🗣️ Introducing LangGraph
We just launched LangGraph, which helps customize your Agent Runtime. You can read more about it on the blog.
LangGraph helps construct a powerful agent executor that allows for loops in logic while keeping track of application state. Agents constructed with LangGraph can handle more ambiguous inputs and accomplish tasks more consistently. These agents can:
- Force call a tool
- Wait for human-in-the-loop approval
- Manage agent steps
- Return outputs in a specific or raw format
We’re excited to give our agents more controllable power. Learn about LangGraph by reading our docs. If you have feedback, we’d love to know!
🦜🔗 LangChain 0.1 is here!
If you haven’t yet upgraded to LangChain 0.1, you should check it out. We wrote more about it on our blog, but here are the big takeaways:
- No more breaking changes on a minor version release.
- LCEL supports parallelization, fallbacks, batch, streaming, and async all out-of-the-box.
- LangChain supports almost 700 integrations now and has improved co-maintenance with integration partners.
- More advanced retrieval to get your RAG apps behaving better.
- Largest library of tools and templates of cognitive architectures for your LangChain Agents.
- 1-click observability with LangSmith.
New in OSS
Introducing the stream_events method
- This new method streams the current step the agent is taking, retrieved sources at the step, and the agent’s response. See docs.
New and improved use case docs
- Q&A with RAG: Get started with your own question answer chat bot.
- Text to SQL: We hear of lots of companies building their own text to SQL bots; this guide will get you started.
- Tool use: Let your LLMs have access to APIs and functions to take action and integrate with existing company software.
🦜🛠️ New in LangSmith
Tracing improvements
- We refreshed the run tree so that the most important calls are displayed in the default view. Keeping traces clean and uncluttered is an important step to getting your bearing on what’s happening faster. If you want more detail, you can always show all in an expanded view.
- Better rendering for tools and function calls in the trace view and playground.
- Use our keyboard shortcuts (J, K) to navigate to the previous or next run while in the trace history view.
- Trace histories now have infinite scroll, so you can get more context.
Annotation improvements
- Faster updates and deletes on annotation labels.
Monitoring improvements
- Drill downs and dynamic stat calculations by tag on the monitoring tab, so that you filter charts by a segment.
- Notice faster rendering of monitoring charts? We’ve re-platformed our database so we can keep scaling and maintain a snappy experience in the UI.
đź‘€ Stay in the Know
Educational Resources:
- Build LLM Apps with LangChain.js by DeepLearning.AI
- LangGraph YouTube Series
- LangChain 0.1 YouTube Series
- Streaming Events with stream_events method
- Chat your PDFs by Austin Vance
Good Reads:
- Multi-agent Workflows
- Ally Financial Collaborates with LangChain to bring PII Masking into LangChain JS
- Building a Wikipedia Chatbot Using AstraDB, LangChain, and Vercel by Carter Rabasa at Datastax
- 10 Lessons from Developing an AI Chatbot Using Retrieval-Augmented Generation by Fiddler
- How Mendable leverages LangSmith to debug Tools & Actions by Nicolas Camara at Mendable
Events you should attend:
- Jan 30th: Build LLM Data Apps with LangChain and Dash
- Feb 8th: Streamlit’s SF Meetup