Introduction
Your own AI model, trained on your data, in a few clicks.
typestrict trains a small decision model on your data: it reads a ticket, a message or a document and answers with a structured, typed decision (a label, a yes or no, a score), the same answer shape as Jev, the market reference model. If your model does not beat Jev on examples it never saw during training, the credit comes back (unless you chose to keep the model anyway, see Refund calculation).
How a training goes
Every training goes through five steps, in the dashboard, the API and the assistants alike:
- Data: you send files, connect Google Drive or OneDrive, or start from a public dataset. Personal data is masked before anything else reads it.
- Review: our agent organizes the data into training examples. You check the proposal and adjust it.
- Base model: you pick the model to train on, cleared for commercial use, and confirm the cost.
- Training: we train it and measure it against Jev.
- Result: you download the model (ONNX and weights) and use it anywhere, without a GPU.
Where to start
Quickstart
An API key and your first calls.
Authentication
API keys and how to send them.
Trainings
Steps, next actions and pages of results.
Refund calculation
How we decide whether your model beats Jev.
Base models
What your model starts from, and on what data.
AI assistants
ChatGPT, Claude, Cursor, Codex, Copilot, Antigravity, Devin.
Run it anywhere
Docker, Cloud Run, Lambda, Firebase, Kubernetes.