Offline AI Tools

An offline transcription tool needs a speech model. A translator needs the languages you plan to use. Choose a tool by what you want to make or edit, then check which files to bring.

Before a trip, open the app with Wi-Fi and mobile data turned off. Try a recording, a paragraph or a photo from the device you’ll take.

Last updated:

A tool for each task

Use the local option in each app. Online extras have separate requirements.

Tools, offline tasks, and what to prepare
ToolYour taskWhere it fitsDownload before you go
LociOur everyday writing pickDraft a reply or summarize an attached document on your device. Local chat is free.iPhone, iPad, Android, Mac, WindowsThe recommended local chat model. Web search needs a connection.
MacWhisperAudio transcriptionSaved audio on a MacA local speech model and your audio files
Draw ThingsImage generationiPhone, iPad, MacA compatible image model
Argos TranslateText translationPython, command-line, or desktop GUI workflowsThe language packages you need
OllamaCoding help through a local modelDesktop, local API, and scriptingA local model and any editor integration
llama.cppLocal inference with file-level controlDeveloper CLI and server workflowsA compatible model and the runtime

Chat
and write

Bring a draft you can check

Download a chat model, open a fresh conversation and paste in a paragraph from your draft. Ask for a shorter version that keeps the names and dates. Compare the two before you keep the edit.

For work on a train or plane, choose a model stored on the device in your hand. A chat session that uses a paired computer depends on reaching that computer. Save any reference material you’ll need before you leave.

Transcribe
audio

MacWhisper for saved recordings

Bring an interview, voice memo, or lecture recording to MacWhisper. Its local speech models turn audio files into text on your Mac. Download the model and keep a local copy of the recording before you disconnect.

Select local transcription. The separate cloud transcription feature uploads audio for server processing. Summaries that use an online provider also need internet access.

For a command-line workflow on a computer, Whisper provides open-source speech recognition. You will need its dependencies and model weights. Review the transcript against the recording, especially for names and numbers.

Make
images

Draw Things for local generation

Draw Things runs image models on iPhone, iPad, and Mac. Choose a model your device can handle, download it, then use local compute to generate from your prompt.

Allow space for model files as well as the images you save. Draw Things also offers cloud compute, which needs a connection. A tool’s platform support does not mean every model will fit every device.

Translate
text

Argos Translate for installed language pairs

Argos Translate is an open-source offline translation tool with a Python library, command line, and graphical interface. Install the language packages you plan to use.

Check that you have the pair in the right direction. Argos can translate through an intermediate language when you have the necessary packages, but that can reduce quality. Use a direct pair where available and review important text with someone who knows the language.

Work
with code

An API for an editor, a runtime for model files

Use Ollama when you want a local model server for an editor or script. Download a local model, then point your tool at the local endpoint. Ollama’s documented local-only mode disables its cloud models and web search.

Choose llama.cpp when you want to manage model files and inference settings at the command line. Its CLI and HTTP server support a range of computer hardware.

A local coding model can explain a function or propose a patch. Package installs, remote repositories, and online tools still need a network. Run the code and its tests before relying on a suggestion.

Pack the files, then disconnect

  1. Install and open the app. Finish any runtime download while you have a connection.
  2. Download the right model. Check device support and available storage.
  3. Save your input files. Keep a copy of each recording or document on your device.
  4. Try a fresh task offline. Turn off Wi-Fi and mobile data, reopen the app, and try a new prompt.

Read the guide to offline model files before choosing a download. For a device-first comparison, see offline AI apps by platform.

Questions and answers

Which AI tasks can I do offline?

With the right downloaded model, you can write and chat, transcribe saved audio, generate images, translate text, or ask for coding help. Choose a tool built for your task. Web search and cloud models need a connection.

What should I download before I disconnect?

Install the app, its model, and any extra runtime or language pack. Save the files you want to work on. Open the tool once while connected, then test a fresh task with Wi-Fi and mobile data off.

Does MacWhisper work offline?

MacWhisper can transcribe files on a Mac with a downloaded local model. Select local transcription. Its optional cloud transcription sends audio to a server, and online AI providers also need a connection.

Can I generate images without internet?

Draw Things supports local image generation on iPhone, iPad, and Mac. Download a compatible model and choose local compute. Cloud compute and new model downloads need internet access.

Is offline translation the same as an offline chatbot?

No. Argos Translate uses installed language-pair packages for text translation. A local chatbot uses a general language model. Check the languages you need before leaving, and review translations where mistakes matter.

Can a coding assistant work without a connection?

A local model can explain code or suggest changes offline. Ollama and llama.cpp can run that model on your computer. An editor may still need the network for other providers, web tools, dependencies, or remote repositories.

Does local AI guarantee that no data ever leaves my device?

Local describes where the model runs. A tool may also have cloud features, online sources, or sharing options. Choose the local workflow and check the settings for the task you plan to do.

How we chose

We used each maker’s documentation to match one local workflow to a task. We looked for the files you need before disconnecting, the platforms that can run them and which features still need a server. The checklist gives you a way to test your own setup.