open-guardrail-langchain

LangChain.js integration for open-guardrail.

Install

npm install open-guardrail-langchain open-guardrail

Usage

import { createGuardrailChain } from 'open-guardrail-langchain';
import { pipe, promptInjection, pii, toxicity } from 'open-guardrail';

const guardrail = createGuardrailChain({
  input: pipe(
    promptInjection({ action: 'block' }),
    pii({ entities: ['email'], action: 'mask' }),
  ),
  output: pipe(
    toxicity({ action: 'block' }),
  ),
  onBlocked: (result, stage) => {
    console.log(`Blocked at ${stage}:`, result.action);
  },
});

// Guard input before sending to LLM
const safeInput = await guardrail.invoke('user message');

// Guard output after LLM response
const safeOutput = await guardrail.guardOutput(llmResponse);

// Batch
const results = await guardrail.batch(['msg1', 'msg2']);

API

createGuardrailChain(options)

Returns a GuardrailRunnable with invoke(), batch(), guardInput(), guardOutput().

| Option | Type | Description | |--------|------|-------------| | input | Pipeline | Guards for user input | | output | Pipeline | Guards for LLM output | | onBlocked | (result, stage) => void | Callback on block |

License

MIT