Connect OpenAI billing

Create a separate organization Admin API key in the OpenAI platform. A project inference key is for model calls and cannot replace the reporting credential. The Token Observe connector needs no additional configuration fields.

Create and confirm your Token Observe account, sign in and create a workspace. In Connect billing, choose the provider, give the connection a useful name and enter its reporting credential. Enter credentials only in the hosted portal’s masked field, never in a chat, worksheet or support message.

Choose Connect and collect costs. A saved credential can still have a failed or partial collection. Open Coverage details and check source status, reporting window, last successful collection and any reason for incomplete evidence. Use Collect latest 30 days after fixing permissions or configuration.

The connector reads the OpenAI Organization Costs API. It groups daily costs by project and line item and preserves the reported currency. ChatGPT subscriptions and historical prompts are outside this connection.

Keep the provider cost total separate from a final invoice, tax, credits and gateway estimates. The cost feed describes the provider account; a project identifier does not identify your internal team or named owner automatically.

The collectors bound a run to 31 completed UTC days, 20 requests or pages and 10,000 rows. A partial or failed collection keeps previous evidence with its status. Missing evidence remains unknown. One billing connection per provider per workspace prevents overlapping account connections.

To replace a credential, choose Rotate credential and confirm it accesses the same account. Changing accounts requires disconnecting first; disconnecting removes that connection’s retained costs. Scheduled collection stops when the trial expires. Retained authorized reads and exports remain subject to the shared traffic quota.