🌙 nocturne.mediagato.com
Local AI assistant for the Mediagato network
🧠 Ollama Models
⚡ FastAPI Streaming
🗂️ SQLite History
🔒 On-Prem Deployment
Nocturne delivers a stylish, fast local assistant with modern UX expectations. Built for Mediagato staff and trusted collaborators, it combines streaming chat, persistent history, and optional web search—all powered by local Ollama models with zero external API dependencies.
🧠 Backend Core
- FastAPI with streaming responses
- Session-based authentication
- SQLite chat history storage
- Optional DuckDuckGo web search
🖥️ Frontend
- SSE-driven streaming UI
- Persistent conversation history
- Model selector dropdown
- One-click message copy
⚙️ Runtime & Infra
- systemd-managed services
- Ubuntu 24.04 LTS on Intel NUC
- 1TB storage for models & history
- CPU-only inference (no GPU required)
🎨 User Experience
- Mediagato dark theme with purple accents
- Glassmorphism UI effects
- Mobile-optimized responsive layout
- Compact history dropdown navigation
Model Inventory
| Model | Size | Primary Role |
|---|---|---|
| qwen2.5:1.5b | 986 MB | Default balanced assistant |
| tinyllama:1.1b | 637 MB | Ultra-fast fallback |
| llama3.2:1b | 1.3 GB | Meta Llama 3.2 compact |
| gemma2:2b | 1.6 GB | Google Gemma v2 |
| qwen2.5-coder:1.5b | 986 MB | Code-focused variant |
| phi3:mini | 2.2 GB | Microsoft Phi-3 |
| deepseek-coder:1.3b-instruct | 776 MB | Lightweight coder |
| deepseek-coder:6.7b-instruct | 3.8 GB | Highest-quality coder |
With Ollama storage relocated to a 1TB volume and systemd supervision for both FastAPI and Ollama, Nocturne maintains rapid local inference without sacrificing retention. All chat history is stored in SQLite, ensuring conversations persist across sessions while keeping data entirely on-premises.