v1.0 - SovNode

SovNode v1.0 — Sovereign AI Node

First public release. A private desktop AI workstation (PyQt6) that runs local models via Ollama, with hybrid memory, live workspace indexing, and sandboxed file/shell tools.

Highlights

• Local-first, with optional Cloud: 100% local inference via Ollama by default, with the option to connect a Cloud engine (Gemini or Anthropic) when you need more horsepower.
• Instant intent routing: a lightweight model classifies every message in milliseconds and picks between a fast conversational path and a deeper reasoning path.
• Hybrid memory & RAG: FAISS + SQLite FTS5 over your conversation history, cached web knowledge, and indexed files.
• Live workspace indexing: add project folders and SovNode re-indexes them automatically in the background.
• Sandboxed tool engine: file read/write and shell execution, all boxed inside your active workspace folder.
• Sandboxed code execution: LLM-generated Python runs through an AST allow-list in an isolated subprocess.
• Anti-hallucination verification: a post-hoc pass checks claims against retrieved evidence.
• Voice & LaTeX: local speech-to-text/text-to-speech, plus inline LaTeX rendering.
• Durable memory (WAL): every turn is fsync'd, exportable as a DPO/SFT training dataset.
• Bilingual: interface and assistant responses in Spanish and English.

Installation

Download SovNode.zip from this release, extract it, and run SovNode.exe — no Python install required. Or run from source (see README).