Own the model. Own the memory.The AI runs on your machine. After installation, nothing leaves it.
Free during the beta — no account, no sign-up.
In an internal test in June 2026, a version of memown before 1.0 ran a model bigger than the graphics card's memory, by also using the computer's RAM and disk.
After installation, nothing leaves your machine: the AI runs locally. On first launch, memown downloads its engine from GitHub; the model is a file you download yourself, for example from Hugging Face.
Your context is saved in a file on your machine (memown.mem) and reused after memown restarts.
A model of about 120 billion parameters started on a laptop with an 8 GB graphics card and 16 GB of RAM, with a version of memown before 1.0, and was ready to answer. Its engine (llama.cpp), tested alone on the same laptop, answered slowly. Measured in June 2026, internally.
measuredAfter memown restarts, it reuses its saved context, and the answer is identical, byte for byte, to one computed from scratch on the same machine. Measured on September 23, 2026 (Windows 11 and Ubuntu 24.04).
measuredLeave your machine after installation. On first launch, memown downloads its engine from GitHub; the model is a file you download yourself, for example from Hugging Face.
by designmemown is built by ZMLabs, a deep-engineering studio in Sète, in the South of France.
On September 23, 2026, memown ran a small model on a Windows 11 virtual machine with 8 GB of RAM and no graphics card, and gave the same answer after a restart.
memown works without an account: download it and run it.
Preview — accounts aren't open yet.