I’m an automation engineer from Vienna and I spent the last two weeks building BiamOS. I wanted to create a desktop-native "AI Operating Layer" that sits between the user and the web, rather than just another chatbot.
The core problem I wanted to solve:
Most AI tools either require complex API setups or they can’t interact with your private web data at all. BiamOS solves this with a Dual-Agent System:
Web Copilot (The "Ghost-Auth" approach): It uses an embedded Chromium WebView with persistent sessions. This allows the AI to act autonomously on the DOM level—summarizing your Gmail or Notion without needing any API keys or OAuth flows.
System Assistant (The Integration Shop): For tasks where APIs are available and preferred, there’s a built-in "Integration Shop". You can import Swagger/OpenAPI schemas or use pre-built plugins (Weather, Crypto, etc.).
Key Technical Details:
6-Stage Intent Pipeline: Translates natural language into categorized intents, extracts parameters, and generates a dynamic UI on a React-based canvas.
100% Local & Private: All data is stored in a local SQLite DB. You decide when the AI reads the screen.
Tech Stack: React 19, TypeScript, Electron 34, Hono, and Drizzle ORM.
I’ve reached a codebase of 36,600+ lines in this sprint and just released the v1.0.0-alpha for Windows. I’d love to get some feedback on the architecture and the "Ghost-Auth" concept.
Hi HN,
I’m an automation engineer from Vienna and I spent the last two weeks building BiamOS. I wanted to create a desktop-native "AI Operating Layer" that sits between the user and the web, rather than just another chatbot.
The core problem I wanted to solve: Most AI tools either require complex API setups or they can’t interact with your private web data at all. BiamOS solves this with a Dual-Agent System:
Web Copilot (The "Ghost-Auth" approach): It uses an embedded Chromium WebView with persistent sessions. This allows the AI to act autonomously on the DOM level—summarizing your Gmail or Notion without needing any API keys or OAuth flows.
System Assistant (The Integration Shop): For tasks where APIs are available and preferred, there’s a built-in "Integration Shop". You can import Swagger/OpenAPI schemas or use pre-built plugins (Weather, Crypto, etc.).
Key Technical Details:
6-Stage Intent Pipeline: Translates natural language into categorized intents, extracts parameters, and generates a dynamic UI on a React-based canvas.
100% Local & Private: All data is stored in a local SQLite DB. You decide when the AI reads the screen.
Tech Stack: React 19, TypeScript, Electron 34, Hono, and Drizzle ORM.
I’ve reached a codebase of 36,600+ lines in this sprint and just released the v1.0.0-alpha for Windows. I’d love to get some feedback on the architecture and the "Ghost-Auth" concept.