Download Memrith.
Free, MIT licensed, and no longer maintained. You do not need it to read the study, and nothing here depends on your installing it. It is available because reproducing any of these findings requires it, and because the people who paid for it when it was a product should get to keep it.
Before you install
You will need an API key. Memrith does not include an AI model. You supply a key from Anthropic, OpenAI or OpenRouter and that provider bills you directly for what you use, which is typically a small amount per day for personal use, and you can set a hard spending cap in their dashboard. Alternatively, point it at a model running on your own machine (Ollama, llama.cpp, LM Studio) and pay nothing at all.
Both operating systems will warn you on first launch. The installers are not code-signed, because signing certificates cost money the project no longer has. This is expected, and it is a one-time click-through:
- macOS: right-click
Memrith.app→ Open → Open. On macOS 13 and later you may instead need System Settings → Privacy & Security → Open Anyway. - Windows: SmartScreen shows "Windows protected your PC" → More info → Run anyway.
First run downloads an embedding model. Roughly 128 MB from Hugging Face, once, at a pinned revision with its checksum verified before use. This is what makes search-by-meaning work locally rather than in somebody's cloud, which is question 1. Without it, search falls back to keyword matching and says so.
No warranty, no support
This is an unmaintained student project. It works, and it keeps working offline, but nobody is developing it and problems you report may never be fixed. There is no support address. The source is public, so forking and fixing it yourself is a faster path than asking.