Mayo Software
Suggested start
Phi-3 Balanced
Good on most phones and PCs
The default. A steady pace for everyday devices, and the one Modem AI offers first.
Modem AIEmpower Your Privacy with Personal AI
Private. On-device. Worth the wait.
The model runs on your phone, tablet, or computer. Prompts and replies stay there — no account, and no cloud chatbot in the middle.
On your device
Prompts and replies are processed where you are. They are not sent off to be answered in the cloud.
Works offline
The model downloads once, about 2–3 GB. After that, chat does not need a connection.
No account
Nothing to sign up for. Free, with no subscription sitting between you and the model.
Worth the wait
Replies take longer than a cloud chatbot. That pace is the price of keeping the model with you.
A private assistant, said plainly
Calm on purpose. The app tells you what is happening on the device, including when it is taking its time.
Chat that stays put
Conversations live on the device, and you can come back to them. Replies can include lists, tables, and code you can copy. The header shows which model is loaded, and how it is pacing.
Voice that stays local
Dictate into the message field. Recognition runs on the device, and you review the words before they send. If on-device speech is not available, Modem AI says so instead of quietly using the cloud.
A model that fits
Choose in plain language: Balanced, Snappier, or Faster. The app looks at the device and suggests a starting point. You can still pick, and every option runs locally.
Shaped for the device
Phones stay focused on the conversation. iPad and Mac open a sidebar. Windows is keyboard-first, with shortcuts for the things you do often. Same green mark, native where it matters.
Controls in front of you
Light, dark, or match the system. Temperature, top-p, and response length show their values as you move them. Voice can stay off, or send only after you have read the transcript.
A snag stays private
When something fails, the conversation has not left the device. You can try again, or send an anonymous report. That report does not include the chat unless you choose to add it.
Choose the pace
Three on-device options, each about 2–3 GB. Privacy is the same on every one of them. Faster, on Windows, uses a capable GPU when you have one and falls back to the CPU when you do not.
Quicker
Phi-3 Snappier
Quicker replies when you are in a hurry
The same private, on-device idea, tuned for a shorter wait on the CPU.
Windows GPU
Phi-3 Faster
Faster on Windows PCs with a capable GPU
Optional. Mac, iPhone, iPad, and Android stay on the CPU. Windows uses the CPU when the GPU is not a fit.
The same assistant, on each device
Green where the brand is, native where the platform already has a good answer.
Chat
Ask in your own words. The reply stays on the machine, and the screen keeps the model name where you can see it.
Settings
Voice stays on the device. Theme, temperature, and response length are labeled while you change them.
iPad and Mac
A sidebar for Home, Chat, Settings, Feedback, and About. The phone keeps that list behind a menu so the chat can have the screen.
About
The promise is in the app too: on-device, about 2–3 GB, and slower than the cloud so the conversation can stay yours.
The tradeoff, said plainly
Running on the device means replies take longer than a fast cloud service, and they can be less polished. That is the exchange. You get a conversation that is not sitting on someone else's server.
What we collect
Anonymous usage statistics, such as performance, so the app can improve. Not your prompts, and not your replies, unless you explicitly choose to share something with us. Voice audio is recognized on the device.
Questions worth asking
The short version of how Modem AI treats your conversations.
No one else. They stay on your device. We do not receive the questions you ask or the answers you get, unless you explicitly choose to share something with us.
The model runs on the device, including when you are offline. We collect anonymous usage statistics, such as performance. We do not collect personally identifiable information, and we do not collect the conversation itself. Speech you dictate is recognized on the device.
Because the model is on your device, not in a data center. That wait is the price of the privacy. Phi-3 Snappier shortens it on the CPU. On a Windows PC with a capable GPU, Phi-3 Faster can go further, and it falls back to the CPU when the GPU is not a fit.
Start with the suggestion. Phi-3 Balanced fits most phones and PCs. Phi-3 Snappier aims for quicker replies. Phi-3 Faster is the Windows GPU option. Each download is about 2–3 GB and happens once. You can switch later, and every model still runs on the device.
No. Dictation uses on-device speech recognition. By default you read the transcript before it sends. You can turn on auto-send in Settings if you want the words to go as soon as you stop listening. If on-device recognition is not available, Modem AI tells you. It does not fall back to a cloud speech service.
Before the Web, people reached bulletin board systems over a phone line. A modem translated between the analog line and a digital computer. Modem is short for modulator-demodulator. Those connections were slow compared with what came later, and they got faster as the hardware improved. Modem AI borrows the name, and the patience.
Download Modem AI
A private conversation, on the device you already have.
By downloading, you agree to our Terms of Use and Privacy Policy.