LangSmith Agent Builder has been renamed to LangSmith Fleet. The new name is reflected across the platform.
You can now pin any experiment as your baseline, so every subsequent run is automatically measured against it. Instead of manually selecting experiments for comparison, your chosen baseline persists across all future runs.
Agent Builder enhancements centered around collaboration. There's now a central Chat agent with access to file uploads and a tool registry, making working with an agent feel like working with a teammate.
The LangSmith Insights Agent now runs on your schedule, no manual triggers required. Set a recurring cadence (daily, weekly, or a custom cron expression) and the agent generates reports automatically.
deepagents v0.4 ships pluggable sandbox support, smarter conversation history summarization, and Responses API as the default for OpenAI models. Agents now have enhanced capabilities for complex tasks.
You can now configure which parts of a trace's inputs and outputs appear directly in the tracing table. This is especially useful for projects with custom data structures.
LangSmith now gives you a unified view of costs across your full agent workflow, not just LLM calls. Submit custom cost metadata for any run, whether that's API calls, compute, or other expenses.
This self-hosted release expands feature parity with LangSmith Cloud while improving performance, security, and operational control. It brings Insights and other cloud features to self-hosted deployments.
Pairwise Annotation Queues in LangSmith are a fast, structured way to compare two agent outputs side-by-side and pick a winner. This makes scoring subjective tasks more efficient.
LangSmith Fetch is a new CLI tool that brings full LangSmith trace access directly into your terminal, coding environment, or IDE, enabling seamless debugging workflows for agent builders.