* feat(chat): Refactor streaming message handling and add math support
* feat(chat): Basic multimodalities
* fix: language regex and python as default
* fix: refactor llm stream event as one onStreamEvent
- Docs, base translation aand tamagotchi are now in Russian
- Russian language remapping in i18n modules
- Fixed en\settings.yaml
🍰
---------
Co-authored-by: Typed SIGTERM <typed.sigterm@gmail.com>
Co-authored-by: Neko <neko@ayaka.moe>
* feat(unocss): centralize font configuration and fix cyrillic font handling (#389)
This commit refactors the UnoCSS font configuration to establish a single source of truth and resolve issues with Cyrillic character rendering.
- Centralizes all font definitions within the root uno.config.ts.
- Updates the theme.fontFamily to use a correct fallback chain, ensuring "Comfortaa" is used for Cyrillic characters.
- Restricts the "Kiwi Maru" font to Latin and Japanese character subsets to prevent incorrect application.
- Simplifies application-specific uno.config.ts files by removing redundant local font overrides.
* fix(ui): ensure provider models reload reactively (#407)
This commit resolves a persistent reactivity issue where the model list would not update automatically after changing provider credentials. This was a recurring problem reported by users on Discord.
Previously, the model list was only fetched when the consciousness settings page was first mounted. This meant that if a user updated an API key and navigated back, the page would still display the old, stale model list until the application was fully restarted.
The fix introduces two key changes:
1. A `watch` effect has been added to the `providers.ts` store. It now monitors `providerCredentials` and automatically refetches the model list for a specific provider whenever its credentials change.
2. The `consciousness.vue` component now uses the `onActivated` lifecycle hook in addition to `onMounted`. This ensures that the model list is refreshed every time the component becomes active, such as when navigating back to it. A `watch` on the `activeProvider` was also added to handle provider switching within the page.
These changes ensure that the model list is always synchronized with the current provider configuration, providing a much smoother and more intuitive user experience.
Closes#407
---------
Co-authored-by: Neko <neko@ayaka.moe>
This commit introduces a new "OpenAI Compatible" provider, allowing users to connect to any service that implements the OpenAI API specification.
Previously, users were limited to the pre-defined list of providers. With the proliferation of many new AI services and local models (like LM Studio, vLLM, etc.) that expose an OpenAI-compatible endpoint, it has become impractical to add and maintain a separate integration for each one.
This new generic provider solves that problem by offering a single, flexible solution. It clarifies the user's choice by separating the official OpenAI service from other compatible alternatives, improving the overall user experience.
Closes#401, #388
Co-authored-by: Neko <neko@ayaka.moe>
---------
Co-authored-by: Neko <neko@ayaka.moe>
Corrects the `satisfiesVersionBy` logic in the `useVersionedLocalStorage` composable. The previous implementation (`v !== '1.0.2'`) was inverted, causing it to return `false` for the valid version. This triggered the `onVersionMismatch` handler, which reset the data, re-triggered the `watch` effect, and resulted in an infinite recursive update loop, crashing the application.
The condition has been changed to `v === '1.0.2' to correctly validate the version and prevent the loop.
I though this is tauri fault, but kek just silly mistake
Fix#414#375#400
Typed SIGTERM