AI that works on your own device is getting more useful
Liquid AI's new LFM3 family offers another path for useful tools. Start by asking which jobs would benefit from working on a phone or computer itself.
Choose where an AI feature should work before choosing how to build it. A weak connection, a quick response, or a clear data boundary can shape the right approach.
AI does not always have to live far away
If you have used AI through a website, it is easy to picture every answer coming from a distant computer. Some AI can also run on the device in front of you. That opens up different choices when you are planning a useful app.
On October 27, Liquid AI introduced LFM3, a family of downloadable AI models designed for focused tasks. The family includes options for text, pictures, and audio, with tools for adapting them to a particular job. Liquid AI says they are built to run on local devices as well as through its online service.
This is a set of building blocks, rather than a finished app you can simply hand to your team. The right option depends on the task, the device, and the results you need.
Read the announcement: Liquid AI: Introducing LFM3.
Begin with the moment someone needs help
Imagine a team member recording a short work update where the connection is unreliable. You might want an app to prepare a draft note on the device, then send an approved version when the connection returns. That is a product idea to test, not a promise that this release delivers it unchanged.
The useful question is where waiting or losing a connection gets in the way. You can answer that from your knowledge of the work, even if someone else handles the programming.
Running AI locally does not automatically make an entire app private or able to work without internet. Other parts may still send information elsewhere. Ask your builder to explain the complete journey in ordinary words.
A simple brief is enough to start
Pick one situation and describe what the person should be able to finish. Use made-up information for an early demonstration.
- Name the job and the device someone will use.
- Write what should still work if the connection disappears.
- Describe what information may leave the device, and when.
- Ask to see the result on the actual device before choosing the approach.
Learn the decision, then get help with the details
You do not need to download a model to benefit from this news. Learning to ask where a feature runs can make you a more confident buyer or builder.
If you want to build it yourself, a guided session can help turn the brief into a first experiment. If you want the tool built for you, ask about device support, setup, ongoing care, and the model's usage terms. Your starting point can stay simple: what should become easier for the person using it?
Read the original reporting.
Our perspective is based on these sources. Product details and availability can change after publication.
Last updated .