# AI Features

AI-powered task creation, voice dictation, text enhancement, and an intelligent co-worker built into your workflow.

import { Aside, Tabs, TabItem, Card, CardGrid } from '@astrojs/starlight/components';

Hypertask is built from the ground up as an AI-native project management platform. AI isn't bolted on as an afterthought — it's woven into every part of the workflow, from creating tasks to communicating with your team.

<CardGrid>
  <Card title="AI Task Writer" icon="pencil">
    Describe what you need. Get detailed, context-aware tasks.
  </Card>
  <Card title="Voice Dictation" icon="seti:audio">
    Speak your thoughts. Hypertask transcribes and structures them.
  </Card>
  <Card title="HyperAI Co-worker" icon="comment">
    Chat with an AI that knows your project — and can take action on it.
  </Card>
  <Card title="Text Enhancement" icon="document">
    Improve grammar and readability of any text with one click.
  </Card>
</CardGrid>

## AI Task Writer

Instead of manually writing detailed task descriptions, tell the AI Task Writer what you need in plain language. It generates fully structured tasks with context-aware details.

**How it works:**

1. Open the AI Task Writer from any board.
2. Describe what you need — a sentence or two is enough.
3. The AI uses your **project context** (existing tasks, descriptions, labels) and any **custom instructions** you've configured to generate a detailed task.
4. Toggle **web search** on if the AI should research external information (documentation, APIs, best practices) while writing the task.
5. Review, edit if needed, and add the task to your board.

The Task Writer maintains a **history of responses**, so you can revisit previous suggestions and reuse them.

<Aside type="tip">
The more context your project has — existing tasks, custom instructions, attached documents — the better the AI Task Writer performs. It learns from your project's patterns.
</Aside>

## Voice Dictation

Capture ideas hands-free with OpenAI Whisper-powered voice dictation. Use it anywhere you can type — task titles, descriptions, comments.

| Feature | Details |
|---------|---------|
| **Keyboard shortcut** | `Shift + Cmd/Ctrl + D` |
| **Continuous mode** | Keep talking without pressing buttons repeatedly |
| **Echo cancellation** | Works in noisy environments and during calls |
| **Noise suppression** | Filters out background noise for cleaner transcription |
| **Multilingual** | Supports all languages that Whisper recognizes |

Speak naturally. The transcription handles punctuation, formatting, and language detection automatically.

## Text Enhancement

Select any text in a task description or comment and click **Improve Readability**. The AI refines grammar, clarity, and style while preserving your meaning.

- Works on any text field in the app
- Preserves technical terms and project-specific language
- **Undo support** — revert to the original text if you prefer it

This is intentionally lightweight. It's not a rewrite — it's a polish.

## HyperAI Co-worker

HyperAI is an AI chat sidebar built into the Hypertask app. It's context-aware: it knows your project, your tasks, and your team structure.

### What makes it useful

- **Project context** — HyperAI can see your boards, tasks, and project configuration. Ask it questions like "What are the highest priority tasks this week?" or "Summarize what happened on the API project yesterday." It also understands archived tasks, so you can query completed or archived work.
- **Take action** — HyperAI can create tasks, assign team members, set priorities, add labels, and manage your board directly from the chat sidebar. Just tell it what you need — "Create a high-priority bug for the login issue and assign it to Sarah" — and it handles the rest.
- **Consistent formatting** — Both AI Chat and HyperAI always return responses in HTML, so output renders correctly in the app regardless of the model or prompt used.
- **@mentions** — reference specific tasks, people, or projects in your messages to give the AI precise context.
- **Ask AI from comments** — see a comment or description you want to dig into? Click the "Ask AI" button on any comment or task description to send it directly into an AI chat, where you can ask follow-up questions or get analysis.
- **Chat sessions with history** — conversations are saved as named sessions. The AI automatically gives each chat a descriptive name, and you can browse your full chat history to pick up where you left off.
- **Available on mobile** — AI Chat is now accessible full-screen on mobile. Tap the chat icon on any mobile view to get the same context-aware assistant you have on desktop.
- **Fast, responsive streaming** — AI chat uses optimized streaming for low-latency responses, with a smooth conversation flow that keeps up with your thinking.

