agentmolt 0.1.0

Agent Control Panel — monitor, control, and kill-switch your AI agents

AgentMolt (ACP) — Python SDK

Monitor, control, and kill-switch your AI agents. Works with OpenAI, Anthropic, LangChain, CrewAI, and any LLM provider.

Install

pip install agentcontrol

With OpenAI support:

pip install agentcontrol[openai]

Quick Start — 2 Lines

import acp
from openai import OpenAI

# Patch OpenAI to auto-monitor all calls
client = OpenAI()
panel = acp.ACP(api_key="acp_...")
panel.patch_openai(client)

# Use OpenAI normally — ACP logs everything automatically
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
    acp_agent_id="my-agent"  # tag which agent this is
)

Budget Enforcement

from acp import ACP, AgentBudget

panel = ACP()
panel.register("research-agent", budget=AgentBudget(
    max_cost_usd=10.00,    # Kill at $10
    max_tokens=100_000,     # Kill at 100K tokens
    max_calls=50,           # Kill after 50 calls
    alert_at_pct=0.8,       # Alert at 80%
))

# Agent is automatically killed when budget exceeded
panel.log("research-agent", model="gpt-4o", tokens=5000, cost=0.025)

Kill Switch

# Manual kill
panel.kill("rogue-agent", reason="suspicious activity detected")

# Check before executing
if panel.is_alive("my-agent"):
    do_work()
else:
    print("Agent is dead. Standing down.")

Decorators

from acp import monitor, budget, kill_switch

@budget("expensive-agent", max_cost_usd=50.0)
@kill_switch("expensive-agent")
@monitor("expensive-agent")
def run_analysis():
    # Your agent code here
    ...

CLI

# Scan for shadow agents
acp scan

# Check agent status
acp status

# Kill a rogue agent
acp kill marketing-gen --reason "unauthorized spend"

How It Works

ACP wraps your LLM client calls (same pattern as OpenTelemetry):

1. Before each call: Check if agent is alive + within budget
2. After each call: Log tokens, cost, latency, model
3. On budget exceeded: Kill the agent, block future calls
4. Background: Flush events to ACP dashboard every 5 seconds

No code changes needed beyond the initial patch_openai() call.

Dashboard

Connect to agentmolt.dev for:

• Real-time agent monitoring
• Cost tracking per agent
• Security scoring
• Audit trails
• One-click kill switch

License

MIT — agentmolt.dev