Spring AI Privacy Guardrails¶

English | 한국어

Send tokens to the model. Give each tool only the originals its policy allows.

Demo videos (YouTube): English · 한국어

Featured in¶

Select the Protection You Need¶

What you need	Start here
PII tokenization and per-tool original disclosure	Privacy setup
Tool visibility and execution based on the user's permissions	Tool authorization setup
Request and response checks using rules, ONNX or an HTTP model	Content inspection setup

Use these features independently or together.

See It in Action¶

A customer asks for a delivery update with a customer ID and an email address. The model calls the lookup tool with tokens. Only the customer ID is restored for the tool, and PII in its result is protected again before it returns to the model.

This example uses a custom detection rule for customer IDs and a per-tool disclosure policy for the lookup tool.

Execution trace

Two model calls surround one customer lookup, and the final response passes output inspection. The application records model/tool calls and guardrail decisions together.

Reference¶

Guide	Covers
Getting Started	Common setup and first steps for privacy protection, tool authorization and content inspection.
Configuration	Dependencies, client creation, feature selection and policy reuse.
Sample Guide	Inspector scenarios, expected results, endpoints, and language selection.
Privacy Protection	Analyzers, tokenization, tool disclosure, output protection and processing limits.
Spring Security Tool Authorization	Optional principal-aware tool discovery and execution checks, Tool Search, and asynchronous context.
Content Inspection	Input and optional output inspection with rules, local ONNX text classification models or HTTP guard models.
Runtime Observation	Recorded normal and blocked requests and Observer usage.
Architecture	Module boundaries, request sessions, evidence resolution, and execution lifecycle.
Threat model	Protected assets, trust boundaries, controls, limitations, and separately managed areas.
Evaluation	Boundary tests, the deterministic analyzer baseline, and repository benchmarks.
Troubleshooting	PII detection, rule selection, tool disclosure, authorization, streaming and inspection setup issues.

Sample¶

The sample application uses fixed examples to illustrate Local Tool, RAG, MCP, Security and Inspection. See the Sample Guide for scenarios and run instructions.