Introducing Modem AI

When I first began working on my open-source project, Generellem, the idea was that the cloud would be the best approach. For various reasons, that didn't work and I set out on a new task. Moving forward, several factors influenced my thinking, though none more so than privacy. Privacy was hands down the salient theme in all of the feedback I received. So, I decided to address that head-on.

Security and privacy are the driving motivation for creating Modem AI. Achieving this meant that I would have to give people control over their own data and that none of their company/personal data or conversations would never be sent to an external server. Instead of making promises and leaning on my reputation, I simply made it impossible for me or anyone working with me to see anyone's data and conversations.

The next question that I asked myself and am sure that you're wondering about is "What are the tradeoffs?" In software, when you pull in one direction, you often take away from another. In this case, the ability to load a Large Language Model (LLM) via an app faces friction with both app store constraints and a decreased ability for users to consume because of limited resources. So, I went with a Small Language Model (SLM), which was Microsoft's Phi-3. Why didn't you go with Phi-4? It's larger and though it's a better model, I wanted to increase the opportunity for users to be able to load it onto their device. There are implications.

This leads to why the name, Modem AI. Back in the day, before most people were on the Internet, there were these things called Bulletin Board Systems (BBS). A BBS allowed people to connect via phone line and participate in conversations, read news, play games, download files and other such fun. The way that people connected to a BBS was via a Modem, which stands for Modulator/Demodulator, translating between digital computer signals and analog phone signals. To say the least, Modems were very slow. The first parallel here is that modems were slow and you had to set there and watch the BBS load the page, much like what you'll experience with Modem AI because the SLM is slower than an LLM and you're resource constrained. Another point quality, where sometimes your line would drop in the middle of a call with the BBS. SLM quality is not as good as an LLM. Shout out to Jennifer Mayo for coining the name after seeing it in (slow) action. So, Modem AI was born.

Don't take my word - be the judge - download Modem AI and give it a shot. Modem AI also has a feedback form, and it would be amazing to hear from you.

Get Modem AI

Microsoft Store
App Page