[360Labs.ai]
{ 360Labs }Case Study
{ Research and Development }· Live

AgentGuard

C++17/Python library for deadlock prevention in multi-AI-agent systems. Extends Dijkstra's Banker's Algorithm. LangGraph integration.

AgentGuard
// overview

ResourceManager orchestrator with SafetyChecker (Banker's Algorithm), RequestQueue, 5 scheduling policies, ProgressTracker, DelegationTracker, DemandEstimator. Direct LangGraph integration via @guarded_tool decorator.

// specs

Specifications

CoreExtended Banker's Algorithm for LLM agents
ImplementationC++17 with Python bindings via pybind11
Tests285 tests (189 C++, 96 Python)
IntegrationNative LangGraph/LangChain support
// features

Features

  • 01Progress Monitor with named metrics + auto reclamation for silent stalls
  • 02Delegation Tracker with BFS/DFS cycle detection for authority deadlocks
  • 03Adaptive Demand Estimator with statistical learning for unknown demands
  • 045 scheduling policies: FIFO, Priority, ShortestNeedFirst, DeadlineAware, Fairness
  • 05Open-source under MIT license on PyPI
// architecture

Architecture

SAFETY PIPELINE

Multi-Agent Requests
ResourceManager + RequestQueue
SafetyChecker (Banker's algorithm)
Grant (run safely)
Defer (wait)
Reject (deadlock risk)
Schedulers + Trackers

Interested
in AgentGuard?

Let's discuss how this can work for you.

Get in Touch →
// end of case studyAgentGuard