🖋️ Private Novel Strategist

Offline Character & Plot Consistency Engine for Ongoing Web Novels

---

📖 Project Purpose

This project was created to help an author maintain consistency in characters, plotlines, and world-building while writing an ongoing online novel.

It acts as:

• 📚 A private story library
• 🧠 A semantic memory engine
• 🔎 A fact-checking assistant
• 🧩 A plot consistency verifier

The goal is to ensure that as new chapters are written, the author can verify:

• Character motivations remain consistent
• Timeline continuity is maintained
• Plot threads are not contradicted
• Previously established facts are preserved

---

🎯 Problem It Solves

In long-running serialized fiction:

• Characters evolve over many chapters
• Minor details are easily forgotten
• Retcons accidentally occur
• Subplots may contradict earlier arcs

This system provides a searchable semantic memory of the entire novel, allowing the author to verify facts instantly instead of manually rereading everything.

Example questions:

• “Has this character ever met the king before?”
• “What was the original reason for the betrayal?”
• “Did I already define this magic rule earlier?”
• “Is this action consistent with Chapter 12?”

---

🧠 How It Works (Architecture Overview)

1. Chapters are stored as .txt files.

2. Text is:

   • Split into chunks
   • Converted into embeddings
   • Stored in a vector database (ChromaDB)

3. When a question is asked:

   • Relevant story fragments are retrieved
   • Injected into the prompt
   • The local LLM (phi3) generates a context-aware response

This creates a lightweight Retrieval-Augmented Generation (RAG) pipeline optimized for fiction tracking.

---

🔒 Fully Offline

• No external APIs
• No cloud dependency
• No data leaves your machine
• Safe for unpublished manuscripts

---

⚙️ Tech Stack

• Streamlit – User Interface
• ChromaDB – Vector Database
• Ollama – Local LLM Runtime
• phi3 – Fast inference model
• nomic-embed-text – Local embedding model

---

🚀 How To Run The Project

Follow these steps carefully.

---

1️⃣ Install Ollama

Download and install from:

👉 https://ollama.com

After installation, pull required models:

ollama pull phi3
ollama pull nomic-embed-text

Verify Ollama is running:

ollama run phi3

Then stop it with Ctrl+C.

Optional (check GPU usage on Apple Silicon):

ollama ps

It should show:

PROCESSOR: gpu

---

2️⃣ Create a Virtual Environment

Inside your project folder:

python3 -m venv venv
source venv/bin/activate

---

3️⃣ Install Dependencies

pip install streamlit chromadb ollama

---

4️⃣ Project Folder Structure

Ensure your structure looks like this:

NovelAssistant/
│
├── app.py
├── story_db/              # Auto-created vector database
├── my_novel_pages/        # Add your .txt chapters here
└── README.md

---

5️⃣ Add Your Chapters

Place your novel chapters as .txt files inside:

my_novel_pages/

Example:

my_novel_pages/
├── chapter1.txt
├── chapter2.txt
├── notes.txt

---

6️⃣ Run The Application

From the project root directory:

streamlit run app.py

Your browser will open automatically.

---

7️⃣ Sync Your Story

Inside the sidebar:

Click:

“Sync New Chapters/Notes”

This will:

• Split chapters into chunks
• Generate embeddings
• Store them in ChromaDB

Once synced, you can start asking questions.

---

🧪 If You Change Embedding Models

If you ever switch embedding models and get:

Embedding function conflict

Delete the vector database:

rm -rf story_db

Then restart and sync again.

---

⚡ Performance Tips (Apple M1 Recommended)

For best performance:

• Use phi3
• Keep chunk size around 600
• Use retrieval k = 2
• Limit output tokens (num_predict = 300)

Expected response time: 1–3 seconds on M1

---

🧩 Future Improvements

• Character profile auto-extraction
• Timeline visualization
• Relationship graph mapping
• Contradiction detection alerts
• Arc-level summarization memory

---

👨‍💻 Built For

Writers of serialized fiction who want:

• Long-term narrative consistency
• A private AI assistant
• Structured plot validation

---

"Write freely. Let the system remember everything."