# Changelog

## 0.0.3
*(2026-10-23)*

#### Improvements
* The list of AI-enabled modules on the Core AI settings screen is shown as styled cards again.

#### Fixed
* An AI request no longer hangs indefinitely when a provider cannot be reached: the connection now gives up after a few seconds even when the request timeout is left disabled, so a blocked or unreachable provider fails quickly instead of stalling the admin action, cron, or page.

## 0.0.2
*(2026-10-06)*

#### Improvements
* Updated the model lists for every provider to current families — Claude (Opus 5.5, Sonnet 5.5), Gemini (3.8 / 3.7 Flash), Grok (4.7 / 4.6 / 4.5) and OpenAI (GPT-6 and GPT-5.6 families); retired OpenAI models (gpt-4, gpt-4-turbo, gpt-3.5-turbo) are dropped from the list and a stored value for one is migrated to a current model.

#### Fixed
* Reasoning / effort-controlled models no longer fail with "temperature is deprecated for this model": the `temperature` parameter is omitted for Claude 4.8+ and OpenAI GPT-5 and later (version-based, so new model generations are covered automatically).
* Claude no longer rejects a request with "each stop sequence must contain non-whitespace": whitespace-only stop sequences (e.g. a bare newline) are dropped before the request is sent.

## 0.0.1
*(2026-10-05)*

#### Features
* Initial release. The AI subsystem extracted from Mirasvit Core into its own package (`Mirasvit\AiCore`): the provider integrations (Claude, Gemini, Grok, OpenAI), the request/response API with image-attachment support, request-header normalization, AI logging, and the admin provider configuration. Modules that use AI now depend on this package instead of carrying the integration in Core.
