Об этом проекте

HALO Console is a local-first AI workbench designed for users who want full control over their AI interactions. Built with Next.js and TypeScript, it connects to locally running Ollama models to enable streaming chat, document-based retrieval (RAG), and intelligent model routing labeled as Quick, Daily, or Heavy based on task complexity. Users can upload text, markdown, log, or PDF files (with text extraction for readable PDFs) and selectively inject document chunks into conversations. A manual Learning Layer lets users curate personal notes that can be optionally included in prompts. The system includes a read-only Runtime Bridge for HomeLab status queries when configured, but no shell access, file browsing, or cloud APIs are required. All data — chats, documents, and learning notes — remain stored locally under HALO-controlled directories. Security boundaries prevent exposure of paths, secrets, or private content. Web search and OCR are disabled by default. The interface is browser-based, session history is saved in localStorage, and the architecture separates UI, API, and model runtime layers cleanly. This public preview (v0.8.1) focuses on demonstrating local-first design, not deploying private data or services.