Docs

Find the quickstart, SDK and API references, plus integration guides for all major agent frameworks.

Quickstart

Install the SDK and inspect content in under 5 minutes.
API Reference

HTTP API reference with request/response schemas.
LangChain Integration

Protect LangChain AgentExecutor with one call.
LlamaIndex Integration

Inspect LlamaIndex query engines and agent pipelines.

SDK Overview

The Repellox Python SDK places a transparent inspection layer around your agent executor or LLM client. It is built for zero-disruption adoption, so your current pipeline code remains unchanged. Add a wrapper, set your authorization policy, and let the SDK inspect each context boundary.

Install it with one pip command. Available on PyPI, the SDK supports Python 3.9 and later. A TypeScript package for Node.js agent frameworks is also offered as a preview release.

Installation

Shell 
pip install repellox

Minimal usage

Python 
import repellox

client = repellox.Client(api_key="rpx_your_api_key")

result = client.inspect(
    content=user_message,
    mode="strict"
)

if result.blocked:
    raise repellox.ThreatDetectedError(result.threat_class)

Core Concepts

Inspection Modes

Repellox offers three inspection modes governing threat blocking:

  • strict: Block threats above the confidence threshold. Production default.
  • monitor: Log, but do not block, threats. Ideal for initial calibration.
  • permissive: Block only high-confidence threats above 0.95. Best for lower-risk contexts where false positives are costly.

Authorization Policies

In an agent framework, create a tool authorization policy specifying the tools the agent may call. Enforcement occurs at the tool-call layer separately from injection detection, and both protections may operate at once.

Context IDs

Every inspection call supplies a context_id connecting related inspection events in the audit log. In multi-turn conversations, reuse that context_id on every turn to support cross-turn anomaly aggregation.

Next Steps