DecisionTune
Train your own decision model.
Coming soon
Turn a spreadsheet of past decisions into a fast, private model that you own. DecisionTune Studio trains it on your computer. You do not need ML experience.
- 1 · Supply examplesA spreadsheet of past tickets, requests or reviews. Each row has the decision that a person made.
- 2 · Train on your computerStudio trains the model on your Mac or laptop. Your data stays there.
- 3 · Use it foreverThe same Python, command line and HTTP calls, with your categories. It works offline.
Studio pricing coming soon. DecisionTune 1.0 stays free under Apache-2.0.
Why Studio
- Private by default.Studio trains on your computer, and the model runs on your computer. Medical, legal and financial records do not go to a cloud API.
- Fast: about 10 ms.On a Mac with Apple silicon, the model uses the GPU through MLX. A short decision takes approximately 10 ms.* There is no network trip.
- Yours forever.Your model is a file on your computer. It works offline. No API can change its price or shut it down.
- No ML team needed.Supply a spreadsheet of past decisions. Studio prepares the data, trains the model and tests it. Training on a Mac takes longer than on a cloud GPU, but your data stays with you.
- Proof before you ship.Studio tests the model on examples that it did not train on. It shows the accuracy for each category.
- Small and free to run.395M parameters and 1.7 GB of GPU memory on a Mac.* No GPU server and no fee for each decision.
* We measured on an M5 Pro Mac with the MLX backend. On a CPU only, a short decision takes approximately 65 ms and uses up to 4.5 GB of memory.
Who it is for
- Teams whose data must stay localHealth and medical records, legal and confidential documents, financial records. Any data that must not go to a cloud service.
- Developers of AI agentsSelect the correct tool for each request, without a cloud call.
- Support and operations teamsSend each ticket to the correct queue. Find the requests that need a person.