OpenAI Admin API (API Platform)
Analytics
OpenAI Admin API (API Platform) data, published as OpenAI Admin API (API Platform) publishes it. Ingest delivers this data to a data warehouse you own as 7 tables, and updates it on the schedule you choose. There is no code to write and nothing to install.
What you get
| Table | Each row is |
|---|---|
costscosts | Daily organization cost in USD, one row per UTC day, split by project, line item, API key, user and API source; that day's money lines are in the costs__results child table, each with the amount in its own currency and the metered quantity. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no spend still has a row, with no money lines. |
project_service_accountsproject_service_accounts | The service accounts in each project (transformer child of projects), with name, project role and when each was created, keyed by the account and the project. Each project's service accounts are re-read once a day. |
project_usersproject_users | The members of each project (transformer child of projects), with name, email, project role and when each was added. A member of two projects is two rows, keyed by the member and the project. Each project's membership is re-read once a day. |
projectsprojects | The organization's projects, archived ones included, with name, status, data residency and when each was created and archived. Read in full on every sync. |
usage_completionsusage_completions | Daily text generation usage (Chat Completions, Responses and Batch), one row per UTC day, split by project, user, API key, model, batch flag, service tier and API source; that day's usage lines are in the usage_completions__results child table with input, cached, cache write, uncached, output, audio and image tokens and the request count. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no usage still has a row, with no usage lines. |
usage_embeddingsusage_embeddings | Daily embeddings usage, one row per UTC day, split by project, user, API key and model; that day's usage lines are in the usage_embeddings__results child table with input tokens and the request count. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no embeddings traffic still has a row, with no usage lines. |
usersusers | The organization's members, with name, email, organization role and when each was added. Read in full on every sync. |
How it works
- Choose OpenAI Admin API (API Platform). Connect your OpenAI Admin API (API Platform) account once, from Ingest.
- Pick where it goes. BigQuery, Snowflake, Databricks, Redshift and 11 more, in an account you own.
- Pick a schedule. Daily is the default, and weekly and monthly are there too.
- Ingest runs it from then on. If a run fails, your dashboard shows which table and why.
Questions
- Do I need an account with OpenAI Admin API (API Platform)?
- Yes. You connect your own OpenAI Admin API (API Platform) account once, from Ingest, and the data comes from what that account can see.
- Which warehouses can OpenAI Admin API (API Platform) data go to?
- Amazon S3, Google Cloud Storage, Azure Blob Storage, Apache Iceberg, Ingest Managed Lakehouse, MotherDuck, Postgres, Amazon Athena, Databricks, Redshift, BigQuery, ClickHouse, ObsessionDB, MySQL / generic SQL and Snowflake, in an account you own.
- How fresh is the data?
- As fresh as your schedule: daily by default, or weekly or monthly. Schedules as often as every 20 minutes are on the Enterprise plan.
- Will running it again create duplicate rows?
- No. Each row is stored once, by its key: when a run reads a row that is already there, it replaces that copy instead of adding a second one.
- Do I need to write code?
- No. Ingest builds, runs and maintains the connector. Someone with access to your warehouse connects it once, following a short guide.
For your data team: technical reference
Connector openai, version 0.1.0, a REST source, status beta, authentication bearer.
Vendor documentation: https://developers.openai.com/api/docs/guides/admin-apis
Configuration
No configuration required.
Resources
| Resource | Status | Plan | Write | Primary key | Overrides |
|---|---|---|---|---|---|
costsDaily organization cost in USD, one row per UTC day, split by project, line item, API key, user and API source; that day's money lines are in the costs__results child table, each with the amount in its own currency and the metered quantity. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no spend still has a row, with no money lines. | beta | window | merge | start_time | lookback_days |
project_service_accountsThe service accounts in each project (transformer child of projects), with name, project role and when each was created, keyed by the account and the project. Each project's service accounts are re-read once a day. | beta | parent_child | merge | id, _ingest_parent_key | - |
project_usersThe members of each project (transformer child of projects), with name, email, project role and when each was added. A member of two projects is two rows, keyed by the member and the project. Each project's membership is re-read once a day. | beta | parent_child | merge | id, _ingest_parent_key | - |
projectsThe organization's projects, archived ones included, with name, status, data residency and when each was created and archived. Read in full on every sync. | beta | full | merge | id | - |
usage_completionsDaily text generation usage (Chat Completions, Responses and Batch), one row per UTC day, split by project, user, API key, model, batch flag, service tier and API source; that day's usage lines are in the usage_completions__results child table with input, cached, cache write, uncached, output, audio and image tokens and the request count. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no usage still has a row, with no usage lines. | beta | window | merge | start_time | lookback_days |
usage_embeddingsDaily embeddings usage, one row per UTC day, split by project, user, API key and model; that day's usage lines are in the usage_embeddings__results child table with input tokens and the request count. OpenAI keeps one year of reporting data, so each sync re-reads the last 7 days by default (the lookback can be raised to 360 days at deploy) and re-read days merge onto their own start time; days older than the lookback are not revisited unless it is raised. A day with no embeddings traffic still has a row, with no usage lines. | beta | window | merge | start_time | lookback_days |
usersThe organization's members, with name, email, organization role and when each was added. Read in full on every sync. | beta | full | merge | id | - |
Where this lands, and what else connects
Setting up the destination is its own short guide, one per warehouse or lake: Amazon S3, Google Cloud Storage, Azure Blob Storage, Apache Iceberg, Ingest Managed Lakehouse, MotherDuck, Postgres, Amazon Athena, Databricks, Redshift, BigQuery, ClickHouse, ObsessionDB, MySQL / generic SQL, Snowflake.
Get OpenAI Admin API (API Platform) data into your warehouse
Last updated .