Free during the beta

Download memown.

memown runs on the machine you already own. On first launch, it downloads its engine (llama.cpp, from GitHub); you also need a model file (.gguf), for example from Hugging Face. After that, answers are computed on your machine. Pick your platform.

Windowsx64

Windows 11 — verified on September 23, 2026.

Download .zip
Linuxx64

Ubuntu 24.04 or newer (glibc ≥ 2.38), with libgomp1 installed. Verified on Ubuntu 24.04 on September 23, 2026. Does not start on Ubuntu 22.04.

Download .tar.gz
macOSApple Silicon

July 2026 build, signed (Apple Developer ID) and notarized — not re-verified since.

Download .tar.gz
macOSIntel

July 2026 build, signed (Apple Developer ID) and notarized — not re-verified since.

Download .tar.gz
Opening it the first time.
  • Windowsmemown.exe is not code-signed. Chrome may warn you about the download: open Chrome's downloads, click the ⋯ next to it and choose Keep. In our test on September 23, 2026, Microsoft Defender detected no threat in memown.exe; you can also check the SHA‑256 of the .zip below. Unzip, then start it from a terminal (see Start below). If a blue Windows protected your PC box appears, click More info → Run anyway.
  • macOSThe July 2026 macOS build was signed with an Apple Developer ID and notarized by Apple; it has not been re-verified since. Extract the .tar.gz, then start ./memown in Terminal as shown in Start below.
  • LinuxNeeds Ubuntu 24.04 or newer (glibc ≥ 2.38); it does not start on Ubuntu 22.04. First install libgomp1: sudo apt install libgomp1 (without it, the engine does not start and memown may wrongly say the model is too big). Extract the .tar.gz (add chmod +x memown if needed), then start ./memown as shown in Start below. Verified on September 23, 2026, on Ubuntu 24.04.
  • Startmemown needs a model file (.gguf), for example Qwen2.5-0.5B-Instruct-Q4_K_M.gguf (about 400 MB) from Hugging Face, the small model used in our test on September 23, 2026. In a terminal, in the folder where you extracted memown, run .\memown.exe serve --model C:\path\to\model.gguf on Windows, or ./memown serve --model /path/to/model.gguf on Linux and macOS. On first launch it downloads its engine (llama.cpp) from GitHub. Once it is ready, it answers on your machine through an OpenAI-compatible API at http://127.0.0.1:8095/v1.
Verify what you download.

Each archive has a published SHA-256. On your machine run sha256sum (Linux), shasum -a 256 (macOS) or certutil -hashfile memown-1.0.0-windows-x64.zip SHA256 (Windows), then compare it with the list.

SHA256SUMS →

What we measured, and when: see the proof section of the home page