![[zaphiri-release-cover.png]] # Zaphiri Release 2026.28 > [!quote] Your AI assistant is here - right where your work lives > > This release brings a full AI assistant into Zaphiri: a side panel that helps you work through tasks, find answers in your help docs, and refine your writing - all without leaving the app. ## New features and improvements ### AI chat assistant Zaphiri now includes a built-in AI chat assistant that lives in the right sidebar. Open it from the sidebar rail and it stays with you as you navigate - no separate tab, no context switching. The assistant works across two modes. In **Tasks mode**, you can ask it to complete, move, or create tasks, and those changes reflect immediately on your board. In **Help mode**, it answers questions about Zaphiri using your actual help documentation - so you get accurate, grounded answers rather than generic responses. Conversations are organised into threads, so you can pick up where you left off or start a new one. Responses stream in real time, and the assistant renders task references as clickable chips and structured notes as callout cards. - **Task actions**: Ask the assistant to complete, move, or create tasks and watch them update live - **Multi-turn threads**: Conversations persist so context carries across messages - **Real-time streaming**: Responses appear as they are generated, no waiting for a full reply - **Callout cards**: NOTE, WARNING, and TIP blocks are rendered visually for easy scanning - **Response feedback**: Rate responses with thumbs up or down to help improve quality See [[AI Chat Assistant]] for full details. ### Help chat The Help assistant is a dedicated entry point for how-do-I questions, accessible from the question mark icon in the sidebar. It draws answers from Zaphiri's help documentation using a search-and-retrieval pipeline - so when you ask "how do I set up repeat tasks?" it finds the relevant article and gives you a direct, accurate answer. Help mode is visually distinct from the task chat so it is always clear which context you are working in. - **Docs-grounded answers**: Responses are based on your help documentation, not general knowledge - **Quick help cards**: Common questions surface automatically so you can get answers without typing - **Conversational flow**: Ask follow-up questions and the assistant maintains context See [[Help Chat]] for full details. ### AI writing assistant When you open a task description and highlight text, a set of inline writing actions now appears: improve, proofread, and refine. Apply the suggestion or reject it - the choice stays with you, and the original text is never overwritten until you accept. This runs directly in the task modal, with no need to open a separate panel or send the text anywhere. - **Inline suggestions**: Accept or reject without leaving the task description - **Highlight to trigger**: Select any text to see writing options appear - **Non-destructive editing**: Original text is preserved until you explicitly accept a change See [[AI Writing Assistant]] for full details. ### Bring your own key Prefer to use your own AI provider account? Under [[AI Configuration|Settings - AI Configuration]], you can now add your own Anthropic or OpenAI API key and use it for chat. Supported models include Claude Haiku 4.5, Sonnet 4, and Opus 4 via Anthropic, and GPT-4o Mini, GPT-4o, GPT-4.1, and GPT-4.1 Mini via OpenAI. Keys are stored encrypted and only a masked hint is shown after saving. You can test a key before setting it as default, and remove it at any time. Zaphiri's built-in key remains available as the fallback if no personal key is configured. New accounts include 100 free messages to get started. - **Anthropic and OpenAI support**: Choose the model that fits your workflow - **Encrypted storage**: Keys are stored securely; only a masked hint is displayed - **Provider selection in chat**: Switch between your key and Zaphiri's key directly from the chat panel See [[AI Configuration]] for setup details. ## Bug fixes and refinements - **AI chat bubble display**: Removed the avatar dot and sender label (Claude/Zaphiri) from AI response bubbles; the Intelligence chat panel now uses full-width layout to match Help chat - **BYOK providers load on mount**: The AI provider selector in the chat panel now populates when you open it, without requiring a page reload - **Provider dropdown dismissal**: Fixed an issue where the provider dropdown in the chat panel would not close when clicking outside of it - **Help guardrail accuracy**: Narrowed the phrase list used to detect action requests in Help mode, reducing false positives that previously blocked legitimate how-to questions like "how do I create a task?" --- _Meaningful productivity - built for how you actually work._ Your [[Feedback Process|feedback]] drives our development. If you have suggestions or encounter any issues, please reach out to our team.