AgentGuard

Detect loops and wasted LLM/tool calls in real time.

Install

pip install agentguard

Usage

1. Decorator Mode (Recommended)

Wrap your functions with @track. Every call is recorded automatically.

from agentguard import start_guard, stop_guard, track

start_guard()

@track
def search(x):
    return x

for x in ["a", "b", "c", "a", "b", "c"]:
    search(x)

stop_guard()

2. Manual Mode (Full Control)

Use the Guard class directly. Best when you need runtime control.

from agentguard.guard import Guard

g = Guard()
g.start()

for x in ["a", "b", "c", "a", "b", "c"]:
    g.track_step("search", x)

    if g.should_stop():
        print("Stopping early")
        break

print(g.stop())

3. Hybrid Mode (Advanced)

from agentguard import start_guard, stop_guard, track

start_guard()

@track
def search(x):
    return x

for x in ["a", "b", "c", "a", "b", "c"]:
    search(x)

stop_guard()

# Hybrid mode allows combining decorator tracking with manual control if needed

Example Output

AgentGuard Report

Verdict: BAD (High Waste) Reason: Too many repeated calls (61%)

Total Calls: 44 Wasted Calls: 27 Waste Ratio: 61%

Status:

• Loop Detected: No
• High Waste: Yes

Warnings:

• High number of repeated calls

Suggestions:

• Cache or reuse previous tool results
• Avoid duplicate LLM/tool calls

Why use AgentGuard?

• Detect infinite loops in AI agents
• Identify wasted LLM or tool calls
• Reduce API costs
• Improve efficiency of workflows

Status

v0.1 — Core detection working