AI
OpenAI integrations that do a specific job and stay under your control
OpenAI offers capable language models through a well-documented API, along with tools for structured output and working with images and audio.
Why we use it
Why we build with OpenAI
OpenAI offers capable language models through a well-documented API, along with tools for structured output and working with images and audio. We use it for narrow, useful tasks: sorting incoming messages, extracting fields from documents, drafting replies for a person to approve, and answering questions from your own content. We prefer small features that save real time over open-ended chatbots.
What we build
What we build with OpenAI
Document and message extraction
Pulling names, dates, amounts, and other fields out of emails, PDFs, and forms into structured records, with a review screen for anything uncertain.
Drafting assistants
Suggested replies, summaries, and descriptions that staff can edit and approve before anything is sent.
Classification and routing
Tagging and sorting requests so they reach the right person or queue faster.
Search over your own content
Question-and-answer tools grounded in your documents, with links back to the sources.
How it fits our stack
What we pair it with
We call OpenAI from a Node.js or Python service, store results and logs in PostgreSQL, and keep API keys in encrypted configuration. Because the model is behind a thin layer in our code, we can swap or add providers, including Anthropic, without redoing the application.
Honest advice
When something else is better
If the task is rule-based, ordinary code is cheaper, faster, and more predictable than a model. If your data cannot leave your own environment, we would discuss options before using any hosted model API. Where Anthropic's models perform better on your particular task, we will recommend that instead, and we test both on your real examples where it matters.
FAQ
Frequently asked questions
Can AI make mistakes?
Yes. That is why we design human review into workflows where errors would matter, and why we log inputs and outputs so problems can be investigated.
Will our data be used to train models?
Provider policies differ by plan and change over time. We review the current terms for the API plan being used and discuss them with you before sending business data.
Do I need a chatbot?
Often not. A focused feature, such as extracting data from an invoice, can save more time than a general chat window. We help you pick the use case that pays off.
What does it cost to run?
Usage-based fees depend on volume and the model chosen. We estimate it from your expected workload and design to keep calls efficient.
Next step
Thinking about OpenAI for your project?
Tell us what you want to build and we will say plainly whether it is the right fit.
- Fixed scope
- Weekly demos
- You own the code