Skip to main content

Install

pip install vaultak

Basic Setup

from vaultak import Vaultak
from vaultak.models import AgentConfig, KillSwitchMode

# Configure once per agent
breaker = Vaultak(AgentConfig(
    agent_id="my-agent",
    kill_switch_mode=KillSwitchMode.PAUSE
))

Wrap Agent Actions

# Wrap any action your agent takes
with breaker.watch("file_write", "/etc/config", {"content": data}):
    # Your agent's action happens here
    write_file("/etc/config", data)
That’s it. Every action is now monitored, risk-scored, and governed in real time.

Response Modes

ModeWhat happens
ALERTLogs and notifies. Agent keeps running.
PAUSEAgent halts. Awaits human approval.
ROLLBACKReverses last N actions. Agent pauses.

Works With Any Framework

# LangChain
with breaker.watch("api_call", "openai.chat", payload):
    response = llm.invoke(prompt)

# CrewAI
with breaker.watch("database_write", "users_table", payload):
    crew.kickoff()

# Custom agents
with breaker.watch("file_delete", "/data/sensitive", payload):
    agent.run()

Next Steps