AI Features

AI is woven through every part of Hypertask — create tasks, filter your board, smart-label workloads, and manage daily operations without leaving the workflow you already use.

1. AI-assisted task creation

   • Click AI in the top-right corner of any board, open the Ask Hypertask AI panel, and describe a task in plain English.
   • AI parses the description and creates a task pre-assigned to you with a title, description, and relevant tags (task types, priorities, area).
   • Review the generated task before saving; edit fields or refine the prompt and generate again.

2. Prompt-based filtering

   • In the Board Filter, type natural-language constraints like:
     • show me top-priority bugs
     • tasks assigned to Valentin and due this week
   • Use combinations such as high priority and not blocked, and press Enter to apply.

3. Smart labels

   • Open a label (or the Manage Tags screen) and create a new label marked as a smart label.
   • In the "Smart label — AI keeps this label on matching tasks" field, define which tasks should carry it, for example:
     • backend issue, involves DB, high impact
   • Save; AI immediately classifies the board up to 500 existing tasks and adds the label to matches. Smart labels show a sparkle icon in label lists.
   • New tasks are classified when created; edited tasks are re-classified when title or description change; updating the prompt re-runs classification for the entire board.

4. AI usage without a signed-in user

   • AI usage can now be recorded and stored as usage tracks even when calls don't originate from a signed-in user.
   • This allows external agents or systems invoking the AI APIs to function without requiring active authentication, while still producing usage records available to admins.

5. Batch-powered background workloads

   • Background smart labels and other AI classification runs that involve many tasks are routed to Anthropic's Batch API at half price, allowing cost-efficient processing at scale.
   • Matching tasks receive their labels automatically; activity is logged in the task activity feed.
   • AI writes a compact representation of your content and sends it to the embed service, which can reject very long or empty text or refuse very long or empty text. On these failures, the upsert silently rotates the text and upserts without final vectors instead of failing, so comments and pages remain searchable with the relevant tags even when vectors aren't generated.
   • You don't see extra UI for this—it happens invisibly in the background with consistent results.

6. AI Chat suggestions for tickets

   • Open AI Chat in a ticket panel to get context-aware suggestions tailored to that ticket's specific information.
   • Previously, suggestions could be generic or fail entirely when the server returned "No output generated"; Haiku 5.5 now returns proper suggestions that match the ticket context.

7. Faster agent task lists

   • AI agents are now much quicker to load their task lists, especially after server warm-up.
   • Previously slow database reads that fetched task lists have been optimized: the slowest 5% of calls improved from ~129 ms down to ~40 ms. The improvement consistently benefits agents on both warm and cold servers, making AI-driven workflows feel far more responsive.

8. AI Chat histories

   • Open AI Chat from the right rail or press Ctrl/Cmd + I to see a list of conversations.
   • In the full-screen chat view (desktop), the history list pages as the window resizes, and you scroll to load more older chats. Pick or filter from that list to open a chat.
   • On mobile, the list is limited; tap an entry to open its history.
   • The docked chat always shows your most recent chats inline; use the full-screen view for history navigation.

9. Comment summaries preserve open questions

   • When AI summarizes your comments, it now accurately reflects whether questions have been answered or remain unresolved.
   • After this fix (HTPR-7046), comment summaries display Open: ... for unanswered questions instead of inventing an answer, and retain "still unresolved" language when appropriate.

10. Bulk archiving avoids server errors

    • Agents that attempt to archive many tasks see a clear refusal message instead of a server crash.
    • An earlier incident had agents trying to archive 31 Done tickets per day, triggering a 500 error. The fix now returns a direct not allowed response, keeping workflows smooth and errors visible for observers.

Where AI appears

Area	What AI does
Task creation	Parses your description into a task with title, description, priority, and tags.
Board filtering	Understands natural-language constraints and builds dynamic filters.
Smart Labels	Keeps labels on tasks that match your AI prompt; changes to tasks or the prompt trigger reclassification.
AI Chat	Global assistant for questions, summarization, and quick actions, with full-screen history paging.
Ticket AI suggestions	Provides context-specific suggestions in ticket panels that mirror the ticket's subject and related data.
Bulk AI workloads	Heavy classification runs use the Batch API to handle large boards efficiently; background jobs go through Anthropic's half-price Batch endpoint. Usage is recorded as usage tracks regardless of whether a signed-in user is present.
Agent task lists	Optimized reads for quick agent workloads; significant speedup for the slowest dataset slices.
Comment summaries	Summarize comments and preserve whether questions remain open for human follow-up.

Task creation with AI writer

1. Open the New Task box — Use Ctrl+J on desktop or Cmd+J to open the New Task creator from boards, the inbox, or any view. On the Tasks page, click New Task in the toolbar.

2. Describe your task — Enter a brief description of the work you need done. You can type directly or let AI expand your ideas.

3. Let AI write the task — Reviews your inputs and generates a complete, structured task with a title, description, priority, and relevant tags. AI respects your board's templates and style when available.

4. Review and refine — Edit any field in the generated task. You can press Ctrl+K (desktop) or Cmd+K (Mac) in the description area to refine the AI's work, regenerate, or add more detail.

5. Save the task — When you're satisfied, click Create (or Enter) to save the task and add it to the board. The task appears immediately in the appropriate column or section.

AI writer failure reasons

Fix When the AI writer fails to generate a task, the Ctrl+J New Task box and the AI Task Writer now show you the real reason for the failure instead of hiding it. Your note is still saved as a task, so you don't lose your work:

• Error messages displayed — Common errors like quota limits, missing context, or prompt format issues show explainable messages.
• Task created anyway — Your original description or note is preserved as a task, so you can retry AI later or edit it manually.
• Applies everywhere — This fix covers task creation from boards, the inbox, and when editing task descriptions.

If AI generation is not available (for example, your team has not configured an AI key), the text you entered appears without modification, keeping your workflow uninterrupted.

AI Task Writer

The AI Task Writer is a prompt-based task creation flow that generates complete, shippable tickets from paste inputs.

How it works today

The Task Writer writes tasks in your team's style. It uses board-aware prompts, pulls task templates from boards, and uses search to find related tickets instead of inventing them.

Using the Task Writer

• From a board: Ctrl+J / Cmd+J and select Create with AI — Task Writer.
• From the command palette: Ctrl+K — search for "Task Write" and select it.
• From a task: Click the AI icon in the task header and select Task Writer.

After the Task Writer generates a draft, you can refine it just like any other AI-generated comment or description.

When AI is not available

Behavior change AI features now show clear guidance when a team does not have an AI key configured. Before this fix (HTPR-7149), teams without an AI key would see server errors when trying to use AI in agents or the CLI. Now, users receive a concise message directing them to add an AI key through the Appearance & AI Models settings page:

"AI features require an AI key. Configure your AI key in Appearance & AI Models."

This applies to AI-powered task creation, filters, smart labels, and agent workflows that invoke AI.

Settings

• Choose your preferred AI model in the Appearance & AI Models settings page. Different models may produce different responses.
• Adjust leniency of AI-assisted task creation in the same settings area if you prefer more precise or broader suggestions.
• Smart label accuracy can be tuned by sharpening or relaxing your prompt; see the "How it stays current" details in the Smart Labels reference.

Note

Smart labels and prompts are a best-effort AI signal. They do not substitute for human review. Ensure your prompt manages edge cases or exceptions directly, for example: "NOT: schema changes on branches not yet merged."

Tip

Use smart labels together with filters and saved views to create reusable patterns for recurring workloads (e.g., on-call issues, quick fillers, enterprise integrations).