LangChain brings managed LangSmith deployments to Azure and Google Cloud

Teams using Azure or Google Cloud can now keep LangSmith’s agent records and deployments in their own accounts without operating the platform themselves. LangChain’s November 3 announcement brings its Bring Your Own Cloud service to both clouds in beta, alongside general availability on AWS. LangChain manages the infrastructure; customers retain the data plane.

Your cloud account, LangChain’s maintenance

BYOC separates the location of customer data from the work of keeping LangSmith running. LangChain provisions and operates the service inside the customer’s virtual network, handling scaling, platform updates, database schema changes and runtime dependencies. Traces—the records of an agent’s activity—stay in the customer’s account, along with evaluations, datasets, prompts and agent deployments.

The service is an alternative to maintaining a self-hosted installation. LangChain says teams with strict compliance boundaries previously needed platform engineers to patch services, configure databases and manage migrations. Its AWS BYOC service launched in August. The company says it delivers new features soon after their LangSmith Cloud release, rather than making customers wait for the next self-hosted release.

That arrangement does not put every part of LangSmith in the customer’s account. The architecture divides the service into a LangChain-run control plane, which handles administration, and a customer-owned data plane, which contains the sensitive workload data. LangChain still operates that data plane on the customer’s behalf.

Three clouds, different deployment details

An organization can have several data planes, each tied to a particular account and region and containing its own workspaces. Teams can separate production from staging, or give business units their own deployments on different clouds. The announcement describes distinct networking and management arrangements for each provider:

• Azure: The data plane runs in a dedicated virtual network with a private endpoint for its Kubernetes service. System telemetry goes to Azure Monitor. LangSmith Sandboxes support is still planned, not available at launch.
• Google Cloud: Data planes can run in any region at launch, but are managed from LangSmith’s US control plane. Resources use names scoped to their namespace to avoid collisions with existing infrastructure.
• AWS: The service runs in an existing virtual private cloud, with its database in customer-selected subnets. Security teams can attach their own security groups to each application role to fit existing traffic-inspection policies.

All three cloud options also run Managed Deep Agents endpoints. The data plane’s LLM Gateway supports models from 11 providers out of the box, plus endpoints compatible with OpenAI or Anthropic. Customers can therefore choose the model provider as well as the cloud hosting their data plane.

Moving an existing installation takes two steps

Existing self-hosted customers can bring their data rather than start over. An open-source migration tool copies workspace resources across all three clouds while leaving the source installation untouched and running. It covers users, roles, datasets, experiments and prompts, plus annotation queue settings, automation rules, dashboards and Fleet resources.

Historical traces, runs and feedback move separately, using internal tooling operated by LangChain. Trace migration is currently available on AWS. That distinction matters for teams moving an established installation: workspace migration across all three clouds does not mean the same trace-migration support is available everywhere.

The Terraform modules used to provision BYOC are public, as is the migration tool. Getting a managed deployment still requires contacting sales: LangChain directs customers to request provisioning for AWS, Azure or Google Cloud through its BYOC announcement.