dream-archivist

• 

  Dream Archivist

Inspiration

Dream-Archivist was born from the idea that dreams are part of who we are, and they deserve better than being forgotten. With powerful open-source models like gpt-oss-20B and 120B, we can now build personal AI agents that live locally, keep our data private, and help us explore our inner world.

What it does

Dream-Archivist is a local-first, offline AI system that allows you to:

1. Log dreams via text.
2. Analyze themes, emotions, archetypes using gpt-oss models.
3. Visualize dream patterns over time with semantic mapping.
4. Keep everything local—no cloud, no internet required.

How we built it

Model: gpt-oss-20B and 120B loaded via Ollama and vLLM for local inference. Frontend: Built with React/Electron. Backend: Python + FastAPI. Storage: Encrypted local SQLite database for dream logs and metadata.

Challenges we ran into

Model performance vs. local hardware: Running large models locally required optimization and memory management (especially for gpt-oss-120B). Privacy: Ensuring true offline functionality required careful tool selection (e.g., Ollama vs. APIs).

Accomplishments that we're proud of

Built a fully offline agent that feels personal, intimate, and intelligent.

What we learned

Offline AI is very possible with smart model optimization and UX design.

What's next for dream-archivist

1. Personalized fine-tuning: Enable users to train the model on their unique dream patterns and language over time.
2. Voice-only interface: Build a dream journaling experience usable from bed via local speech-to-text.
3. Dream-to-calendar sync: Explore integrations with personal productivity apps for subconscious goal-tracking.
4. Dream sharing (optional & encrypted): Create opt-in, anonymous dream exchanges for collective dream mapping.
5. Symbol expansion: Train additional models on cultural and historical dream interpretation texts.

Built With

• electron
• javascript
• ollama
• python
• react