AI assistant

Personal AI Access and shared Providers & Credentials offer Use the provider’s fallback model, off by default and hidden on older servers. The switch explains the recommendation source, possible cost/latency/capability changes, and administrator notification. The switch appears only for OpenAI OAuth, Gemini, xAI API, and Grok OAuth, whose providers supply dynamic defaults or rolling aliases. It is absent for OpenAI API, Anthropic, and local endpoints. No replacement model is entered or pinned. The shared preference also applies to a remediation-model override. Chat renders the selected model, replacement attempt, and reason in a separate notice with an AI Access link; this notice is never sent back as assistant text. Administrator fallback issues offer a provider-settings link, and remediation run history records replacements.

• Multi-provider chat with incremental SSE streaming on native and web, visible tool activity, and a poster carousel for results. Every user can bring a personal Anthropic, OpenAI, Gemini, or xAI Grok API key, or link a subscription account with a browser device code: OpenAI (OAuth) via ChatGPT or xAI Grok (OAuth) via SuperGrok / X Premium+. Admins can configure the same choices as an included server profile and grant it per user. Personal overrides fail closed instead of silently spending shared quota.
• Server-side tools: the assistant searches (movies, TV, books, and music), checks availability, and requests on your behalf; book and album results ride the same carousel and open the matching detail page with their catalog identity and selected instance. Admins can manage libraries and downloads, troubleshoot issues, and update supported connected-service settings conversationally.
• Configuration receipts: explicit admin requests can update supported connected-app settings in one turn, without copying a confirmation command back into chat. Supported profile and custom-format writes return a trusted review receipt; quality-profile update receipts also lead to a one-time guarded restore when the live state still matches. Assistant prose never creates controls.
• Persistent session: the focused /assistant workspace keeps one conversation alive across navigation (30-minute idle expiry).
• Capability-aware introduction: the welcome and suggested prompts use the server’s current per-user capability summary, refreshed when opening the assistant or starting a new chat. Books and music require an accessible Chaptarr or Lidarr instance; requests and administrative actions also require their enabled tools. Older servers or unreadable capability data receive a neutral greeting without capability claims.

View the maintained source for this page.