@promptpack/langchain

LangChain.js integration for PromptPack - organize, version, and manage AI prompts with built-in governance and tools.

Features

• 🎯 Prompt Registry - Load and manage PromptPacks from JSON/YAML
• 🔧 Tool Integration - Built-in tool calling with governance policies
• ✅ Validation - Response validation and guardrails
• 🔄 LangChain Native - Seamless integration with LangChain templates
• 📦 Type Safe - Full TypeScript support

Installation

npm install @promptpack/langchain

Peer Dependencies:

npm install @langchain/core langchain

Quick Start

1. Create a PromptPack

Create a JSON file customer-support.json:

{
  "$schema": "https://promptpack.org/schema/v1.0/promptpack.schema.json",
  "id": "customer-support",
  "name": "Customer Support Pack",
  "version": "1.0.0",
  "description": "Customer support prompts with multiple task types",
  "template_engine": {
    "version": "v1",
    "syntax": "{{variable}}"
  },
  "prompts": {
    "support": {
      "id": "support",
      "name": "Support Agent",
      "version": "1.0.0",
      "system_template": "You are a {{role}} for {{company}}. Help customers with their {{issue_type}} issues.",
      "variables": [
        {
          "name": "role",
          "type": "string",
          "required": true,
          "description": "The role of the assistant",
          "example": "helpful support agent"
        },
        {
          "name": "company",
          "type": "string",
          "required": true,
          "description": "Company name"
        },
        {
          "name": "issue_type",
          "type": "string",
          "required": false,
          "default": "general",
          "validation": {
            "enum": ["billing", "technical", "general"]
          }
        }
      ],
      "parameters": {
        "temperature": 0.7,
        "max_tokens": 1500
      }
    }
  }
}

2. Use with LangChain

import { PromptPackRegistry, PromptPackTemplate } from '@promptpack/langchain';
import { ChatOpenAI } from '@langchain/openai';

// Load the pack
const pack = PromptPackRegistry.loadFromFile('./customer-support.json');

// Create a template
const template = new PromptPackTemplate({
  pack,
  promptId: 'support',
});

// Use with LangChain
const model = new ChatOpenAI({ modelName: 'gpt-4' });
const chain = template.pipe(model);

// Invoke with variables
const response = await chain.invoke({
  role: 'helpful support agent',
  company: 'TechCorp',
  issue_type: 'billing',
});

console.log(response.content);

Documentation

• Getting Started - Installation, quick start, basic usage
• Core Concepts - Registry, templates, variables
• Tools - Function calling with governance
• Validators - Response validation and guardrails
• API Reference - Complete API documentation
• Examples - Code examples and tutorials

Core Concepts

Registry

Load and manage PromptPacks:

import { PromptPackRegistry } from '@promptpack/langchain';

// Load from file
const pack = PromptPackRegistry.loadFromFile('./pack.json');

// Or from directory
const registry = PromptPackRegistry.fromDirectory('./packs');

// Get a specific prompt
const prompt = registry.getPrompt('pack-id', 'prompt-id');

See Core Concepts for details.

Templates

Use PromptPackTemplate with LangChain:

const template = new PromptPackTemplate({
  pack,
  promptId: 'support',
});

const chain = template.pipe(model);
const response = await chain.invoke({ role: 'agent', company: 'TechCorp' });

See Core Concepts for variables, validation, and advanced features.

Tools

Define tools at pack level with governance policies:

const tools = template.filterTools(allTools);
const model = new ChatOpenAI({ model: 'gpt-4o-mini' }).bindTools(tools);
const chain = template.pipe(model);

See Tools Documentation for governance, policies, and examples.

Validators

Validators work like tools - define them in your PromptPack, provide custom implementations:

import { OOTB_VALIDATORS, CustomValidatorFn } from '@promptpack/langchain';

// Custom validator implementations
const customValidators: Record<string, CustomValidatorFn> = {
  sentiment: (response, validator) => {
    // Your validation logic
    return { passed: true, validatorType: 'sentiment' };
  },
  pii_detection: (response, validator) => {
    // Your PII detection logic
    return { passed: true, validatorType: 'pii_detection' };
  },
};

// Create chain - validators come from pack definition
const chain = template
  .pipe(model)
  .pipe(template.createValidationRunnable(customValidators));

// Use with LangChain features
const result = await chain
  .withRetry({ stopAfterAttempt: 3 })
  .invoke(input);

Validator Types:

• OOTB (always available): banned_words, max_length, min_length, regex_match
• Custom (you provide): sentiment, pii_detection, toxicity, etc.

See Validators Documentation for complete guide.

Examples

The examples/ directory contains complete working examples:

• examples/basic-usage.ts - Basic usage and JSON pack loading
• examples/conversation.ts - Multi-turn conversations with memory
• examples/integration.ts - LangChain features (streaming, batching, chaining)
• examples/tools.ts - Tool calling with governance
• examples/validation-native.ts - LangChain-native validation with .pipe(), .withRetry(), .withFallbacks()
• examples/validation-error-test.ts - Missing validator error handling

Advanced Usage

For advanced features like model overrides, fragments, pipelines, and caching, see:

• Advanced Documentation
• API Reference

Contributing

Contributions welcome! See our contributing guidelines.

License

MIT

Links

• PromptPack Specification
• LangChain.js
• GitHub Repository