OpenAI Platform setup - Configview Support

ConfigView pulls your OpenAI API organization — spend, token usage, projects, members, and API keys — through an Admin API key created in the OpenAI platform console. You will end up with 1 secret in ConfigView (OPENAI_ADMIN_KEY) when setup is complete.

This is not the same as ChatGPT. OpenAI splits into two products with two separate consoles, two key types and two ConfigView apps. This page covers platform.openai.com — the developer API you are billed for per token. If you want to inventory ChatGPT Enterprise seats and workspace activity, that is the OpenAI ChatGPT integration and a different credential entirely. Most organizations that use both will want both apps enabled.

Step 1: Create the Admin API key

Sign in to the OpenAI platform console as an organization Owner
Go to Settings → Organization → Admin keys
Create a new admin key and name it ConfigView
Copy the key when it is shown. You will not be able to see it again.

An admin key is not a project API key. The usual sk-proj-... key you use to call models will authenticate and then return 401 or empty results on every organization endpoint. If ConfigView reports “authenticated but no data”, the wrong key type is the first thing to check.

Step 2: Add the secret to ConfigView

Go to your ConfigView dashboard: https://{companyname}.configview.com/admin/secret/
Click Add Secret
Secret name: OPENAI_ADMIN_KEY
Secret value: Paste the admin key from Step 1
Click Save

Step 3: Enable the OpenAI Platform scripts in ConfigView

Go to: https://{companyname}.configview.com/admin/cron/
You should see OpenAI Platform in the list of available apps
Select the scripts you want to run:

Spend and usage

Script	Notes
Cost Breakdown	Spend by line item and project. This is the one that answers “what are we paying for”.
Completions Usage	Token usage for chat/completions, split by model, project, user and API key. Includes cached and audio token counts.
Embeddings Usage	Token usage for embeddings.
Images Usage	Image generations by model, size and source.
Audio Speeches Usage	Characters synthesised.
Audio Transcriptions Usage	Seconds transcribed.
Moderations Usage	Moderation calls.
Vector Stores Usage	Bytes stored per project.
Code Interpreter Usage	Code interpreter sessions per project.

Access and governance

Script	Notes
Organization Users	Everyone in the org, with their org-level role.
Pending Invites	Invitations not yet accepted, with expiry.
Projects	Every project, including archived ones. Start here; three scripts depend on it.
Project Membership	Who is in which project, and with what role. Depends on Projects.
Project Service Accounts	Non-human members of each project. Depends on Projects.
Admin API Keys	Organization-level admin keys, with owner and last-used date.
Project API Keys	Per-project keys, with owner and last-used date. Depends on Projects.
Audit Logs	Org audit events — who changed what, from where.

Click Save

Note: Project Membership, Project Service Accounts and Project API Keys all depend on Projects. ConfigView wires these dependencies automatically — Projects always runs first.

Step 4: Verify

Go to: https://{companyname}.configview.com/admin/cron/
Run the OpenAI Platform health check

If a check fails:

401 — The key is wrong, revoked, or is a project key rather than an admin key. See the note in Step 1.
Usage tables are empty but cost is populated — Usage endpoints are time-bucketed. A freshly created org, or one with no API traffic in the window queried, legitimately returns nothing.
Audit Logs returns 403 — Audit logging is not available on every OpenAI plan. If your org does not have it, leave that script disabled.

Data Tables

Once the scripts run, the corresponding tables are created in your database. All tables include a run_at column for historical tracking.

Table	Source (`/v1/organization/...`)	Key Columns
`openai_cost`	`costs`	starting_at, ending_at, line_item, project_id, amount, currency
`openai_usage_completions`	`usage/completions`	starting_at, project_id, user_id, api_key_id, model, input_tokens, output_tokens, input_cached_tokens, input_audio_tokens
`openai_usage_embeddings`	`usage/embeddings`	starting_at, project_id, model, input_tokens, num_model_requests
`openai_usage_images`	`usage/images`	starting_at, project_id, model, size, source, images
`openai_usage_audio_speeches`	`usage/audio_speeches`	starting_at, project_id, model, characters
`openai_usage_audio_transcriptions`	`usage/audio_transcriptions`	starting_at, project_id, model, seconds
`openai_usage_moderations`	`usage/moderations`	starting_at, project_id, model, input_tokens
`openai_usage_vector_stores`	`usage/vector_stores`	starting_at, project_id, usage_bytes
`openai_usage_code_interpreter`	`usage/code_interpreter_sessions`	starting_at, project_id, num_sessions
`openai_users`	`users`	user_id, email, name, role, added_at
`openai_invites`	`invites`	invite_id, email, role, status, invited_at, expires_at, accepted_at
`openai_projects`	`projects`	project_id, name, status, created_at, archived_at
`openai_project_users`	`projects/{id}/users`	project_id, user_id, email, role, added_at
`openai_project_service_accounts`	`projects/{id}/service_accounts`	project_id, service_account_id, name, role, created_at
`openai_admin_api_keys`	`admin_api_keys`	api_key_id, name, redacted_value, created_at, last_used_at, owner_email
`openai_project_api_keys`	`projects/{id}/api_keys`	project_id, api_key_id, name, redacted_value, created_at, last_used_at, owner_email
`openai_audit_logs`	`audit_logs`	audit_log_id, event_type, effective_at, actor_email, actor_ip, project_id, details

Things worth knowing

Cost and usage are separate questions, and they do not reconcile line for line. openai_cost is what OpenAI bills; the openai_usage_* tables are what was consumed. Discounts, minimums, and free tiers live in cost and not in usage, so summing tokens and multiplying by list price will not reproduce your invoice. Use cost for finance and usage for attribution. last_used_at on API keys is the highest-value column here. A key that has never been used, or has not been used in months, is the cleanest signal for revocation — and both key tables carry the owner, so you know who to ask before pulling it. Usage rows are pre-aggregated buckets, not individual calls. Each row covers a time window (starting_at to ending_at) for a given model/project/user combination. You cannot get per-request detail from this API, so questions like “which exact call cost the most” are not answerable here. Archived projects still appear. openai_projects.status and archived_at tell you which are live. Filtering on them matters for seat and key reporting, or archived projects will inflate your counts.

​Step 1: Create the Admin API key

​Step 2: Add the secret to ConfigView

​Step 3: Enable the OpenAI Platform scripts in ConfigView

​Step 4: Verify

​Data Tables

​Things worth knowing