Skip to main content
The LangSmith data migration tool copies data from an existing LangSmith Cloud organization or self-hosted instance into a BYOC data plane. The tool copies data (as opposed to moving data over), so the source instance is unchanged and stays available during the move. Run the migration after your data plane is active and reachable. Trace data is not migrated, so plan the manual steps alongside the tool.

Supported resources

The tool migrates the following resources between a source instance and a destination data plane: Each resource type has its own command, flags, and caveats. For the command reference, see the tool README.
Resources belonging to a feature that BYOC does not support yet, such as Managed Deep Agents, cannot be migrated. See Available features.

Migrate with the tool

1

Install the tool

Install the latest release with uv, uvx, or pip. For the current install commands, see Installation.Verify the install:
2

Create the destination workspaces

Workspace IDs differ between instances, so create the target workspaces in the data plane first. See the Create workspaces section of the BYOC onboarding docs.Migration commands run per workspace pair. Repeat them for each pair.
3

Create the destination tracing projects

The tool does not create tracing projects. Create the projects you need in the destination workspace before you migrate automation rules or charts, or those mappings have no target.
4

Configure the connection details

Set the source and destination credentials and endpoints:
OLD refers to your current LangSmith deployment, and NEW refers to the destination BYOC data plane:
  • LANGSMITH_OLD_API_KEY: An API key from your current deployment with permission to read all resources you want to migrate. For LangSmith Cloud at smith.langchain.com, create it under Settings > API Keys. For another Cloud region or a self-hosted instance, use that deployment’s UI.
  • LANGSMITH_NEW_API_KEY: An API key created under Settings > API Keys at aws.smith.langchain.com. Scope it to a workspace in your destination BYOC data plane. See Create an API key.
  • LANGSMITH_OLD_BASE_URL: Your current deployment’s API URL. For LangSmith Cloud in GCP US, use https://api.smith.langchain.com. For another Cloud region, use its regional API URL. For self-hosted LangSmith, use your instance’s API URL.
  • LANGSMITH_NEW_BASE_URL: Your destination BYOC data plane’s API URL, including https://. Find it under Settings > Data Planes. See BYOC onboarding.
The destination key must be scoped to a workspace in the target data plane. An organization-scoped key does not work.To migrate Fleet agents with an owner, use a personal access token (lsv2_pt_*) as the destination key. Workspace API keys carry no user identity, so agents created with them have no owner.
5

Test both connections

6

Run the migration

Use the interactive wizard to walk through every resource type:
To migrate one resource type at a time, run its command instead, such as datasets, prompts, or fleet. Add --dry-run to preview a step without writing, and -v for verbose output. Use langsmith-migrator resume to retry pending or failed items from a previous session.For per-command flags, workspace mapping, and project mapping, see the tool README.

Complete the Fleet manual steps

Fleet resources migrate with the fleet command, but some values cannot cross instances. Complete the following in the destination workspace after migration:
  • Re-enter secret values: The Fleet API does not return secret values. The tool creates workspace secrets and auth provider client secrets as empty placeholders.
  • Re-authenticate OAuth connections: Per-user agent connections, such as Gmail, Slack, and GitHub, are tied to individual user tokens. Each user must reconnect.
  • Re-share agents: Per-user access lists keep only user IDs that exist on the destination. The tool reports which users it removed.
  • Check agent models: The tool substitutes a model when the destination catalog does not offer the source model. It logs every substitution.
  • Configure infrastructure-level settings: OAuth providers, the GitHub App, and the Slack app can be recreated in the Settings > Oauth Providers and Fleet > Integrations UI.
Fleet migrators never overwrite resources that already exist on the destination, so re-running fleet is safe.

Complete the Engine manual steps

Engine issues migrate with the issues command. The command creates any tracing project that has Engine data and is missing on the destination. Issues migrate as metadata only, and the command skips issues and project configurations that already exist on the destination. To run Engine in the data plane, complete these steps in order:
  1. Enable Engine for the data plane: A LangChain operator enables Engine for your data plane. Contact your LangChain account team to request it.
  2. Turn on Engine: An Organization Admin goes to Settings > Engine, selects the data plane, and turns on Engine.
  3. Add model provider credentials: An Organization Admin adds credentials for your own model providers. Credentials are not migrated. See Engine on BYOC.
  4. Reconnect GitHub: Configure your GitHub App in the destination, then reconnect repositories in each Engine project. Repository settings migrate, but they work only after the destination has a GitHub connection. See Configure GitHub for Engine on BYOC.
  5. Recreate notification destinations: Connect your Slack app and add Slack, Jira, or webhook destinations again. See Engine notifications.
  6. Let Engine re-detect traces: Migrated issues have no linked traces, pull requests, or suggested actions. Engine links new traces to the issues as it analyzes traces in the data plane.

Move data the tool does not migrate

Traces

The tool does not migrate traces. To keep historical traces beyond the lifecycle of your old instance, bulk export them to an S3-compatible bucket. To move tracing applications, create API keys scoped to a workspace in the data plane, then repoint the application:

Deployed agents

Redeploy agents that run on your existing instance into a workspace in the data plane. Use the LangSmith UI or the API. You may also need to recreate service accounts, image pull secrets, and other resources in the data plane cluster. For agent databases and caches, choose one of the following:
  • Create new instances in the data plane and point the redeployed agents at them. Start empty, or restore from a backup of the old instance.
  • Point the redeployed agents at your existing database instances, and reach them from the data plane cluster over PrivateLink or VPC peering.
For more information, see LangSmith Deployment.

Insights reports

Insights reports are not migrated across instances. To keep a report, open its details in the LangSmith UI and select Download.

See also