Compute
Build and use AI models by telling Orchard Studio what you need.
Describe the model you want to create. Orchard finds a starting model and suitable data, prepares the run, trains and evaluates the model, and can deploy it for your app. You only approve the estimate before paid work begins.
Ask in Orchard Studio
- "Build a model that can identify damaged products from photos."
- "Use the sales data I uploaded to train a demand forecasting model."
- "Deploy this model, keep it private, and give me an API for my app."
From idea to working model
- Describe: tell Orchard in chat what the model should do and how you want to use it.
- Prepare: Orchard uses the files you share or recommends a model and data, then preprocesses and validates the data for training.
- Approve: review the estimated cost and time, then set the maximum cost and runtime.
- Build: Orchard trains and evaluates the model, saves the result, and deploys it when requested.
The environment is handled for you
You do not need to configure a local GPU or set up a separate training environment. Orchard prepares the compatible software and hardware for each model before the run begins.
If the environment is not compatible, Orchard stops before training and explains what needs to change instead of wasting the approved budget.
Saved models and data
Orchard keeps the inputs, logs, reports, results, and model files needed to continue your work. Saved data is charged from the first byte at $0.02 per GB per month, based on how long it is stored. Delete it when you no longer need it to stop future storage charges.
Private by default 🔒
Saved data and deployment packages are encrypted, including while they are transferred. Orchard does not retain inference inputs or responses after processing them. Deployed models stay private unless you choose to change their access.
Deploy without retraining
Orchard checks the completed model, builds the managed deployment, and repeats its sample request before reporting Ready. The managed build and first smoke test do not count toward your usage.
Once Ready, call the model from your app's backend with a project API key from Model details. Access can stay private, be limited to your app's signed-in users, or be made public.
One cost limit
Saved data, approved runs, and deployed inference are shown separately in Billing and count toward one Billing Usage cap.
Prices vary with live availability and the work your model needs. No paid work starts until you approve the estimate, maximum cost, and runtime. Deployed models cost nothing while no replica is running.
What to know
- Cloud runs, managed deployment, and inference require an active Builder plan and Billing Usage cap.
- Every paid run requires a current estimate and your explicit maximum-cost approval.
- Recommendations and reports do not guarantee model quality, safety, or license suitability.
- Do not include passwords, tokens, or other secrets in the files you provide.