In ProgressAI2026

Project O.L.I.V.I.A.

AI companion with persistent memory, voice, and dreaming

Role

Solo Developer & Researcher

Duration

Ongoing

Tech Stack

6 technologies

About This Project

Project O.L.I.V.I.A. started as a question: what would it take to build an AI companion that actually remembers you? Not just within a single conversation, but across weeks and months of interaction. The result is an experimental system that pushes the boundaries of personal AI by combining persistent memory, natural voice, and a unique "dreaming" mechanism.

At its core, O.L.I.V.I.A. uses vector embeddings to store and retrieve memories from past conversations. But raw retrieval isn't enough, so the "dreaming" system periodically consolidates memories, finding patterns and connections that inform future interactions. This gives the AI a sense of continuity that feels qualitatively different from typical chatbots. The personality layer sits on top, evolving subtly based on interaction patterns while maintaining a coherent identity.

The technical stack is built for experimentation, and it runs entirely on local hardware. There are no cloud APIs anywhere in the pipeline. FastAPI orchestrates the whole system around a fine-tuned 8B LLM served through Ollama. faster-whisper handles speech-to-text and ChatterBox Turbo handles speech synthesis, so voice never leaves the machine either. Long-term memory lives in a tiered ChromaDB store, with the dreaming process consolidating and reorganizing those tiers over time. The project remains actively developed, with the core architecture open-sourced to gather community feedback and contributions.