Offline AI lists usually mix two kinds of tools: apps that run the AI on your own computer, and cloud apps that only keep working for a few minutes when the Wi-Fi drops. I use two of the popular picks myself, and one of them isn't offline at all. Below are 10 tools that keep working with no connection at all, and how to find out which ones your computer can run.
⚡ Short answer: Yes, AI works without internet. Download a free app like LM Studio, Ollama or Jan, download a model while you're online, and it runs entirely on your device afterwards. A laptop with 16 GB of RAM runs a capable chat model. For transcription use MacWhisper or Buzz; for images, Draw Things; on phones, Google AI Edge Gallery.
🆕 What changed in this update (October 14, 2026)
- Moved Descript out of the offline picks: it's cloud-first, even though I use it daily.
- Added a hardware picker, phone apps and Draw Things.
- Updated models to Gemma 4, Qwen 3.6 and gpt-oss.
- Flagged that GPT4All hasn't shipped a release since early 2025.
Some links are affiliate links · they support this site at no extra cost to you · Full disclosure
Can AI work without internet?
Short answer: Yes. Offline AI runs a model file stored on your own computer or phone, so nothing is sent to a server. You need internet once, to download the app and the model; after that it works on a plane, in a cabin or with Wi-Fi switched off. The trade-off is speed and size: local models are smaller than ChatGPT and know nothing newer than their training data.
Three things make a tool offline AI in the real sense. The model runs on your processor or graphics card, not a data center. Your prompts, files and recordings never leave the device. And it keeps working with networking switched off, not just for a few minutes until a sync fails.
That's also why offline AI matters for privacy, not only travel. A local model can't send your client notes, medical questions or unpublished work anywhere, because it has nowhere to send them.
Key idea: Test any "offline" tool the same way: download what it needs, turn off Wi-Fi, restart the app and try the job you care about. If it complains, it isn't offline.
Which offline AI can your computer run?
Short answer: Memory decides it. With 8 GB of RAM you can run small models for drafting and summaries; 16 GB runs a capable chat model like gpt-oss-20b; 32 GB or more runs models close to cloud quality. Apple Silicon Macs do well because the processor and graphics share the same memory.
Pick your memory and what you want to do. The picker suggests an app, a model to download first, and what to expect.
| Your memory | Model to start with | Good for |
|---|---|---|
| 8 GB | Gemma 4 E4B, Qwen 3.5 4B | Drafts, summaries, quick questions |
| 16 GB | gpt-oss-20b, Gemma 4 12B | Everyday writing, research notes, light coding |
| 32 GB | Qwen 3.6 27B (4-bit) | Long documents, coding, near-cloud quality |
| 64 GB+ | Qwen 3.6 27B at higher precision, larger models | Heavy use, long context, several models at once |
These are starting points from BenchLM's October local-model rankings and each model maker's memory guidance. Real use depends on context length and the app, so open your system monitor the first time you load a model.
Best offline AI for chat and writing
Short answer: LM Studio is the easiest place to start: a free desktop app for Windows, Mac and Linux with a built-in model search. Ollama suits people who want a local AI server for other apps, and Jan is the open-source pick with a clean chat window. GPT4All still works, but it hasn't shipped a release since early 2025.
Search, download and chat with open models in one window. It shows whether a model fits your memory before you download it, which saves a lot of trial and error. It's free for personal and work use.
Ollama runs models in the background and exposes a local API that other apps can use, including Obsidian plugins and code editors. It now has a desktop chat app too, so you don't need the terminal for basic use.
A tidy ChatGPT-style window for local models. Jan can also connect to cloud providers, so check the model picker: a chat only stays local while a local model is selected.
Built to run on ordinary processors, with a LocalDocs feature that answers questions from your own files. The catch: its last tagged release on GitHub is v3.10.0 from February 2025, so newer models arrive late or not at all.
All four use the same family of open models, so the choice is about the interface. If you write in Obsidian, connect it to Ollama with a local-LLM plugin and your notes app becomes a private assistant too. For background on the models themselves, see my guide to open-source AI models.
Best offline AI for transcription
Short answer: MacWhisper on a Mac, Buzz on Windows or Linux. Both run OpenAI's Whisper speech model on your own computer, so interviews and meetings never leave your device. MacWhisper has a free tier and a €59 (about $69) one-time Pro license; Buzz is free.
Drop in a recording and get a transcript with timestamps, speaker labels and subtitle export. Everything runs on your Mac once the model is downloaded.
A simple desktop app around Whisper with SRT subtitle export. Slower than MacWhisper on older laptops, but free and cross-platform.
Developers can use whisper.cpp directly from the command line; it's the engine many of these apps are built on, and it runs on a plain processor.
Best offline AI for images
Short answer: Draw Things on a Mac, iPhone or iPad with Apple Silicon: it's free and runs image models like FLUX and Stable Diffusion on the device. DiffusionBee is the simpler Mac alternative. On Windows, offline image generation needs an NVIDIA graphics card with at least 8 GB of video memory.
Text-to-image, editing and upscaling on your own device, with a large model library you download once.
One-click Stable Diffusion for Apple Silicon Macs. Fewer options than Draw Things, but less to learn.
Offline AI video editing: CapCut Desktop
Short answer: CapCut Desktop's core editor works offline: trimming, cutting, text, filters, basic audio and exporting media already on your computer. Its AI effects, templates and cloud sync need a connection. Download what you need before you go offline.
