15:I[981,["511","static/chunks/511-fc0d2f24caf5741b.js","183","static/chunks/183-6beeeb801c274aaa.js","474","static/chunks/app/%5Btype%5D/layout-347af3707a752b1a.js"],"Posts"] 16:T70b,I'm wary of sending private data to cloud AI services, but local models aren't always powerful enough. So I built PromptMask, an open-source local-first privacy layer.

It uses a trusted local LLM (via Ollama, llama.cpp, etc.) to intercept your prompt before it goes to a cloud service like OpenAI. It finds sensitive data and replaces it with semantic placeholders. Instead of a generic [REDACTED] that breaks context, it creates a map like {"John Doe": "${PERSON_1_NAME}"}.

Only the anonymized prompt is sent to the cloud. When the response returns, PromptMask uses the map to restore your original data, so that your secrets never leave your machine.

This is practical on consumer hardware because the local model's job is small: it only outputs a JSON map of secrets, not a full text rewrite. My benchmarks show even sub-1B parameter models are effective at this.

Two ways to integrate it into your AI activities:

1. Python lib for developers: A drop-in replacement of OpenAI SDK. `from promptmask import OpenAIMasked as OpenAI`

2. API Gateway for client-side apps: Run `promptmask-web` to get a local reverse-proxy endpoint (localhost:8000/v1/chat/completions) that secures requests from any OpenAI-compatible app.

I'd love to hear your feedback.

GitHub Repo: https://github.com/cxumol/promptmask (MIT Licensed)

Blog post "How Not to Give AI Companies Your Secrets", a deep dive of "how"s and "why"s: https://xirtam.cxumol.com/promptmask-how-not-give-ai-secrets...

Reverse-engineering Roadsearch Plus, or, roadgeeking with an 8-bit CPU

Show HN: MaskWise: Redact, mask, and anonymize data in training files for LLMs

The non-orientable spacetime of the eternal black hole

Spotify Messages

GNU Artanis – Web app framework for Scheme

China uses electronic devices to spread cultural influence in North Korea

Ask HN: Do You Believe in Aliens?

Breaking the creepy AI in police cameras

Universal Algorithmic Intelligence

Just-in-Time UI

Estimating the Effects of Political Pressure on the Fed (2024)

A Research-Backed Cheat Sheet to Transitioning Mental States

Elon Musk Appears to Be Completely Addicted to Anime Gooner AI Slop

Zero Trust Revisited

Show HN: CodeBoarding – Visualize Code Structure

Building Invincible APIs

Holga 120-degree Wide Pinhole Camera

Show HN: I made PromptMask, a local-first privacy layer for LLMs

AI 'deadbots' are persuasive and primed for monetization

So many conferences are mid. Why was BSC 2025 so good?

Glass bottles found to contain more microplastics than plastic bottles

Apple and the Ghosts of Companies Past

Drama Model Inference Efficiency Boosted by 1.7x-2.3x

Wayback Machine Save Page

2-D Sparse Parallelism for Deep Learning Recommendation Model Training

Show HN: ProStore BETA – The best app store for iOS just got better

Show HN: ProStore – The best alternative app store for iOS

Outperforming cuBLAS on H100 [video]

Building Realistic Benchmarks for Evaluating Retrieval on Technical Documents

Google X Head on Moonshots: 10x Is Easier Than 10 Percent (2013)

Ask HN: What is the best TTS model out there?