Upgrading the AI Packages
Update the base AI module and the providers you use together. Retain your normal database/configuration backup and deployment process, and check custom integrations against the extension contracts.
Settings and Credentials
Existing provider settings retain their hyva_ai/... storage paths. The general execution default is Keep existing feature choices when absent, so an update does not automatically reroute saved requests. Saved provider/model options remain supported. From Hyvä AI 1.3.1, Stores → Configuration → Hyvä AI → Settings and the provider connection test need Hyvä AI Section (Hyva_Ai::config) instead of Content, and existing roles are not migrated.
Choose models in each feature's settings; a provider's connection-test model does not set every feature's model. There is no general requirement to recreate AI settings. Re-enter and save a provider key only when it is missing, invalid or cannot be decrypted on this installation. An encrypted value copied from an installation with a different Magento encryption key is not a usable credential.
For DeepL, save the correct Free/Pro key and run its usage-based connection test. An update cannot repair an unreadable encrypted value automatically.
Retain older model selections only while the provider still accepts them. A saved model remaining visible does not guarantee upstream availability. See Model Guidance.
Verify the Update
- Complete
setup:upgrade, configuration-cache refresh, DI generation and admin static deployment as appropriate for your environment. Restart long-running consumers. - Open Hyvä AI → Settings and confirm only the installed providers appear. Check saved credentials with each available provider test.
- Check Run requests using globally and in each installed feature.
- Test representative requests and review their results.
Direct requests use bounded deadlines and reject incomplete output. If a request exceeds its budget or is truncated, reduce its size or use a feature designed for longer work. The modules do not change your server's timeout configuration.
For Commerce integrations, also follow AI Worker Operations and Feature Configuration.