llama-index-callbacks-vaultak

Runtime security for LlamaIndex agents, powered by Vaultak.

Intercept every agent action, tool call, and LLM query in real time — score risk, enforce policies, mask PII, and automatically block dangerous behavior before it reaches your production systems.

Install

pip install llama-index-callbacks-vaultak

Quick Start

from llama_index.core.callbacks import CallbackManager
from llama_index.callbacks.vaultak import VaultakCallbackHandler

# Initialize the handler
handler = VaultakCallbackHandler(api_key="vtk_...")
callback_manager = CallbackManager([handler])

# Use with a query engine
query_engine = index.as_query_engine(
    callback_manager=callback_manager
)

# Query — every action is now monitored and secured
response = query_engine.query("Summarize our Q3 revenue data")

Set Globally

from llama_index.core import Settings
from llama_index.core.callbacks import CallbackManager
from llama_index.callbacks.vaultak import VaultakCallbackHandler

Settings.callback_manager = CallbackManager([
    VaultakCallbackHandler(api_key="vtk_...")
])

Use with Agents

from llama_index.core.agent import ReActAgent
from llama_index.core.callbacks import CallbackManager
from llama_index.callbacks.vaultak import VaultakCallbackHandler

handler = VaultakCallbackHandler(
    api_key="vtk_...",
    agent_name="my-production-agent",
    risk_threshold=6.0,
)

agent = ReActAgent.from_tools(
    tools,
    callback_manager=CallbackManager([handler]),
)

What Gets Monitored

Configuration

handler = VaultakCallbackHandler(
    api_key="vtk_...",           # Required
    agent_name="my-agent",       # Label in the Vaultak dashboard
    block_on_high_risk=True,     # Block actions above threshold
    risk_threshold=7.0,          # 0-10 scale
    verbose=True,                # Log all scored actions
)

Links

• Vaultak docs
• LlamaIndex docs
• PyPI
• GitHub