langchain-security

LangChain Security Integration

Add security scanning to LangChain.js applications using Weave Protocol.

When to Use

• Building LangChain chains or agents that handle untrusted input
• Implementing RAG pipelines that need document scanning
• Protecting against prompt injection in production LLM apps
• Adding approval gates for dangerous tool operations
• Redacting PII from retrieved documents

Installation

npm install @weave_protocol/langchain @langchain/core

Core Patterns

1. Callback Handler (Recommended)

import { WeaveSecurityCallback } from '@weave_protocol/langchain';

const callback = new WeaveSecurityCallback({
  action: 'block',        // block | warn | log
  minSeverity: 'medium',  // low | medium | high | critical
});

// Works with any LangChain component
const chain = new LLMChain({
  llm,
  prompt,
  callbacks: [callback],
});

2. Secure Tools

import { createSecureTool, createHighRiskTool } from '@weave_protocol/langchain';

// Basic security scanning
const secureTool = createSecureTool(myTool, {
  name: 'my-tool',
  security: { action: 'block' },
});

// Require user approval
const approvedTool = createHighRiskTool(dangerousTool, async (input) => {
  return await askUser(`Execute: ${input}?`);
});

3. Secure Retrievers

import { createSecureRetriever } from '@weave_protocol/langchain';

const secureRetriever = createSecureRetriever(vectorStore.asRetriever(), {
  name: 'docs',
  scanDocuments: true,
  redactSensitive: true,  // Auto-redact PII
});

Presets

import {
  createStrictSecurityCallback,    // Blocks medium+
  createWarningSecurityCallback,   // Logs only
  createProductionSecurityCallback // Blocks high+
} from '@weave_protocol/langchain';

Event Handling

new WeaveSecurityCallback({
  onSecurityEvent: (event) => {
    if (event.threats.length > 0) {
      logger.warn('Threat detected', {
        category: event.threats[0].category,
        mitreId: event.threats[0].mitreId,
      });
    }
  },
});