An intimate, local-first, and privacy-focused intelligence for your personal journal. Visualize, explore, and talk to your past.
I have been journaling since I was 10. A few years ago, I digitized everything, creating an archive of over 500 entries spanning my life. I was struck by two episodes of Black Mirror: "The Entire History of You" (a memory archive) and "Be Right Back" (a digital resurrection). I realized I had accidentally created the source material for both.
So, I decided to build it. I named it Smriti—Sanskrit for memory.
This project is evolving rapidly. Features may break, change, or disappear. It is not yet ready for general use, but you are welcome to explore and contribute.
Smriti's stance on privacy is simple: your data is yours. It is designed from the ground up to be a completely private, local-first application.
- 100% Local First: All models, data, and processing happen on your machine. Nothing is ever sent to the cloud.
- Zero Persistence: The application database exists only in temporary memory while the app is running. It is completely destroyed when you shut it down, leaving no trace.
Visualize your emotional history. Red days were rough; green days were better.
Smriti runs a language model locally to answer questions based only on the context from your journal. It's like having a conversation with a digital ghost of who you were.
Me: What do you fear the most?
Discover when you felt a certain way by grouping entries by weekday, hour, or month.
- 💬 Generative Q&A: Ask your journal anything and get answers based on your own words.
- 🔍 Semantic Search: Find memories by meaning and feeling, not just keywords.
- 📊 Emotional Landscape: Visualize your sentiment over time with interactive heatmaps and charts.
- 🔗 Connection Engine: Discover hidden relationships between people, places, and ideas.
- 🏷️ Automated Insights: Automatically discover recurring topics and named entities (
People,Places,Organizations).
Prerequisites: You must have Git, Docker, and Docker-Compose installed. For the demo, you'll also need Python 3.
git clone https://github.com/bvrvl/Smriti.git
cd SmritiTo download the language model, you must agree to Google's Gemma license terms.
- Visit the Gemma 3 4B IT model page and accept the terms.
- Generate a Hugging Face Access Token with
Readpermissions. - Create a file named
.envin the project root and add your token:HF_TOKEN=hf_xxxxxxxxxxxxxxxxxxxxxxxxxxxx
You have two options: use the demo diary or add your own entries.
This repository includes pre-processed entries from The Diary of a Young Girl for a powerful demonstration. To use them, create a data directory and copy the demo files into it.
From your terminal in the project root, run:
# Create the data directory
mkdir data
# Copy the demo entries into it
cp -r anne_frank_diary_entries/* data/This will create a data/ directory filled with the formatted diary entries, ready for Smriti to analyze.
Place your journal entries as .txt or .md files inside a data/ directory in the project root. Smriti will parse the creation date from metadata (e.g., Created: August 11, 2024 7:11 AM) or the filename (e.g., 2024-08-11.md).
With Docker running, execute the following command from the project root:
docker compose up --build- Note: The first build will take a while as it downloads the ~3GB language model. Subsequent builds are much faster.
- Once complete, access the Smriti frontend at http://localhost:5173.
- Backend: FastAPI (Python) with SQLAlchemy
- Frontend: React (TypeScript) with Vite
- Containerization: Docker and Docker-Compose
- LLM Engine:
llama-cpp-pythonrunning Google's Gemma 3 4B-IT (GGUF) - NLP & Embeddings: Sentence-Transformers, spaCy, NLTK, Gensim
The vision is to provide the deepest possible personal insight. Features being explored include:
- Uncovering Internal Contradictions: Automatically identify moments of cognitive dissonance across your journal history to illuminate areas for personal growth.
- Mapping Core Beliefs & Values: Move beyond what you wrote to understand why you wrote it by identifying the persistent, underlying belief systems that guide your reflections.
If you're an engineer or researcher excited by these challenges, your contributions are welcome!
- Gemma 3 Model: This project uses a quantized GGUF version of Gemma 3 provided by @bartowski on Hugging Face, enabling it to run on consumer hardware.
- LLM Backend: As of July 2025, Gemma 3 support is not yet in the main
llama-cpp-pythonlibrary. This project relies on an experimental fork from GitHub user @kossum. The backendDockerfileis configured to use these specific community dependencies.
Contributions, ideas, and feedback are welcome. Please open an issue or submit a pull request.
Led and developed by @bvrvl as part of Kritim Labs, an independent creative technology studio.
This project is licensed under the MIT License. See the LICENSE file for details.