LangSmith launches Preview Builds for agent testing

LangSmith has launched a new deployment feature called Preview Builds in public beta, designed to simplify how developers test AI agent modifications before merging them into production. Currently available for deployments connected via GitHub integration on LangSmith Cloud, the feature automatically spins up a temporary, production-like staging environment directly from a pull request branch. This eliminates the need for developers to manually configure new deployments or force collaborators to clone repositories and replicate local setups to review code changes.

The system offers two main triggers for generating these temporary environments: Every PR, which builds a preview for any pull request opened against the deployment branch, and Label only, which restricts builds to pull requests tagged with a specific GitHub label. When a developer pushes a new commit to an active pull request, LangSmith automatically generates a new revision of the preview deployment. To help teams manage resources, the platform includes an Idle TTL setting to automatically tear down inactive previews, alongside a Max concurrent previews limit to control how many environments run simultaneously.

For development teams, this workflow bridges the gap between local testing and production deployment. It allows non-technical stakeholders, such as product managers, domain experts, and quality assurance testers, to interact with the live agent, validate its behavior, and check terminology. Meanwhile, engineers can use the same preview environment to inspect traces, monitor tool calls, and debug failure paths. Preview deployments inherit secrets from their parent deployment by default, though teams can override these credentials to secure sensitive services against unauthorized access from external contributors.