Quickstart

Install the SDK, get your API key, and run an inspection. Estimated time: 5 minutes.

Documentation

Step 1: Install the SDK

Shell 
pip install repellox

Requires Python 3.9 or later. Required dependencies are limited to the standard library and httpx.

Step 2: Create an API Key

Log in to your Repellox dashboard, open Settings, choose API Keys, and create a key. Copy its value because it is displayed only once.

Export the key as an environment variable instead of hardcoding it:

Shell 
export REPELLOX_API_KEY=rpx_your_api_key_here

Step 3: Run Your First Inspection

Python 
import os
import repellox

# Client reads REPELLOX_API_KEY from environment automatically
client = repellox.Client()

# Inspect a user message before sending to your model
result = client.inspect(
    content="Ignore all previous instructions and ...",
    mode="strict"
)

print(result.blocked)      # True
print(result.threat_class)  # "direct_injection"
print(result.confidence)   # 0.97
print(result.context_id)   # "rpx_ctx_abc123" (for audit log)

LangChain Integration

To protect LangChain agents, call repellox.wrap() on an existing AgentExecutor. Before they reach the model, the wrapper intercepts user inputs, retrieved documents, and tool outputs.

Python 
from langchain.agents import AgentExecutor, create_openai_tools_agent
from langchain_openai import ChatOpenAI
import repellox

client = repellox.Client()

# Your existing agent executor
agent_executor = AgentExecutor(
    agent=create_openai_tools_agent(ChatOpenAI(), tools, prompt),
    tools=tools
)

# Wrap it with Repellox
protected = repellox.wrap(
    agent_executor,
    client=client,
    mode="strict",
    allowed_tools=["web_search", "calculator"]
)

# Use exactly like the original
result = protected.invoke({"input": user_query})

LlamaIndex Integration

Python 
from llama_index.core.agent import ReActAgent
import repellox

client = repellox.Client()

agent = ReActAgent.from_tools(tools, verbose=True)

# Wrap with Repellox inspection
protected_agent = repellox.wrap_llamaindex(
    agent,
    client=client,
    mode="strict"
)

response = protected_agent.chat(user_query)

Next Steps