yashmahajan10/llm-differential-privacy-gateway

View on GitHub ↗

Noisegate: a differential privacy gateway that lets an untrusted LLM agent query sensitive data over MCP (Model Context Protocol), with a formal guarantee no individual's record can leak even if the agent is adversarial - enforcement lives in trusted code below the model, validated by a runnable attack gallery.

26 ★3 forksPythonUpdated 1mo ago

What you need to know

Noisegate, a differential privacy gateway for LLMs. An MCP server that adds calibrated noise to LLM outputs or structured records so private data is protected during AI access.

Key features

  • ✓Applies differential privacy noise to LLM outputs
  • ✓Privacy budget management
  • ✓MCP-based gateway integration

Best for

Adding privacy guarantees to LLM applications that use sensitive data

Reviewed 2026-08-11

Topics

ai-agentsai-safetyanthropicclaudedata-privacydifferential-privacyduckdbfastapilaplace-mechanismllmllm-securitymcpmodel-context-protocolprivacyprivacy-engineeringpythonsecuritystreamlittrust-boundary
Stars
26★
Forks
3
Language
Python
License
Apache-2.0
Created
2026-06-30
Last push
2026-08-02