-3
DWN.BRIDGE – Open-source MIT License. C# client for secure database querying and sandboxed AI execution
(programming.dev)
Hy,
I wanted to share a desktop client I've been building and recently open-sourced under the GPL-3.0 license: DWN.BRIDGE (C# WPF, targeting .NET 10). It is designed to solve the data privacy issue when using web LLMs (like Gemini) to analyze files or query local databases. It exposes a local orchestration bridge to your browser session via WebView2 automation, bypassing developer API paywalls while keeping your raw data offline. Key Features:
- Zero-Knowledge Schema Mapping: It extracts table headers and column metadata locally. It sends only this structure to the web LLM. The LLM acts purely as a logic generator, and the client runs the SQL query locally using native providers (Dapper/SqlClient). Raw data rows never leave your hard drive.
- Process Sandboxing & Consent: Every shell script execution, SQL write, or file modification requested by the agent is paused. The client prompts you with a WPF modal showing a git-style diff of the command/file for manual approval.
- Coder Compilation Loop: If the agent generates scripts, it compiles and runs unit tests locally, feeding error logs (stderr) back to the context window until it builds successfully. The codebase is completely audit-friendly. I'd love to get feedback on the security consent model or contributions on expanding local DB support! 📺 Demo Video: https://www.youtube.com/watch?v=dCtOsXAuPgc 🔗 GitHub: https://github.com/MarckDWN/DWN.BRIDGE
