Portable, Persistent AI Memory.
"Your Ghost travels with you. The Shell is just borrowed." — inspired by Project 2501, Ghost in the Shell (Masamune Shirow)
Your AI memory dies every time you switch models. 2501 fixes that.
ChatGPT doesn't remember you when you move to Claude. Claude doesn't remember you when you move to Llama. Every conversation starts from zero. Your memory belongs to their servers, not to you.
2501 is different. Your memory — your Ghost — lives encrypted in a folder you control, or on a USB stick in your pocket. Plug it into any machine, and your AI is there. Switch models whenever you want. The Ghost stays.
your Ghost (encrypted USB) + LLM (Local/Cloud) = your personal AI
The Ghost is yours. The machine is just a Shell you borrow.
- Portable USB Mode — Auto-deploy the entire system to a USB stick. Works on any machine without installation.
- Cross-Platform — Native launchers for Linux (
run.sh) and Windows (run.bat), plus direct bootstrapping frompython3 2501.py. - Local venv bootstrap —
2501.pycreates a repository-localvenv/on first run and installsrequirements.txtautomatically. - Portable Dependencies — USB mode still supports a local
libsfolder for portable installs on FAT32/exFAT sticks. - Multi-Provider LLM — Use local Ollama (suggested, very good even with 4b or 9b models :) ), OpenAI, Google Gemini, Anthropic Claude, and also ReiLab API ! (reilabs.org).
- Secure Key Vault — Your API keys are stored encrypted inside your Ghost. Only you can unlock them.
- Hierarchical LLM Wiki — Obsidian-style tree view. Automatically groups memories into categories (
project,knowledge,user). - Multimodal — Send PDFs, images, and documents for analysis.
- Wake/Sleep Cycle — Automatic memory consolidation after 45 seconds of conversation pause.
# Clone the repository
git clone https://github.com/lordpba/2501-deepmemory
cd 2501-deepmemory
# Launch directly (Linux / macOS)
python3 2501.py
# Or launch via the Linux wrapper
bash run.sh
# Launch on Windows
run.batRun the script and choose "Yes" when asked to deploy to USB. 2501 will detect your USB stick, copy all files, and set up a portable environment.
Once deployed, just plug the stick into any computer and run:
- Linux:
bash run.sh - Windows: Double-click
run.bat
If you use your Ghost on the go via a USB stick, its memories will evolve independently. To synchronize the newly learned memories from the USB back to your primary Desktop Ghost, run:
python 2501.py --sync-from-usbNote: For security, you will be asked for the Desktop Ghost's password before the sync can overwrite your local data.
You can configure your LLM by clicking the Settings (gear icon) in the top right of the UI:
- Ollama: Specify a local or LAN endpoint (e.g., a DGX server).
- Cloud APIs: Select OpenAI, Gemini, Claude, or ReiLab and enter your API Key.
- Ghost Security: All settings and memories are Fernet-encrypted (AES-128-CBC + HMAC) using your Ghost password.
For local LLM support, download and install Ollama from ollama.com. 2501 includes automatic Ollama detection and service startup on most systems. You can also specify remote LAN endpoints (e.g., a DGX server) in the settings.
Your Ghost is a set of encrypted markdown files. Under the encryption, it's plain text — readable by humans, editable, and compatible with Obsidian.
ghost/
├── identity/ ← encrypted identity, config & API keys
├── wiki/ ← hierarchical memories (Obsidian-style)
│ ├── project-x.md
│ ├── knowledge-y.md
│ └── index.md
└── sessions/ ← full conversation logs
Most AI products put the intelligence in the model. 2501 puts the intelligence in the memory.
A small local model with a rich Ghost outperforms a large cloud model with no memory of you. The Ghost is the product. The LLM is just the voice.
MIT