i got tired of raw data leaks and api billing loops, so i built an open-source local AI agent bridge

hey builders,

wanted to share the story of my side project: dwn.bridge.

like many of you, i wanted to use AI agents to automate tasks in my local workspace—like querying databases, organizing files, and debugging code. but i hit two major walls:

1. data privacy: uploading raw client database rows or private spreadsheet files to cloud LLMs is a massive security risk.

2. cost: letting a coding agent run in recursive loops to fix bugs was setting my credit card on fire with API token bills.

so i spent my nights building a local-first desktop harness (written in c# wpf).

how it works:

- it uses browser automation (webview2) to bridge to your active web session. result: $0 api key costs.

- it maps schemas locally: it only sends empty database headers to the cloud. the llm writes the query, but the client executes it locally on your hard drive. your raw data rows never leave your computer.

- offline test loops: if the model writes a script or a nodejs app, the local client runs compiler checks and test suites (jest, etc.) offline, feeding errors back to the model until the code works.

- user-in-the-loop: every CLI execution or write command is intercepted and stopped until you click "approve" on a desktop dialog showing a diff of the changes.

it's fully open source. the code is on github: https://github.com/MarckDWN/DWN.BRIDGE
Look at some short video demo here :
- short video w/sql example : https://www.youtube.com/watch?v=dCtOsXAuPgc
- another app building example : https://www.youtube.com/watch?v=X0VprcUYlyM

i'm bootstrapping this alone and keeping it free. would love to get your feedback on how you manage data security when running ai agents in your local environment!