Decision AI

Decision AI connects Jev through the System One API to the Decision Model component. Configure the connection once, then select it in nodes across your streaming and request-response scenarios.

Decision Model evaluates event content using choices, ordered scores and yes/no probabilities. Use its typed results in filters, switches and Decision Tables. For chat or text generation, use the separate AI Models integrations.

Configure the integration

1. Open Integrations in the Admin panel and add Decision AI.
2. Give the integration a name, for example jev.
3. Set Base URL. The default is https://api.typesafe.ai/v1; you can change it for your endpoint. Enter the base URL, without the /systemone suffix.
4. Enter your Jev API key. The field is optional for endpoints that do not require authentication; when present, the key is sent as a Bearer token.
5. Click Check connection to retrieve available models, then select the models to expose in scenarios and save the integration.
6. Add a Decision Model node and select the integration and one of its saved models. Run a test with your context and questions to verify inference.

Setting	Meaning
Base URL	Used for model discovery at GET <base URL>/models and inference at POST <base URL>/systemone
API key	Optional Bearer credential for the configured endpoint
Models	Saved model IDs offered in the node's Model selector

Check connection lists models; it does not run inference or prove that a particular question set will succeed. After changing the URL, check the connection again and confirm the selected models. Scenario validation uses saved configuration and does not contact the provider.

A model alias may move to a newer version. Prefer a versioned model when available and inspect #assessment.model in results. The returned name may itself be an alias or request echo, so it is not proof of an immutable model version.

The API key field masks its value on screen. The current Configurator API returns configured keys to callers with access; masking is not a separate secret store.

Supported service and limits

This integration currently supports Jev through System One. Clef through Cloudflare Workers AI and decision models served by Ollama are not supported by this integration yet.

Advanced integration settings control request timeout, request size, concurrency, queue capacity and question/criteria limits. They apply to every node using that integration; concurrency is scoped to each invoker rather than shared across the deployment. See Decision Model for runtime defaults, result modes and error handling.

Next steps

Follow the SIM replacement reply example to configure questions manually or with Nu Assistant. Keep business actions in downstream rules: the model provides an assessment, while the Decision Table defines the policy.