LangChain .NET Documentation

LangChain .NET supports a functional, composable syntax for building pipelines called Chains. Chains allow you to pipe operations together using the | operator, creating a declarative workflow.

Common chain components include:

• Set(value): Sets the initial input value.
• RetrieveSimilarDocuments(collection, embeddingModel, amount): Fetches relevant documents from a vector collection.
• CombineDocuments(outputKey): Merges retrieved documents into a single string assigned to a specific key.
• Template(promptTemplate): Injects variables into a prompt string.
• LLM(model): Sends the final prompt to the language model.

To execute a chain, use await chain.RunAsync(inputKey).

var promptTemplate = @"Use the following context: {context}\nQuestion: {text}\nAnswer:";

var chain = 
    Set("Who was drinking a unicorn blood?")
    | RetrieveSimilarDocuments(vectorCollection, embeddingModel, amount: 5)
    | CombineDocuments(outputKey: "context")
    | Template(promptTemplate)
    | LLM(llm.UseConsoleForDebug());

var chainAnswer = await chain.RunAsync("text");

You can build a Retrieval-Augmented Generation (RAG) pipeline by manually orchestrating models, vector databases, and document loaders using asynchronous methods.

This approach involves:

1. Initializing an LLM and an Embedding model via a provider (e.g., OpenAiProvider).
2. Creating a vector database (e.g., SqLiteVectorDatabase).
3. Loading documents from a source (e.g., a PDF URL) using a loader (e.g., PdfPigPdfLoader) and storing them in a collection.
4. Retrieving similar documents using GetSimilarDocuments.
5. Passing the retrieved context to the LLM via GenerateAsync.

// Initialize models
var provider = new OpenAiProvider(
    Environment.GetEnvironmentVariable("OPENAI_API_KEY") ??
    throw new InconclusiveException("OPENAI_API_KEY is not set"));
var llm = new OpenAiLatestFastChatModel(provider);
var embeddingModel = new TextEmbeddingV3SmallModel(provider);

// Create vector database from Harry Potter book pdf
using var vectorDatabase = new SqLiteVectorDatabase(dataSource: "vectors.db");
var vectorCollection = await vectorDatabase.AddDocumentsFromAsync<PdfPigPdfLoader>(
    embeddingModel, 
    dimensions: 1536, 
    dataSource: DataSource.FromUrl("https://example.com/book.pdf"),
    collectionName: "harrypotter",
    textSplitter: null);

// Find similar documents
const string question = "Who was drinking a unicorn blood?";
var similarDocuments = await vectorCollection.GetSimilarDocuments(embeddingModel, question, amount: 5);

// Use similar documents and LLM to answer
var answer = await llm.GenerateAsync($"... {similarDocuments.AsString()} ... Question: {question}");

The LangChain .NET CLI provides several common options that can be used across different commands to control input, output, debugging, and model configuration.

Input Options

• --input, -i: Specifies the input text. Defaults to an empty string.
• --input-file: Specifies the path to an input file.

Output Options

• --output-file: Specifies the path where the output should be saved.

Configuration and Debugging

• --debug: A boolean flag to enable showing debug information. Defaults to false.
• --model: Specifies the model to use for the command. Defaults to o3-mini.
• --provider: Specifies the provider to use for the command.

# Example usage of common flags
langchain-cli <command> --input "Hello world" --model "gpt-4o" --provider "OpenAI" --debug

# Example usage with files
langchain-cli <command> --input-file ./data.txt --output-file ./result.txt