One simple setup.
Tap Set up free AI. MyAi handles the choice and checks the download.
MyAi handles it
Made for
your device.
You don’t need to know which AI to choose. MyAi checks your device and picks a suitable free option.
Free to use.
No account or subscription for local chat. You use your own device’s storage, memory, and battery.
Your device. Your pace.
More capable devices can handle larger tasks. MyAi checks available resources before running.
Advanced information and model sources
A starting point
General guidance only. This website does not read your hardware.
A lighter model is a good start.
Small quantized models leave more room for conversations and other apps. Begin with a short context and increase it only after a successful load.
What matters
A little context
before the download.
File size is only the beginning. A model also needs working memory, space for your conversation, and a compatible inference engine.
Leave room to work.
Available RAM matters more than the number printed on the box. GPU memory can help, but it is not interchangeable with system RAM.
Longer uses more.
A longer conversation can need more memory and processing time. Start with a modest context window and increase it carefully.
Choose for the task.
Small models can be useful and fast, but they can make mistakes. Coding, vision, and tools require support in both the model and runtime.
Model sources and licenses
Model families under evaluation
Open choices.
Individual terms.
These links lead to the model publishers. They are research references, not a claim that every model is supported in MyAi.
Qwen
ResearchText, coding, reasoning, and vision models across a range of sizes. Check the exact model card: licenses and capabilities vary by release.
View publisher informationGemma
ResearchOpen-weight models, including smaller options intended for efficient devices. Gemma terms are separate from the MyAi software license.
View publisher informationgpt-oss
ResearchOpen-weight reasoning models. Memory, quantization, and tool behavior need to be checked for the exact runtime and model.
View publisher informationGranite
ResearchIBM’s open model family, with task-specific options. Verify the exact model’s license and context requirements before selecting it.
View publisher informationDeepSeek
ResearchOpen-weight releases with different architectures and practical hardware needs. A family name is not a device compatibility guarantee.
View publisher informationThe application’s license covers its code. Model weights have their own licenses. MyAi keeps that distinction visible.