Haven is a local AI assistant designed to help teachers with some of the work that usually takes up far too much time: marking, worksheets, lesson planning, and general admin. The main idea behind Haven is simple — schools should be able to use AI without sending student data to the cloud.
Everything runs locally, and Parallax handles the compute across whatever machines are available on the network.
Teachers are often overloaded with tasks that AI could easily help with, but most AI tools can’t be used in schools because they rely on external servers and third-party processing. Haven is meant to show that a fully local, privacy-safe alternative is possible.
Upload a student’s work and Haven produces structured feedback, key strengths and weaknesses, and a score with justification. Conversations stay saved locally.
Give Haven a topic and it creates a clean, ready-to-use worksheet with questions and answers.
Haven generates a full slide deck with simple, organised formatting and properly credited images.
Teachers can keep separate conversations, and each one is stored locally so nothing is lost after refreshing.
A small RAG setup (using Chroma) lets Haven remember useful details to improve responses over time.
Haven shows token usage, most-used features, and how much cost is saved by running everything locally instead of paying for cloud inference.
- The user opens Haven in a browser.
- Parallax detects available devices on the network.
- Each tool (marking, worksheets, planning) runs on local models.
- Outputs are stored on the machine, not on an external server.
- The system stays completely inside the school network.
- Frontend: Next.js, Tailwind
- Backend: FastAPI, SQLAlchemy
- Database / Memory: ChromaDB
- AI Runtime: Parallax for local inference and device coordination