Spend over time
Daily spend, not the month-to-date total. A total only grows; a daily line shows the day something changed. After a few weeks you know each use's usual range, and a surprise is measured against that, not against the budget.
Guide · OpenAI API
Your OpenAI API spend,explained.
The few numbers worth watching, what OpenAI's own alerts and limits do, and how to catch a surprise before it reaches the invoice. Written for the person who gets asked why the bill moved.
Updated October 1, 2026 · About a seven-minute read · OpenAI is the first provider in Braid's pilot
What to watch
Four numbers, not forty.
OpenAI's usage dashboard shows a great deal. For catching surprises, most of it reduces to four questions, and the first one carries the other three.
Where it comes from
OpenAI breaks usage down by project, API key and model. Those are billing names, not business ones, so the useful habit is one project per use and one key per deployment, named for the work. Then a change points at something you recognise.
Unexpected changes
A weekday-sized weekend. A model nobody chose. A key you don't recognise. If spend and request volume move differently, check the model and token usage before drawing conclusions.
Budget versus actual
What's spent so far this month, what's left, and where the month lands at the current pace. The pace is the part that warns you early. The total only confirms it afterwards.
What a surprise looks like
A step, not a slope.
A sudden increase is easier to spot on a daily chart than in a month-to-date total. The daily view shows when spend changed, so you can check what happened that day.
The chart is the approximate shape of the incident that led to Braid. A bug kept calling an OpenAI endpoint for almost two weeks. Usual spend for the period would have been roughly $20 to $30. The loop drove daily spend above $90 and cost $1,051.96 before it was stopped. Finding the responsible project, key and model afterwards took real work in a dashboard organised around billing.
Daily spend, approximate
Illustrative shape, based on the founder's incident
Usual, about $2 a day
Day 1Day 14
OpenAI's built-in controls
Use what OpenAI gives you.
OpenAI's spend controls are a useful first line, and nothing further down this page replaces them. At the time of writing they cover the following.
What they do
- Organization budget and alerts. Set a monthly organization spend amount and alerts. A separate hard limit can stop affected requests when tracked spend reaches the configured amount.
- Project budgets. Projects can have their own spend alerts and optional hard limits. A hard limit can stop affected requests, though enforcement is not instantaneous.
- Usage breakdowns. OpenAI's usage exports can group activity by project, API key and model. Cost exports can show daily amounts by project or line item.
- Rate limits. Request and token rate limits control throughput. They can slow unexpected traffic, but they are not a dollar cap.
Where they stop
- They're organised by billing. A threshold alert tells you a spend amount was reached. Your team still needs to connect the provider project to a business use and owner.
- They're per provider. Add a second provider and you have two sets of limits, two dashboards and two ideas of what a project is.
- They don't know what's usual. A fixed threshold is either high enough to miss a doubling or low enough to fire every month-end.
- They don't carry context. A new key with no name is just a new key. Who made it and what it's for lives somewhere else, if anywhere.
A weekly routine
Five habits that catchmost of it.
A disciplined team can run all of this with OpenAI alone. The cost is the discipline: the mapping lives in someone's head, the Monday look gets skipped, and the same questions get answered from scratch each time.
Name your uses, not your teams.
One OpenAI project per thing the company does with AI, and one key per deployment. “Support assistant, production” beats “Backend”.
Write down an owner.
For each project, one person who would know whether a change is expected. Keep it where the next person can find it.
Set limits in layers.
An organization budget with an alert well below it, and a project limit sized to each use. Hard-stop only what can tolerate rejected requests.
Look at daily spend once a week.
By project, against the last few weeks. Five minutes on Monday finds a Tuesday step before it's two weeks old.
When something moves, answer three questions.
What changed, what is it for, who should look. If any of them takes more than a few minutes, that's the gap to close.
Where Braid comes in
Built for the three questions.
What changed, what is it for, and who should look. Braid reads usage and cost from OpenAI through read-only access and adds the context the provider cannot have: what each project and key is for, and who owns it.
A map of what the spend is for
Implementations are the uses your company recognises, like a support assistant or a document pipeline. Each one maps the OpenAI projects and keys behind it, with an owner, a team and the models you expect. A key that isn't mapped shows up as something to look at.
A weekly report people can read
Every Monday, an email about last week: what was spent by Implementation, what changed and why, what needs attention, and where the month is landing against budget. Plain language, ready to forward.
Alerts with the answer attached
Braid flags material spend changes and selected budget thresholds in the app and by email, with the affected Implementation and source details when the data supports them. Budgets can be set for the organization and for each Implementation.
OpenAI is the first provider, included in the pilot. Anthropic is planned next, followed by Google Gemini. Braid never changes your OpenAI account, and it never sees prompts or responses.
Why Braid exists
One looping script ranfor almost two weeks.
On our founder's team, a bug kept calling an OpenAI endpoint. It cost $1,051.96 in avoidable spend before anyone stopped it, and tracing the responsible key, project and model took substantial work afterwards. This happened before Braid existed, so we won't claim Braid would have caught it. We will say that the questions which took days to answer are the ones Braid answers first.
$1,051.96 avoidable spend, before Braid
Questions
For anything else, ask the founder directly at hello@getbraid.app.
- Does Braid replace OpenAI's spend limits?
- No. Keep the provider controls that fit your team. Braid reads usage and cost; it does not stop requests or change your OpenAI account.
- What access does Braid need?
- An admin key for your OpenAI organization, used only to read usage and cost. Braid never sees your prompts or responses, and you can revoke the key from OpenAI at any time.
- Which providers are in the pilot?
- OpenAI is included. Anthropic is planned next, followed by Google Gemini.
- What happens when I ask to start?
- Leave your work email on the form below. Justin, Braid's founder, follows up personally before any payment or provider connection. Founding pilots are $250 for 30 days.
Founding pilots · $250 for 30 days
Give every OpenAI projectan owner and a clear report.
During the 30-day pilot, we'll connect OpenAI, map your Implementations with the founder, prepare weekly reports, and set up budgets and spend alerts together.
Leave your work email.
Justin, Braid's founder, follows up personally. Nothing is paid and nothing is connected until you've talked.
Read-only access. Braid never sees your prompts or responses.