Based in Downtown Brooklyn
OpenAI
OpenAI and ChatGPT integration
- We build with it
- We use it ourselves
OpenAI has the widest reach of any provider — most automation platforms and SaaS products integrate it first — which makes it the pragmatic choice when a workflow has to pass through tools you do not control. We build on the API, and set teams up on ChatGPT properly.
The relationship
How we work with OpenAI
- Build
We build with it
We build against the API: classification, extraction, drafting and search inside your own applications and automations, with spend capped and failures handled rather than retried forever.
- In-house
We use it ourselves
We use these models in our own work, so the advice on which one fits a task comes from having run them rather than from a comparison table.
Sound familiar?
OpenAI problems we fix
You don’t need to know the technical name for it. Tell us what’s happening, and we’ll work out what needs attention.
-
“The API spend is climbing and nobody set a limit.”
An endpoint that can call a paid model without a ceiling is a bill waiting to happen. We meter every route, cap it at a figure you agreed, and right-size the model to the task.
-
“You need the model inside a tool that already integrates OpenAI.”
This is where OpenAI’s reach wins — it is usually the first provider a platform supports. We wire it up through whatever the tool exposes, rather than building around it.
-
“The team is on ChatGPT but nothing is set up properly.”
We get accounts, access and licensing right, set out what should and should not be pasted into it, and show people the handful of uses that actually save them time.
-
“Nobody is sure what happens to data on the consumer tier.”
Consumer and business tiers differ materially on retention and training. We put the current terms in writing, and move work onto the tier that suits the data rather than the one somebody signed up for personally.
-
“A model you depend on is being deprecated.”
OpenAI retires models on a schedule. We keep the model behind a configuration setting and test against the replacement before the deadline rather than on the morning of it.
-
“Output is inconsistent between runs.”
Usually a prompt doing too many jobs at once, or no structure imposed on the response. We constrain the output format, add evaluations that catch a regression, and keep prompts in version control.
Something else with OpenAI? Tell us about it.
Where it fits
The services behind the OpenAI work
Claude
Google Gemini
Zapier
Make
n8nQuestions we get
OpenAI, answered plainly
Is ChatGPT the same as building on the API?
No, and conflating them is the most common confusion we see. ChatGPT is a product a person opens and uses. The API is what you build an automation on, so it runs without anybody opening anything. Getting a team onto ChatGPT is worth doing and we will help with it, but it is not automation and it will not do anything while nobody is at a desk.
Which model should we use?
Per task, not per company. The largest model everywhere is the most common and most expensive default: classification and extraction usually run fine on a smaller, cheaper, faster one, and the difference at volume is substantial. We size it to the work and keep the choice changeable.
Should we be worried about our data?
You should know the answer rather than worry about it. Business and consumer tiers differ on retention and on training, and the difference matters if client material is going into it. We put the current terms in writing before anything is built, and say which of your data would leave your systems at all.
What happens when OpenAI changes something?
Models get retired and prices change — that is the nature of building on somebody else’s platform. We keep the provider and model behind configuration rather than welded into the code, so a change is a change rather than a rebuild, and keeping up with it is part of the ongoing arrangement.
Start the conversation
Got a OpenAI headache?
Tell us what’s giving you trouble, or what you’d like to stop managing yourself. You don’t need a technical brief. A few sentences is a good place to start.
Thanks for reaching out.
We’ve received your message about OpenAI and will get in touch about next steps.