Run support queries against PostgreSQL

This Helm repository contains queries to produce output that the LangSmith UI does not currently support directly (e.g. obtaining trace counts for multiple organizations in a single query).

This command takes a PostgreSQL connection string that contains an embedded name and password (which can be passed in from a call to a secrets manager) and executes a query from an input file. The following example uses the pg_get_historic_trace_counts_daily.sql input file in the support_queries/postgres directory.

Prerequisites

Ensure you have the following tools/items ready.

1. kubectl

   • https://kubernetes.io/docs/tasks/tools/

2. PostgreSQL client

   • https://www.postgresql.org/download/

3. PostgreSQL database connection:

   • Host
   • Port
   • Username
     • If using the bundled version, this is postgres
   • Password
     • If using the bundled version, this is postgres
   • Database name
     • If using the bundled version, this is postgres

4. Connectivity to the PostgreSQL database from the machine you will be running the migration script on.

   • If you are using the bundled version, you may need to port forward the postgresql service to your local machine.
   • Run kubectl port-forward svc/langsmith-postgres 5432:5432 to port forward the postgresql service to your local machine.

5. The script to run a support query

   • Download the PostgreSQL support query script

Running the query script

Run the following command to run the desired query:

sh run_support_query_pg.sh <postgres_url> --input path/to/query.sql

For example, if you are using the bundled version with port-forwarding, the command might look like:

sh run_support_query_pg.sh "postgres://postgres:postgres@localhost:5432/postgres" --input support_queries/postgres/pg_get_historic_trace_counts_daily.sql

which will output the count of daily traces by workspace ID and organization ID. To extract this to a file add the flag --output path/to/file.csv

Export usage data

All export methods produce the same data: LangSmith trace counts, LangSmith Deployments node usage, and Fleet run counts across all workspaces and organizations.

The UI and API exports require both of the following:

• The organization:manage permission.
• The caller’s email must be listed in USAGE_EXPORT_ADMIN_EMAILS, or ORG_ADMINS_INSTALLATION_USAGE_EXPORT_ENABLED must be set to true.

USAGE_EXPORT_ADMIN_EMAILS='["[e-mail]", "[e-mail]"]'
ORG_ADMINS_INSTALLATION_USAGE_EXPORT_ENABLED=true

Export from the UI (recommended)

1. Navigate to Settings > Usage and billing > Usage export.
2. Click Export usage data.
3. A ZIP file containing all usage data will download.

Export via API

If you prefer to export usage data programmatically, you can call the export API endpoint directly.

curl -OJ \
  -H "X-API-Key: <your_api_key>" \
  https://<langsmith_url>/api/v1/orgs/current/usage/backfill-export

Export via SQL scripts

You can also run SQL scripts directly against your PostgreSQL database to export usage data. This requires database access credentials — no application-level permissions apply.

To export trace usage (requires Helm chart version 0.11.4 or later):

sh run_support_query_pg.sh <postgres_url> \
  --input support_queries/postgres/pg_usage_traces_full_export.sql \
  --output ls_export.csv

To export node usage (requires Helm chart version 0.11.4 or later):

sh run_support_query_pg.sh <postgres_url> \
  --input support_queries/postgres/pg_usage_nodes_full_export.sql \
  --output lgp_export.csv

To export Fleet run counts (requires Helm chart version 0.13.25 or later):

sh run_support_query_pg.sh <postgres_url> \
  --input support_queries/postgres/pg_usage_agent_builder_full_export.sql \
  --output ab_export.csv

To export usage snapshots (daily entity counts such as workspaces, projects, datasets, prompts, and active users):

sh run_support_query_pg.sh <postgres_url> \
  --input support_queries/postgres/pg_usage_snapshots_full_export.sql \
  --output usage_snapshots_export.csv

---

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.

Edit this page on GitHub or file an issue.