Barndoor Automations now supports OpenAI models
OpenAI models now sit alongside every other model in Barndoor Automations.
Jay Parisi · · 3 min read
TL;DR
- Add GPT-5.6 Luna or GPT-5.6 Terra to Automations from the Add model dialog in a few clicks
- Bring your own OpenAI key to run on your own account and bill, or let Automations host the model with no credentials to manage
OpenAI models now sit alongside every other model in Automations. Open Add model, choose the OpenAI family, pick a model, and decide how you want to run it: with your own OpenAI key, or hosted by Automations with no credentials at all.
Your Blueprints don't change either way. The model is a setting you choose at run time, not something baked into the workflow, so the sessions, tool scopes and output schemas you've already written run as they are.
Two ways to run it
When you add an OpenAI model, the Provisioning setting gives you two options.
Bring your own key
Your Blueprints call OpenAI under your own account, on your contract and your bill. Paste an API key into the dialog, or pick one you've already saved in your secrets so it never has to be pasted again.
Choose this when you already have an OpenAI agreement, need the calls to sit under your organisation's account, or want usage to show up in your own OpenAI dashboard.
Hosted by Barndoor
Automations provides the model for you, so there's no API key to create, store or rotate. Choose it and you can run a Blueprint on an OpenAI model straight away.
Choose this when you want to try a model without setting up an OpenAI account first, or when a team just needs a working model and not another credential to manage.
Adding an OpenAI model
- Open Add model.
- Give it a Name your team will recognise, like "Production GPT".
- Set Family to OpenAI ChatGPT.
- Pick a Model: GPT-5.6 Luna or GPT-5.6 Terra.
- Choose Provisioning: Hosted by Barndoor, or Bring your own key and add your key.
- Set Visibility. Personal keeps the model to yourself.
- Leave Enabled on to offer the model when you run a Blueprint. Turn it off to keep the model configured without offering it.
Then select Add model, and it's ready to use.
The Blueprint stays the same
Switching models is where a lot of AI workflows quietly break: a different model phrases its answers differently, and whatever reads those answers falls over.
A Blueprint is built to absorb that. Each session declares what it can see, which tools it may call, and the schema its answer must match, and Frags checks every answer against that schema before the next step or the caller can use it. Moving a Blueprint to an OpenAI model changes who does the reasoning, not the contract the reasoning has to meet.
That also makes comparing models straightforward. Run the same Blueprint on a different model and compare the typed output, field by field.
Running OpenAI locally with Frags
If you run Blueprints on your own machine with the open-source Frags CLI, OpenAI is supported there too. Point Frags at it in your .env:
CHATGPT_BASE_URL also accepts any OpenAI-compatible endpoint. Running Frags Locally covers the full setup.
A note on "experimental"
The OpenAI family is labelled experimental in the dialog for now. We'd like to hear what you find as you run it on real Blueprints — especially anything