Run 44 open-source models entirely on your device. No internet, no accounts, no telemetry. Your conversations never leave the phone.
44 open-source models from 12 research labs
Meta
Google
Microsoft
Qwen
Mistral
Liquid AI
DeepSeek
IBM
NVIDIA
Hugging Face
Stability AI
01.AIEvery prompt and response is processed locally on your iPhone or iPad. No cloud round-trip, no telemetry, no analytics. Your data is encrypted at rest with the iOS Secure Enclave.
Vision-capable models analyze photos, screenshots, and documents on-device. Decode signage abroad, summarize a chart, or extract text from a receipt, without uploading a single byte.
Pick from open-source models that speak 50+ languages. Translate, explain idioms, or learn a new phrase, all running entirely offline at conversation speed.
Swap models any time. Tiny ones for speed, larger ones for reasoning. All open-source, all on-device.
Gemma 3 4B
RecommendedGoogle · 2.5 GB
Llama 3.2 3B
MultilingualMeta · 1.9 GB
Qwen 3 VL 2B
VisionAlibaba · 1.2 GB
LFM2.5 1.2B
NewLiquid AI · 0.7 GB
SmolLM3 3B
FastHugging Face · 1.8 GB
Granite 4.0 Micro
MultilingualIBM · 1.5 GB
Ministral 3B
MultilingualMistral · 1.8 GB
KernelAI vs the cloud-based assistants you already know.
KernelAI bundles optimized open-source models that run on your iPhone's neural engine and CPU. After a one-time download over Wi-Fi, the model lives on your device. Every prompt is processed locally, with no internet round-trip.
Yes. No subscription, no usage caps, no upsells. There are no servers to fund. Every model runs on your hardware.
Your conversations never leave your device. There is no telemetry, no analytics, and no cloud sync. Data is encrypted at rest with iOS Secure Enclave keys.
Completely. Once a model is downloaded, you can use it in airplane mode, underground, or anywhere with no signal.
Anything running iOS 16 or later. Newer devices with more RAM run larger models faster, but smaller models like SmolLM3 and Phi run smoothly on older hardware.
Yes. Each model has its own customizable system prompt and parameters. Tune temperature, persona, and context length to match your workflow.
Still have questions? Get in touch.
Tested it on a flight with no wifi. It just worked. Fast, surprisingly smart, no signing in.
Finally a local-AI app that doesn't feel like a science project. The model picker alone is worth the install.
I keep KernelAI for anything I'd be uncomfortable typing into ChatGPT. Same speed, none of the receipts.
Download KernelAI and run private, offline AI on your iPhone in under a minute.