### Attachments in AI Chat

You can attach files directly to messages in AI Chat, giving the AI concrete reference material to work with alongside your prompt.

**Supported content:**
- Images (screenshots, mockups, diagrams)
- Documents (PDFs, text files, spreadsheets, and other common formats)

**How to attach:**
1. Open AI Chat.
2. Click the attachment icon (📎) in the message composer, or drag and drop a file into the chat input.
3. Select your file. It appears as a preview alongside your message.
4. Type your prompt and send — the AI receives both the file and your text.

<Aside type="tip">
Attachments are great for asking the AI to analyse a screenshot of a bug, review a spec document, or extract action items from a PDF — all without leaving your project context.
</Aside>

### Keyboard shortcuts

| Shortcut | Action |
|----------|--------|
| `Cmd/Ctrl + Shift + ?` | Minimize / restore the chat sidebar |
| `Cmd/Ctrl + Shift + O` | Start a new chat |

### What AI Chat can and cannot see

AI Chat's access to your workspace is intentionally kept in line with what the CLI and MCP tools can read and write. This means AI Chat shares the same scope and limitations as those interfaces.

| Workspace data | AI Chat | CLI / MCP |
|---|---|---|
| Tasks (title, description, status, priority, size) | ✓ | ✓ |
| Assignees & followers | ✓ | ✓ |
| Comments | ✓ | ✓ |
| Labels | ✓ | ✓ |
| Sections / board columns | ✓ | ✓ |
| Subtask hierarchy | ✓ | ✓ |
| Due dates | ✓ | ✓ |
| Archived / completed tasks | ✓ | ✓ |
| Saved views | ✗ | ✗ |
| Tags | ✗ | ✗ |

<Aside type="note">
**Saved views and tags are not currently accessible to AI Chat, the CLI, or MCP tools.** All three interfaces share the same access boundary — if you need to work with views or tags, use the Hypertask web UI directly. This parity is intentional: AI Chat is powered by the same underlying tools as the CLI and MCP, so their capabilities stay in sync.
</Aside>

## Auto-Summary

When a task accumulates a long comment thread, Hypertask generates an **automatic summary** so you can get up to speed without reading every message.

Auto-summaries highlight:

- Key decisions made in the thread
- Current status and blockers
- Action items that emerged from the discussion

<Aside>
Auto-summaries update as new comments are added, so they always reflect the latest state of the conversation.
</Aside>

## Custom AI Instructions

Tailor how AI features behave on a per-project basis.

### Project-level configuration

- **Custom instructions** — write free-text instructions that guide AI behavior. For example: "Always include acceptance criteria in generated tasks" or "Use British English."
- **Attached documents** — upload files that give the AI domain-specific knowledge (RAG). Product specs, API docs, style guides — anything that helps the AI understand your project's context.
- **Model selection** — choose which AI model powers your project's AI features.
- **Visibility** — set instructions as public (visible to all members) or private (visible only to you).

### Supported models

<Tabs>
  <TabItem label="OpenAI">
    - **GPT-5.2** — most capable OpenAI model
    - **GPT-5 mini** — faster and more cost-effective
  </TabItem>
  <TabItem label="Anthropic">
    - **Claude Opus 4.6** — strongest reasoning and analysis
    - **Claude Sonnet 4.6** — balanced performance and speed *(default)*
    - **Claude Haiku 4.5** — fastest, ideal for high-volume tasks
  </TabItem>
  <TabItem label="Google">
    - **Gemini 3 Pro** — most capable Google model
    - **Gemini 3 Flash** — optimized for speed
  </TabItem>
</Tabs>

<Aside type="tip">
Different models have different strengths. Use a more capable model (Opus, GPT-5.2, Gemini 3 Pro) for complex task generation and a faster model (Haiku, GPT-5 mini, Gemini 3 Flash) for quick text improvements and summaries.
</Aside>
