OpenAI integration
LiveOpenAI cost tracking, per project and per model
The OpenAI dashboard tells you what you spent. CloudQuell puts that spend where the rest of your bill lives — per project and per model, next to AWS, Anthropic, and Snowflake — with anomaly baselines watching it and allocation rules routing it to the same cost centers as everything else.
It connects with a read-only admin key that can't run models or modify your account, reads OpenAI's Costs and Usage APIs on a daily schedule, and backfills up to a year of history on connect.
What CloudQuell ingests
Dollars from the Costs API — the billing source of truth — joined to token counts from the Usage API, so cost and usage sit on the same row.
- Cost, per model per day
- Daily spend by model from the Costs API, the same number that reconciles to your OpenAI invoice.
- Token usage
- Input, cached input, output, and audio tokens plus request counts from the Usage API — so you can see the input:output shape that actually drives a bill.
- Per-project attribution
- Spend broken out by OpenAI Project — the free attribution you get from splitting keys by product surface or environment.
- Service tier
- Batch versus standard traffic, so the roughly 50%-cheaper batch path is visible as an adoption opportunity, not a mystery discount.
- Per-key usage
- Token and request attribution per API key. OpenAI exposes no per-key dollars, so this dimension is usage, not cost — CloudQuell shows it as usage and doesn't invent a number.
What the data looks like normalized
Per-project, per-model OpenAI spend lands in the same ledger as your cloud bill, on the same schema as an EC2 line item.
| Normalized field | OpenAI source | Example |
|---|---|---|
| Date | Costs API day | 2026-08-04 |
| Provider | constant | OpenAI |
| Model | Usage API | gpt-5.4 |
| Scope | Project | doc-extraction-prod |
| Cost (USD) | Costs API | $214.80 |
| Usage | token types | 41.2M in · 6.1M out |
| Attribution | project / key | key: extraction-svc (usage) |
How it connects
One read-only admin key. It can read your usage and costs and nothing else.
- Credential
- A read-only OpenAI Admin API key (sk-admin-…). Project keys won't work by design — only the admin key can read org-wide costs. A read-only admin key can never run a model, incur usage, or modify your organization.
- Access scope
- Read-only. CloudQuell validates the key live against OpenAI on connect, stores it in AWS Secrets Manager (never in its database), and lets you rotate it any time.
- Prerequisites
- An OpenAI Organization Owner to mint the admin key, and a CloudQuell admin to add it.
- Time to first data
- The first ingest starts right away and backfills up to 12 months. Sync runs daily and re-pulls a trailing three-day window so late-arriving usage reconciles.
- What's out of scope
- Azure OpenAI (that's the Azure connector), AWS Bedrock (the AWS connector), and ChatGPT Team/Enterprise seats — this is the OpenAI API platform bill.
What it unlocks
OpenAI spend stops being a second tab and becomes part of the same operating cadence as your cloud bill.
One ledger
Per-project and per-model OpenAI spend in the same dashboard, cost centers, and reports as AWS, Anthropic, and Snowflake — one number for what a product costs to run.
Anomaly alerts
Per-model baselines catch a retry loop or runaway agent the afternoon it starts — not on next month's invoice — routed to Slack, Teams, or email.
AI budget alerts →Allocation
Route OpenAI cost to the same teams and cost centers as the rest of your spend, so the AI line has an owner like every other dollar.
Batch headroom
See how much traffic could move to the batch tier's ~50% discount, quantified against your actual usage rather than guessed.
The how-to companion
This page is about putting OpenAI on one ledger with a tool. If you're building attribution yourself first — projects, an admin-key pull, per-feature logging — our full DIY playbook covers all three levels and where each one stops.
How to track OpenAI API costs →Common questions
- How does CloudQuell track OpenAI costs?
- It connects with a read-only admin key and reads OpenAI's Costs API for dollars and Usage API for tokens, daily. Spend lands per project, per model, and per token type in the same dashboard as your cloud, Anthropic, and Snowflake bill, with anomaly baselines and allocation rules applied to it like any other source.
- Is the OpenAI admin key safe to share?
- The key is read-only: it can read usage and costs but can't run a model, incur spend, or change your organization. CloudQuell validates it on connect, stores it in AWS Secrets Manager rather than its database, and lets you rotate or revoke it any time. Project keys (sk-proj-…) deliberately won't work — only the admin key can read org-wide cost.
- Can I see cost per project or per feature?
- Per project and per model, yes — that's what OpenAI's APIs expose and what CloudQuell ingests. Per feature or per customer is finer than any provider's API reaches; you get there by tagging at the call site, which our OpenAI cost-tracking playbook walks through. CloudQuell then puts the project-level total on the same ledger as everything else.
- Does this replace an observability tool like Langfuse?
- No — different job. Observability tools trace individual requests (which prompt cost $3). CloudQuell is the finance side: every AI dollar next to every cloud dollar, allocated and alarmed. Plenty of teams run both; our writing on LLM cost management tools explains the split.
Free under $10K/month of tracked spend. Read-only admin key, per-project and per-model, flat monthly price.