Skip to content

msb-innovation/DAGE-Security-Architecture

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 

Repository files navigation

DAGE Security Architecture Model

Deterministic Agent Gateway Engine
by MSB Innovation GmbH

DAGE is a deterministic security architecture model for AI agent systems with strictly deterministic permission and tool control.

The architecture separates probabilistic language processing from all security-critical system functions.

Core principle:

The language model may think, analyze and formulate,
but it must never act directly.

Website:
https://msb-innovation.de/DAGE.html


Security Goals

DAGE follows a deterministic zero-trust architecture for AI agent systems.

  • Prevent prompt injection attacks
  • Prevent unauthorized tool execution
  • Enforce deterministic user isolation
  • Isolate LLM reasoning from execution
  • Protect local enterprise data
  • Enforce deterministic permission control

Threat Model

DAGE is designed to mitigate:

  • prompt injection
  • unauthorized tool execution
  • cross-user data leakage
  • insecure RAG access
  • uncontrolled agent autonomy
  • privilege escalation through LLM reasoning

Architecture Concept

The DAGE architecture enforces a strict separation between:

  • probabilistic AI reasoning
  • deterministic execution control

The language model never receives direct access to:

  • filesystems
  • APIs
  • databases
  • operating system processes
  • networks
  • external tools

The LLM can only generate text-based tool requests.

All validation, permission checks and execution decisions are handled deterministically by the DAGE gateway layer.


Deterministic Context Control

DAGE does not only control tool execution.

The architecture also deterministically controls:

  • system prompts
  • user isolation
  • RAG retrieval
  • memory access
  • document injection
  • tool responses
  • context assembly
  • conversation history
  • external knowledge inclusion

The language model never determines, modifies or authorizes its own security-relevant context.

All context construction and filtering is performed deterministically by the DAGE gateway layer.


High-Level Architecture

User Request
  ->
DAGE Gateway
  ->
LLM
  ->
Text-Based Tool Request
  ->
DAGE Validation Layer
  ->
Authorized Tool Execution
  ->
Deterministic Result Handling
  ->
DAGE Gateway Response
  ->
User

About

Deterministic security architecture model for AI agent systems with deterministic permission and tool control.

Topics

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors