CLM — Cognitive Load Manager

Real-time metacognitive middleware for LLM agents. Detects when your agent is cognitively overloaded and intervenes before it hallucinates, drifts, or crashes.

Installation

# Lightweight install (numpy only, ~10MB)
pip install clm-plugin

# Full install with embedding support (~1.5GB)
pip install clm-plugin[embed]

Quickstart

from clm import CLM

clm = CLM(verbose=True)

# In your agent loop, add one line:
result = clm.observe_raw(llm_output)

if result.action == "interrupt":
    # CLM detected cognitive overload
    prompt = f"Clarification needed: {result.clarification}"
elif result.action == "patch":
    # Use compressed context in next call
    context = result.context

What it monitors

CLM tracks 4 cognitive signals:

• Branching — too many tasks in flight
• Repetition — agent going in circles
• Uncertainty — excessive hedging language
• Goal drift — wandering from original intent

Combines these into a CLM score (0–100) and acts:

Configuration

from clm import CLM, CLMConfig

config = CLMConfig(
    weights=[0.30, 0.25, 0.25, 0.20],  # [branching, repetition, uncertainty, goal_distance]
    green_max=40.0,
    amber_max=70.0,
    no_embed=False,  # Auto-set to True if sentence-transformers not installed
)

clm = CLM(config, verbose=True)

Adapters

LangChain

from clm.adapters import CLMCallbackHandler

handler = CLMCallbackHandler(verbose=True)
agent.run("your task", callbacks=[handler])

Loop Decorator

from clm.adapters import CLMLoop

loop = CLMLoop(verbose=True)

@loop
def agent_step(prompt: str) -> str:
    return openai_client.chat(prompt)

Observability

clm.get_score()       # Current CLM score (0–100)
clm.get_zone()        # "Green" | "Amber" | "Red"
clm.get_history()     # Step-by-step intervention log
clm.summary()         # Session statistics

Links

• Documentation: GitHub README
• Source: GitHub Repository
• Issues: Bug Reports

License

MIT