nana-wiki 0.0.4

Your sovereign, multilingual AI knowledge assistant for Obsidian

NANA - NAna Not AI

NANA is your assitant for writing better notes and managing your vault.

"I don't cook dinner yet, but I'll help you organize your mind so you have more time to enjoy yours." — Nana

---

What is Nana?

Nana is not just another AI wrapper. She is your digital Babcia (Grandma). While you do the hard work of reading and thinking, Nana handles the tedious bookkeeping of your personal knowledge base.

She works on top of your Obsidian vault — respecting your existing structure, never modifying source material, and always compounding your knowledge over time.

The Philosophy: LLM Wiki, Not RAG

Most AI tools treat your documents as a temporary search index. Nana treats your notes like a persistent, compounding codebase.

• Compounding — The more you add, the smarter the wiki becomes.
• Sovereign — Works locally or with any cloud provider via OpenRouter. Your data stays in your Markdown files.
• Human-Centric — Nana doesn't write your notes for you; she prepares the "ingredients" so you can cook the "meal."

---

Features

Command	Description
clean [path]	🧹 Renames messy files (e.g. Untitled_123.md) intelligently based on their content
extract	📚 Converts raw web clips from /raw into structured, interlinked pages in /wiki
finish [path]	✍️ Finds notes tagged #incomplete and suggests conclusions or missing sections
tag <path> <tag>	🏷️ Adds a tag to all markdown files in a folder that don't already have it
recipe	🍲 Asks Nana for a secret recipe — she's been known to share pierogi

---

Installation

Requirements: Python 3.12+, an Obsidian vault, an OpenRouter API key.

Recommended — pipx (no Python prefix ever needed)

pipx install nana-wiki
nana

One-line installer

curl -sSL https://raw.githubusercontent.com/mlemiec/nana/main/install.sh | bash

From source (dev)

git clone https://github.com/mlemiec/nana.git
cd nana
pip install -e .
nana

What is pipx? It installs Python CLI tools into isolated environments and exposes them globally — like npm install -g but for Python. Install it with brew install pipx or pip install pipx.

---

Configuration

Create a .env file in the project root (or run python src/cli.py to be guided through setup):

NANA_VAULT_PATH=/path/to/your/obsidian/vault
NANA_API_KEY=sk-or-v1-your-openrouter-key
NANA_MODEL_NAME=google/gemma-3-27b-it:free

Nana uses OpenRouter by default, giving you access to hundreds of models — including free tiers.

---

Vault Structure

Nana expects two folders in your Obsidian vault:

YourVault/
├── raw/        ← Drop your web clips and PDFs here (Nana never modifies these)
└── wiki/       ← Nana builds your interlinked knowledge base here
    ├── index.md
    └── log.md

Obsidian Web Clipper tip: Set the default save folder to raw/ in the clipper settings. Then run nana clean to rename and nana extract to process.

---

Usage

Interactive Shell

python src/cli.py
>> clean
>> extract
>> finish /path/to/specific/note.md
>> tag "/path/to/folder" #incomplete
>> recipe
>> exit

Direct Commands

python src/cli.py clean --path /path/to/folder
python src/cli.py extract
python src/cli.py finish --path /path/to/note.md
python src/cli.py tag /path/to/folder "#incomplete"
python src/cli.py recipe

Tip: Use ↑/↓ arrow keys in the interactive shell to cycle through command history.

---

AGENT.md / AGENTS.md — Behavior Instructions

Nana reads a AGENT.md file from the project root and injects it as a system prompt into every model request. This means you can control Nana's personality, formatting rules, and vault conventions without touching any code.

Edit AGENT.md to customise how she processes your notes, what format she uses for wiki pages, and how she handles Obsidian wiki-links vs. citations.

---

Multilingual

Nana speaks your language. Set the locale in src/cli.py:

lang = LocaleManager("pl")  # or "en"

Language	Code	Status
English	en	✅ Complete
Polish	pl	✅ Complete
Other	—	Contributions welcome!

---

Testing

.venv/bin/python -m pytest tests/ -v

17 passed in 0.36s

All AI model calls are mocked — tests run instantly with no API calls.

---

Project Structure

nana/
├── src/
│   ├── actions/
│   │   ├── housekeeping.py   # clean command
│   │   ├── extractor.py      # extract command
│   │   ├── writer.py         # finish command
│   │   ├── tagger.py         # tag command
│   │   └── recipe.py         # recipe command
│   ├── locales/
│   │   ├── en.yaml
│   │   └── pl.yaml
│   ├── cli.py                # Entry point & interactive shell
│   ├── request_handler.py    # OpenRouter API + AGENT.md injection
│   ├── locale_manager.py
│   ├── setup_wizard.py
│   └── utils.py
├── tests/
│   ├── test_housekeeping.py
│   ├── test_tagger.py
│   └── test_writer.py
├── AGENT.md                 # Nana's instruction manual
├── pyproject.toml
└── .env                      # Your secrets (gitignored)

---

Roadmap

• nana index — auto-generate and update wiki/index.md
• nana lint — audit the wiki for orphan pages and contradictions
• nana chat — multi-turn conversation grounded in your wiki
• Local model support via Ollama (Bielik, Llama 3)
• es, de, uk locale additions

---

Contributing

Want to teach Nana a new language or a new skill? PRs are welcome!

1. Fork the repo
2. Add your locale file to src/locales/<code>.yaml
3. Run the test suite
4. Open a pull request with a clear description

---

Made with 💜 and a lot of tea.

v0.0.1 — She's just getting started.