[Week of 3/4] LangChain Release Notes

✂️ Extraction Use Case

One of the main use cases for LLMs is extracting structured data from unstructured text. We've added a lot of functionality to LangChain to best support this, as well as more high level end-to-end guides for this use case!

• We've added a with_structured_output method to language models to make it easy to get back structured data
• We've updated our use case docs to dive into this (Python and JS docs)
• We've also added a repository explicitly aimed at extraction. It exposes a service to create extraction endpoints and then run documents through them
• Check out our in depth blog post on extraction here

🦜 Open Source LangGraph

We've made a bunch of improvements to LangGraph over the past few weeks:

• Automatic rendering of the created graph. This was a much requested community feature. You can now visualize a created graph - both in ascii and as a PNG! This will make it very easy to understand more complex graphs
• Conditional entry points. This allows you to define logic that controls how you enter into the graph. This function will be run as soon as the graph is invoked, and depending on the output you will be routed to a particular node downstream
• Prebuilt Tool Calling Executor. An off-the-shelf, prebuilt graph that uses OpenAI tool calling abilities. This makes it easy to get started with a common agent

📹 In Case You Missed It from LangChain

A few YouTube videos we've recently released that you should check out:

• Building a self-corrective coding assistant. LangGraph makes it easy to engineer flows with various cycles and decision points. In this video, we show how recent ideas from the AlphaCodium paper can be implemented in LangGraph to build a coding assistant from scratch.
• Universal Document Loader with langchain-airbyte. Loading documents is the first step in creating your RAG pipeline. It can often involving connecting to a variety of sources. Airbyte's recent release can help with this!
• Building long context RAG with RAPTOR. RAPTOR (Sarthi et al) is a novel indexing strategy: docs are clustered and clusters are summarized to capture higher-level information across similar docs.

🤝 From the Community

• LangSmith: The LLM Ops Breakthrough. Check out this 15 part series highlighting LangSmith from AI Accelera.
• Build a Perplexity-Inspired Answer Engine. Create your own AI search engine powered by Groq, Mistral, LangChain, Brave, and OpenAI embeddings. Repo here and Youtube tutorial here by Developers Digest.
• Chat with MySQL Database with Python | LangChain Tutorial. Discover how to interact with a MySQL database using Python and LangChain in this tutorial by Alejandro AO.
• LangGraph: An intuitive framework for AI agents workflow. Read this review of LangGraph by Mario Bonilla.