destabilization.net/projects/oracle-vii
Betav10·Updated June 2, 2026

Oracle VII

PythonFlaskAIOllamaLocal LLM

A local AI terminal built on Flask with a retro CRT browser interface. Runs entirely on your machine — no API keys, no cloud, no data sent anywhere. Supports Ollama models with tool-calling, live web search via DuckDuckGo, a persistent SQLite knowledge base, streaming token-by-token output, voice input and TTS, and a full multi-chat sidebar.

Latest Release
Oracle VII v10
↓ Download
SHA-256
65cb9eadaa56…96a7582b9d51
Features
Retro CRT browser interface at localhost:5000
Supports Ollama — no API key, nothing leaves your machine
Tool-calling loop: AI decides when to search, recall memory, or fetch URLs
Live web search via DuckDuckGo — no API key required
Persistent SQLite knowledge base — AI remembers facts across sessions
Streaming token-by-token output with markdown and syntax highlighting
Voice input (Web Speech Recognition) and text-to-speech (sentence-by-sentence)
Persistent multi-chat with sidebar — all chats saved in browser localStorage
Settings panel: system prompt, temperature, voice, rate, pitch
Export current chat as Markdown
Keyboard shortcuts: Enter send, Shift+Enter newline, Ctrl+N new chat, F1 help
Fully offline after setup — no internet required beyond optional web search
Requirements
Windows 10 or 11, 64-bit
Ollama installed and running
At least one model pulled
e.g. ollama pull llama3
A modern browser to access localhost:5000
Known issues
!
Tool-calling requires a model that supports it — Llama 3, Mistral, and Phi-3 work well.
!
If the interface shows a connection error, check Ollama is running: ollama serve
!
Windows Defender may flag the exe — standard PyInstaller false positive. Scan on VirusTotal to verify.
Changelog
v10
  • Latest release
  • Improved tool-calling loop and model compatibility
  • Graceful fallback for models without tool support
  • Bug fixes and stability improvements
v9
  • Persistent multi-chat with sidebar
  • DuckDuckGo web search integration
  • SQLite knowledge base
v4
  • Initial public version
  • Streaming markdown rendering
← back to projectsfaqcontact