OpenAI spend hides across projects. SuperOrgs surfaces all of it.
GPT-4o in the product card. GPT-4.1 in an analytics side project. A dozen Assistants API builds nobody remembers standing up. Custom GPTs scattered across workspaces. SuperOrgs ingests the OpenAI Admin API, joins every project and every key to a human owner, and shows you the bill before the bill arrives.
What we surface from OpenAI.
Three layers of visibility pulled from the OpenAI Admin API, refreshed nightly, reconciled to the humans in your HRIS.
Every project and every key
OpenAI projects are a powerful isolation primitive that most orgs underuse. SuperOrgs enumerates every project, every service account, every API key, and every Assistant in your organization. The long tail of forgotten projects shows up within minutes of the connector going live.
Every custom GPT and every Assistant
Your builders launched an internal sales enablement GPT six weeks ago. A product manager stood up three Assistants for a launch and never decommissioned them. SuperOrgs treats each one as a first-class agent with an owner, a purpose, and a review cadence. No more hidden GPTs.
Every dollar, by model tier and project
GPT-4o, GPT-4.1, o3, o4-mini, the embeddings endpoint, moderation, Whisper, fine-tuned models. Every model tier has its own price. SuperOrgs breaks out usage by model, by project, by owner, and lets finance set a budget per tier rather than chasing a single aggregate number.
Live product sliceYour OpenAI agents, across the lifecycle.
Every OpenAI-backed agent in the kanban has a real owner and a real spend attribution. Drag to move an agent between recruiting, onboarding, active, review, and terminated. The same board is the one your product team uses in the live demo.
How it works
How the OpenAI connector works.
Four-step configuration. One admin key. Read-only scope. OpenAI never receives a single byte of prompt content or organizational metadata from SuperOrgs.
- 01
Mint an OpenAI admin API key
Generate an admin key from your OpenAI organization settings. The admin key has read access to projects, service accounts, user memberships, API keys, and usage. It cannot create Assistants, cannot send messages, and cannot mutate your OpenAI org. You remain the only writer.
- 02
We enumerate projects, Assistants, and keys
Within seconds SuperOrgs lists every project, every API key under each project, every Assistant, every custom GPT with a public share link, and every service account. Objects that span projects are cross-referenced so a single key is never double-counted against the organization total.
- 03
We attribute each agent to a person
OpenAI projects have members. Those members are emails. Emails match against Workday, BambooHR, Rippling, or Okta. SuperOrgs builds an ownership graph in which every agent has a primary owner, a backup reviewer, and a department. Off-boarded employees are flagged the moment their HRIS record flips to inactive.
- 04
We break spend out by model, project, and owner
OpenAI usage is rolled up per model tier per day in the Admin API. SuperOrgs applies your organization's negotiated pricing, normalizes across model generations, and produces a per-agent monthly cost that reconciles within three percent of the invoice you receive from OpenAI.
Cost intelligence
OpenAI spend, broken out the way FP&A asks for it.
$563
7 agents under management
$6,756
based on current monthly trajectory plus +8.2% MoM
$1,149
per employee, peer cohort median
A mid-market SaaS company with a GPT-4o coding assistant, a GPT-4.1 analytics pipeline, and three to five Assistants in production usually reports between 3.8 thousand dollars and 22 thousand dollars in monthly OpenAI spend. SuperOrgs plots your organization against that band with seasonality already baked into the forecast so a single busy quarter does not warp next year's budget.
Governance
OpenAI-specific governance, beyond what the platform ships with.
OpenAI gives you projects and usage. SuperOrgs adds the approval, audit, and anomaly detection layer OpenAI leaves to the customer.
Approval workflow for new OpenAI projects
Require a budget owner before a new project is stood up. Auto-approve projects under a monthly spend ceiling. Require a data-residency review when a project's members span regulated jurisdictions. Every approval is logged with who approved, when, and why.
Assistant and custom GPT inventory
Every Assistant and every shared custom GPT is indexed with its instructions, its tools, and its file attachments. When an Assistant is retired, SuperOrgs checks that the underlying files are unshared. When a custom GPT is modified, the audit log captures the diff.
Model-tier spend caps and rate limits
Set a monthly ceiling per model tier per project. GPT-4o can have one budget, o3 another, embeddings another. The policy engine pauses a project that overshoots its tier budget and routes the overage decision to the project's approver rather than silently letting the bill grow.
Anomaly detection on usage shape
A sales Assistant that suddenly consumes five times its usual context window. A coding project that switches from GPT-4o to o3 without a pricing review. SuperOrgs watches the usage shape, not just the total, and flags shifts your CFO will ask about before they notice.
FAQ
Questions teams running OpenAI ask us.
Does SuperOrgs support the OpenAI Projects primitive?
Yes, natively. OpenAI projects are a first-class entity in our data model. Projects carry their own budget, their own membership, and their own governance settings. SuperOrgs can also report on organizations that pre-date the projects primitive, treating the default project as the boundary.
Can SuperOrgs manage both API usage and ChatGPT Team seats?
Today we cover every surface that exposes an API or usage endpoint. That includes API keys, Assistants, fine-tuned models, and custom GPT share links. ChatGPT Team and Enterprise seat management is on the roadmap for the next release once OpenAI opens the user-level usage endpoints we need.
What about security and data handling?
SuperOrgs is SOC 2 Type II. All OpenAI admin keys are stored in a customer-scoped encrypted vault using per-tenant keys. We read only. We never send prompt content to OpenAI. The admin key lives in your tenant and can be rotated or revoked without any data loss on our side.
How does SuperOrgs compare OpenAI spend to our other model providers?
Every model provider is normalized to the same agent object in SuperOrgs. You can rank your Claude, GPT, Gemini, and Llama spend against one another per project or per owner. Switching costs and multi-vendor strategy decisions become legible because the comparison is like for like.
“Three Assistants that nobody remembered creating. Two projects that nobody remembered funding. SuperOrgs surfaced all of it inside a day and paid for itself in the first month of decommissioned spend.”
VP Engineering, Series B vertical SaaS company