no monkeys required
inv is a private AI assistant that runs on your own computer. Download the app, pick a model, and talk to it with no account and no cloud.
Does anything I type leave this computer?
Only web searches, and only while you have the internet switch on. The model, your chats and your files stay on this machine.
What do I need to run you?
A Mac with Apple silicon and at least 16 GB of memory, and about 10 GB of free disk. A Windows version is on the way.
./get-inv --os=mac
models, one is enough to start
cat requirements.txt
Memory is what matters most. The model has to fit in memory together with your conversation, and longer conversations take more: about 0.15 GB per 1,000 tokens of context on top of the model.
| model | computer | download | memory, min | recommended | longest chat |
|---|
macOS: Apple silicon (M1 or newer) and macOS 14 Sonoma or later. Intel Macs are not supported. Windows: Windows 10 or 11, 64-bit; an NVIDIA graphics card with the listed video memory, or a slower CPU-only mode.
cat install.txt
- Download inv for macOS and one model above.
- Open the app zip and drag inv into Applications.
- The first time, right-click inv in Applications and choose Open, then Open again. macOS asks once because the app comes from this site rather than the App Store.
- Double-click the model zip, then move the folder it makes into inv's models folder:
~/Library/Application Support/inv/models/
In inv, open models folder in the model menu takes you straight there. - Pick the model in inv's model menu. The first load takes about a minute.
- Optional, for running C++ code: open Terminal and run
xcode-select --install
sha256sum --check
Each download lists a SHA-256 fingerprint. To confirm a file arrived complete and unchanged, compare it with:
shasum -a 256 ~/Downloads/inv-v0.2-agent-mlx-8bit.zip # macOS certutil -hashfile %USERPROFILE%\Downloads\inv-app.zip SHA256 # Windows
whoami
William Cheriton made inv and this site.
Questions, bugs or ideas: [email protected]
inv is built on K2 Horizon 7B from the Institute of Foundation Models (MBZUAI), released under the Apache 2.0 licence, and fine-tuned independently. It is not an IFM release.