LangChain announced two major upgrades at its Interrupt conference designed to tackle the debugging nightmare of AI agents: a new database called SmithDB and an automated troubleshooting engine called LangSmith Engine.
Old infrastructure couldn't keep up with the growing flood of trace data (execution logs). The new SmithDB ditches local storage for an object-based architecture, boosting query performance on core workloads by up to 15 times.
On top of that faster foundation, LangSmith Engine automates the whole "fix the bug" process. It continuously monitors production environments, automatically groups failed calls, and pinpoints which piece of code is causing the problem. Then it drafts a pull request with the fix and a corresponding eval suite — without a human lifting a finger.
For complex agents, manually sifting through mountains of traces has become the biggest bottleneck. With this update, LangChain essentially turns the entire debug cycle — find the error, locate the code, auto-fix, add tests — into a fully automated pipeline.