Install MnemosyneC: Windows

Offline-first · No account required · Right-click → Run as Administrator

← Back to homepage · Download


NANO vs FULL Install

MnemosyneC ships one installer that sets up two modes:

ModeWhat it includesWhen to use
NANOqwen2.5:0.5b (bundled, tiny, instant-on)First install, low-RAM machines, quick start
FULLGemma 4 12B (downloaded at runtime via Ollama)Full substrate power, higher accuracy, needs 8 GB+ RAM

Start with NANO. You can upgrade to FULL from inside the app: Settings → AI Engine → Switch to Gemma 4 12B.


Five-Step Install Guide

Open PowerShell as Administrator and run:

irm mnemosynec.org/install | iex

The canonical install script handles cleanup, Ollama + gemma4:12b, MnemosyneC setup, and a Test My Box self-smoke through the substrate mesh.

Manual .exe download (secondary):

→ Download MnemosyneC-Lean-Setup-0.9.41.exe

Before running the .exe manually: verify the SHA-256 hash.

Get-FileHash MnemosyneC-Lean-Setup-0.9.41.exe -Algorithm SHA256

Compare to the hash on the GitHub release page. If it matches: proceed. If it doesn’t: do not run the file, download again from the official link above.

Step 2: SmartScreen Warning

Windows may show a Microsoft Defender SmartScreen notice while this release builds reputation. The installer is digitally signed; before running it, confirm Windows shows the publisher shown on the certificate, and you may compare the file’s SHA-256 with the value on the release page. If Windows shows “Windows protected your PC,” choose More info, confirm the publisher, then Run anyway only if the publisher matches.

To proceed:

  1. Click “More info”
  2. Click “Run anyway” only after the publisher name matches

If you verified the SHA-256 hash in Step 1, you have confirmed the binary is the exact file we published.

Step 3: Run as Administrator

Right-click the installer → “Run as Administrator”

Administrator privileges are required to:

Step 4: Wait for First-Launch Initialization

On first launch, MnemosyneC initializes:

  1. Ollama runtime: the local AI engine (bundled)
  2. Substrate directory: your personal Eblet™ store
  3. NANO model: qwen2.5:0.5b, pulled from Ollama registry on first run

This takes 1–5 minutes on first run depending on your internet speed (model download) and hardware speed (model initialization). After that: instant-on.

You will see a progress indicator. Wait until the UI shows “Ready.”

Step 5: Ask It Something

Type a question in the Ask bar. You will get an answer, local, private, no cloud.

Offline test: disconnect from the internet and ask again. It still works.


Post-Install: Upgrade to FULL

To switch from NANO (qwen2.5:0.5b) to FULL (Gemma 4 12B):

  1. Open MnemosyneC
  2. Settings → AI Engine
  3. Select “Gemma 4 12B (FULL)”
  4. Click “Download & Switch”
  5. Wait for the model download (~8 GB)

Requirements for FULL: 8 GB+ RAM recommended. 16 GB for comfortable operation. The NANO model remains available as fallback.


Binary Integrity

Every MnemosyneC release is:

Your substrate is stored at:

%APPDATA%\MnemosyneC\substrate\eblets.jsonl

This is a plain JSONL file. You can read it, copy it, verify it, and export it at any time.


Troubleshooting

“The application failed to start because its side-by-side configuration is incorrect” → Install Microsoft Visual C++ Redistributable (latest, x64). Available from Microsoft.

“Ollama port already in use” → If you already have Ollama installed, MnemosyneC will use your existing Ollama instance. No conflict. If port 11434 is blocked by firewall, add an exception.

Model download stalls → The NANO model download requires internet access on first launch only. If it stalls: close MnemosyneC, check your connection, relaunch. The download resumes where it left off.

Substrate directory missing after reinstall → Your substrate (%APPDATA%\MnemosyneC\substrate\) is preserved across reinstalls. Uninstall does not delete it. If you want a clean slate: delete the substrate directory manually before reinstalling.


Install Guide · MnemosyneC v0.9.41 · Windows

← Back to homepage · All proofs · Community