System One and System Two
Start with questions or a policy. Keep the model and send it with each execution.
aityx turns your questions or policy into an executable decision model. You keep that model and choose the rules your application runs. The engine returns answers and a receipt of each execution.
Choose how to start
| You send | What happens | Pricing |
|---|---|---|
| Questions and state | aityx creates or reuses a model, executes it, and returns the definition, answers and receipt. | System Two, including cache hits. |
| A full model and matching JSON state | The engine executes the supplied definition without AI input extraction. | System One; output is free. |
| A prompt or policy document | AI writes a model for you to inspect and test. | System Two. |
| A current definition and a change request | AI revises that definition. You keep both the original and revised files. | System Two. |
| Content and an output contract, without a prompt or files | Validate a definition and inspect the stage report. | Free. |
The endpoint and the pricing category are separate. POST /v1/systemone supports both questions-based
calls and execution of a model you supply. Sending questions does not become a System One-priced call just
because the service reuses a cached model.
Keep the model to run it for less
Once you have reviewed a definition, send the complete model with the facts for each new case. Matching JSON inputs skip the AI reader and run directly on the decision engine. System One input billing includes the model in the request.
Text or JSON that does not establish the required inputs may need the AI reader first. Input extraction uses System Two pricing in addition to the base execution charge. Missing required facts stop execution. See State.
What the cache does
Repeated questions may reuse the same generated definition from a temporary cache. This improves latency without changing the questions-based rate. The cache holds models, not answers, customer inputs or receipts.
The cache is disposable. If it no longer holds a definition, the same questions can generate a new model. For repeatable execution, keep the model you reviewed and submit it directly.
What is retained
The console holds models, inputs and receipts during the current session. Download what you need before refreshing or leaving; upload a model to continue later. The preview has no saved-model library or receipt history. Accounts, access and accounting records are retained separately.
Where AI sits
AI writes and revises models and, when needed, reads facts into typed inputs. The decision engine executes the rules and produces the answer. The receipt records the inputs and rules behind that execution.
Pricing and limits explains the two rates and the shared credit balance.