Chains

use langchainx::{
    chain::{Chain, ConversationalChainBuilder},
    llm::openai::OpenAI,
    memory::SimpleMemory,
    prompt_args,
};

let chain = ConversationalChainBuilder::new()
    .llm(OpenAI::default())
    .memory(SimpleMemory::new().into())  // Arc<Mutex<dyn BaseMemory>>
    .build()?;

// Turn 1
chain.invoke(prompt_args! { "input" => "My name is Alice" }).await?;

// Turn 2 — chain remembers history
let reply = chain.invoke(prompt_args! { "input" => "What is my name?" }).await?;

The default input key is "input". History is stored under "history" in the prompt.

use langchainx::chain::{SequentialChainBuilder, Chain};

let chain = SequentialChainBuilder::new()
    .add_chain(chain1)   // Box<dyn Chain>
    .add_chain(chain2)
    .build()?;

// Output of chain1 is automatically piped as input to chain2
// using the output_key of chain1 as an input key for chain2

Stuffs retrieved documents into the prompt context.

use langchainx::chain::StuffDocumentBuilder;

let chain = StuffDocumentBuilder::new()
    .llm(OpenAI::default())
    .build()?;

Used internally by ConversationalRetrieverChain and question_answering::load_qa_with_sources_chain.

use langchainx::{
    chain::{Chain, ConversationalRetrieverChainBuilder},
    llm::openai::OpenAI,
    memory::SimpleMemory,
    vectorstore::Retriever,
    prompt_args,
};

let retriever = Retriever::new(vector_store, 5);

let chain = ConversationalRetrieverChainBuilder::new()
    .llm(OpenAI::default())
    .retriever(retriever)
    .memory(SimpleMemory::new().into())
    .build()?;

let answer = chain.invoke(prompt_args! {
    "input" => "What does the document say about X?"
}).await?;

use futures::StreamExt;

let mut stream = chain.stream(prompt_args! { "input" => "Tell me a story" }).await?;
while let Some(chunk) = stream.next().await {
    match chunk {
        Ok(data) => print!("{}", data.content),
        Err(e) => eprintln!("stream error: {e}"),
    }
}

Note: chains with memory do NOT automatically save memory when using stream() — only call() and invoke() save to memory.