXO, how's my schedule today?") and prepares it for processing.
The EXO Desktop App routes your message to your locally hosted large language model. This process leverages your aggregated compute resources—whether that's your CPU, GPU, or even multiple networked Macs. The model has access to your personal context (calendar, emails, media history) while keeping everything local and private.
Once your local LLM generates a response, the Desktop App uses AppleScript to instruct the Messages app to send an SMS back to your phone. From your perspective, it's just like receiving a text from any other contact—except this one happens to be your personal AI assistant.
This integration demonstrates several key principles that make EXO unique. At its core is privacy-first personalization - your LLM can deeply integrate with your digital life while never leaving your personal compute environment, eliminating the need to share private data with third-party servers. Just as importantly, we've achieved universal accessibility - any device that can handle SMS can interact with your LLM, proving that cloud-based solutions aren't the only path forward for AI assistants. Your AI assistant will soon become your exocortex, owned by you while remaining accessible from any device, anywhere. Of course, SMS won't be the only way you can access your exocortex and we will be rolling out more integrations such as radio (off the grid), email, and more.
The EXO Desktop App has native support for SMS via the Messages app. You can download the latest version here. In the app you can enable the integration by selecting "Enable iMessage integration" in the Menu Bar.
We're at the halfway point of the 12 Days of EXO, and there's still so much more to share. We're excited to share what else we've been working on.
If you try running your AI assistant on an old Nokia phone (or any vintage device), we'd love to hear about it! To inspire you, here's our Day 4 post about how we got an LLM running on Windows 98. You can join our Discord community where we share our retro experiments.