The most practical free editor for social video. Plan around the split: edit offline, then reconnect for auto captions, AI effects and uploads.
🧪 My experience: I've edited real projects in CapCut Desktop with no connection. Cutting, text and export worked; anything with an AI badge or a template asked me to reconnect.
| Feature | Works offline? |
|---|---|
| Trim, split, text, filters, basic audio | ✅ Yes |
| Export of local media | ✅ Usually (fails if the project uses cloud assets) |
| Templates and stock assets | ❌ Need internet |
| AI effects and AI tools | ❌ Need internet |
| Sharing to TikTok or YouTube | ❌ Need internet (file still saves locally) |
How to run offline AI on your phone
Short answer: Install Google AI Edge Gallery (Android and iPhone, free) or PocketPal AI (free, open source), download a small model over Wi-Fi, then switch to airplane mode. Phones with 8 GB of RAM handle 3–4B models well; 6 GB phones should stay with the smallest ones.
Google's own app for running Gemma models on-device: chat, image questions and audio. Gemma 4 E2B needs about 4 GB of RAM; E4B needs about 6 GB and an 8 GB phone.
PocketPal AI is the flexible alternative: it loads almost any small open model from Hugging Face, including Qwen, Llama and Gemma. On iPhone, the phone itself needs 2–3 GB for iOS, so the rest is what your model can use. Expect slower answers than on a laptop and more battery drain.
Tools that look offline but aren't
Short answer: Descript, the ChatGPT and Claude desktop apps, and most "AI" features in Notion, Canva or Grammarly need a connection. A desktop app isn't the same as offline AI: if the model runs in a data center, the app stops answering when the Wi-Fi does.
| Tool | What people assume | What happens offline |
|---|---|---|
| Descript | Desktop editor, so it works offline | Cloud-first: transcription and sync need internet, and users report offline edits lost on sync |
| ChatGPT / Claude desktop apps | Installed apps run locally | No answers without a connection: the models run on remote servers |
| Notion AI, Canva, Grammarly | Built into apps I already have | The AI features stop; some basic editing may still work |
🧪 My experience: I use Descript every week for transcripts and edits, and I count on a connection for it. That's why it's no longer in my offline picks, even though it was in earlier versions of this post. If you work on planes, transcribe with MacWhisper or Buzz first and bring the text into Descript later. My Descript review covers what it does well online.
Offline AI tools compared
| Tool | Best for | Platforms | Cost | Fully offline? |
|---|---|---|---|---|
| LM Studio | Chat, writing, coding | Win · Mac · Linux | Free | ✅ |
| Ollama | Local AI server for other apps | Win · Mac · Linux | Free | ✅ |
| Jan | Open-source chat | Win · Mac · Linux | Free | ✅ (local models) |
| GPT4All | Older laptops, LocalDocs | Win · Mac · Linux | Free | ✅ (no release since Feb 2025) |
| MacWhisper | Transcription | Mac | Free / €59 Pro | ✅ |
| Buzz | Transcription | Win · Mac · Linux | Free | ✅ |
| Draw Things | Images | Mac · iPhone · iPad | Free | ✅ |
| DiffusionBee | Images | Mac | Free | ✅ |
| CapCut Desktop | Video editing | Win · Mac | Free + add-ons | 🟡 Core editing only |
| Google AI Edge Gallery | Phone chat | Android · iPhone | Free | ✅ |
What are the downsides of offline AI?
Short answer: Four trade-offs: models take 2–20 GB of disk each, they need a lot of memory while running, they know nothing newer than their training data, and there's no sync between devices. Local models are also less capable than the biggest cloud models on hard reasoning tasks.
- Storage: a small phone model is about 2 GB; a 27B laptop model is 15 GB or more. Keep only the one or two you use.
- Memory and battery: running a model uses most of your RAM and drains a laptop battery fast. Plug in for long sessions.
- No live information: offline AI can't check today's prices or news. Use it for writing, thinking and your own files.
- No sync or sharing: chats stay on one device. That's the privacy upside, and the collaboration downside.
Set up offline AI in 15 minutes
- 1Install LM Studio from lmstudio.ai while you're online.
- 2Pick a model that fits. Use the picker above, search for it in LM Studio and check the green "fits your memory" label before downloading.
- 3Download it over Wi-Fi. 5–15 GB is normal, so don't do it on a hotel connection.
- 4Turn off Wi-Fi and test. Restart the app, open a new chat and ask something. If it answers, you're set.
- 5Add the extras you need: a Whisper model in MacWhisper or Buzz for transcription, Draw Things for images, AI Edge Gallery on your phone.
Which offline AI should you pick?
- Choose LM Studio if you want one free app for offline AI chat and writing on a laptop with 16 GB of RAM or more.
- Choose Ollama if you want other apps, like Obsidian or a code editor, to use your local model.
- Choose MacWhisper or Buzz if your offline job is turning recordings into text.
- Choose Draw Things for images on Apple devices, and CapCut Desktop for video edits you finish before reconnecting.
- Choose Google AI Edge Gallery if your only device on the trip is your phone.
Frequently Asked Questions: Offline AI
🔐 Keep building your private AI setup
Offline AI covers the tools you choose; these cover the AI that comes switched on.