nudgeops 0.2.2

Runtime semantic guardrails for AI agents. Detect loops. Nudge agents back on track.

NudgeOps

Runtime semantic guardrails for AI agents.

Detect loops. Nudge agents back on track. Stop runaway costs.

Features

• Pattern Detection: Stutter, insanity, phantom progress, ping-pong loops
• Intent-Level Protection: LLM-based thought normalization to detect strategy repetition
• Two-Level Blocking: Block exact action repeats AND exhausted strategies
• LangGraph Integration: Drop-in guard for LangGraph workflows
• Observability: Track blocks, saves, and ROI

Installation

pip install nudgeops

# With LangGraph support
pip install nudgeops[langgraph]

# With OpenAI for thought normalization
pip install nudgeops[openai]

# Everything
pip install nudgeops[all]

Quick Start

Basic Usage (Pattern Detection)

from nudgeops import UniversalGuard

guard = UniversalGuard()

# In your agent loop
result = guard.check(state)
if result.blocked:
    print(f"Loop detected! {result.reason}")

Smart Guard (Intent-Level)

from nudgeops import SmartNudgeOps, MockLLMClient

# Create guard
nudgeops = SmartNudgeOps(llm_client=MockLLMClient())

# Check before each action
result = nudgeops.check(
    state={"page": "search"},
    thought="I should search for XYZ-9999",
    tool_name="search",
    args={"query": "XYZ-9999"}
)

if result.blocked:
    print(f"Blocked: {result.reason}")
    print(f"Nudge: {result.nudge_message}")

LangGraph Integration

from nudgeops import SmartNudgeOps
from langgraph.graph import StateGraph

# Build your graph
builder = StateGraph(AgentState)
builder.add_node("agent", agent_node)
builder.add_node("tools", tool_node)

# Apply NudgeOps - wraps tool nodes with guard
nudgeops = SmartNudgeOps(llm_client=my_llm)
nudgeops.apply(builder)

graph = builder.compile()

How It Works

Two-Level Protection

Level 1 (Action): Block exact action repeats after 2 attempts

search({q:"XYZ-9999"}) → fails
search({q:"XYZ-9999"}) → BLOCKED (same action twice)

Level 2 (Intent): Block exhausted strategies after 3 variations

"search XYZ-9999"  → "find product by ID" (intent)
"try XYZ9999"      → "find product by ID" (same intent!)
"try XYZ 9999"     → "find product by ID" (same intent!)
"try XYZ--9999"    → BLOCKED (intent exhausted)

License

MIT