New in LangSmith
- Streaming in the playground: The playground now supports streaming! You can enter the playground by clicking on the “Open in Playground” button on the upper right hand corner of select runs.
- Infrastructure changes to support more users: We’ve beefed up our backend. That means more of you off the waitlist soon!
- UI and navigation enhancements: We’ve made a lot of quality-of-life improvements to the UI, mainly around the run details page. Biggest change is the trace tree is now on the left hand side. See the screenshot below for an up-to-date picture.
New in Open Source
- LangChain Experimental: We’re moving select chains into a separate package to make LangChain leaner, more focused, and safer. This does induce some backwards incompatible changes. Read the GitHub discussion here.
- Docs restructures: We’ve made a lot of improvements to docs. First, we ported over the skeleton from Python JS, so they are matching. Second, we restructured docs to better distinguish between modules and use cases.
- Web retriever: We went deep on a particular use case–a web research assistant. Turned into an interesting and novel retriever. Source code here.
- Llama 2 Support: We quickly added support and examples for Llama 2, including some Llama 2 specific prompts.
In case you missed it
- LangChain x Pinecone Webinar: Supercharge Llama 2 with RAG Webinar, recording here.
- Diagnosing and Fixing Slow Chatbots in collaboration with Daniel at Zep.
- Earning Sage, How to Transform AI into a Savvy CFO in collaboration with Yuze at LeptonAI.
- Using Open Source tools to create an AI companion in collaboration with RealChar.
- Automating Web Research by Lance and Harrison from Team LangChain.
- Support for teams and organizations to share work: Bring colleagues together in LangSmith.
- Increasing the composability of chains: We’re working on some syntax to allow for chains and components to be truly composable.
- Better use case sections in the documentation: All use case sections will be getting a face lift.
Use-cases we love
- CLAIRe, a conversational Learning AI with recall from Suvan
- Q&A over *data* interact with your Snowflake database using natural language queries from Streamlit
- A personal financial advisor and copilot from Fintual
- An AI companion by Realchar
Want more of these in your inbox? Sign up here.