Submitted my entry for the WebMCP Challenge: Albugent WebMCP.
Traditional data governance tools usually require shipping sensitive datasets to external servers to scan for PII and compliance risks. I wanted to see if we could do the entire pipeline locally, right inside the browser.
Albugent runs SQLite WASM inside a Web Worker. All SQL queries, anomaly profiling, and data masking happen strictly in browser RAM—zero data leaves the client. The AI agent registers tools natively using the document.modelContext WebMCP API, scans for unmasked PII, and proposes SQL remediation scripts. The human engineer always stays in the loop to review and approve the SQL before anything executes.
Building a zero-backend governance engine with reactive state sync between WASM worker threads and React was definitely an intense engineering sprint, but really rewarding to see working end-to-end.
What’s Next:
Albugent was built as a privacy-first proof-of-concept to show what’s possible when WASM and AI agents run entirely on the client. To turn this into a fully usable, zero-cost utility for everyone, the next step on the roadmap is adding a BYOK (Bring Your Own Key) option and local model support (via Ollama/WebLLM), along with drag-and-drop CSV and SQLite file loading.
If you'd like to try it out or look through the code:
Thanks to the team behind the WebMCP Challenge for organizing this!
P.S. Built this as a true human-AI collaboration — shoutout to AI coding tools for acting as a 24/7 pair programmer throughout the build.
Top comments (0)