The AI dashboard is currently in beta and may be unavailable or have limited functionality in certain regions.

Tutorial

Dashboard Editor Deep Dive

A section-by-section walkthrough of the dashboard edit page: planning, AI spec edits, tasks, buckets, datasources, draft thread, and assistants.

1

Map the editor sections

The editor is a draft workspace. Changes remain in draft until you publish. The key areas are: Planning prompt, AI edit this spec, Tasks, Buckets, Datasources, Draft Thread, and Assistants.

Dashboard editor map with labeled sections
Treat this page as one cohesive workflow: plan, edit, validate, and publish.
2

Planning prompt vs AI edit this spec

Planning prompt is intent-first and broad. Use it to describe desired outcomes like "add SLA risk indicators and escalation trends". AI edit this spec is precision-first and applies focused structure changes to the JSON spec.

In short: planning prompt answers what to build; AI edit this spec answers what to change.

Comparison between planning prompt and AI edit this spec
Start broad with planning, then use AI spec edits for exact refinements.
3

How Tasks work

Tasks are the runtime units in your dashboard definition. A task has a title, kind, status, and description. The Up and Downcontrols reorder tasks in the draft.

Use tasks for chart widgets, pipeline steps, integration actions, and trigger-linked work. If a task row is blank or unclear, fix it before publish.

Task list showing reorder controls and task fields
Tasks define what executes and what users see in the resulting dashboard flow.
4

How Buckets work and where they come from

Buckets are grouping/layout containers. They can be generated by planning, created manually, or inserted via AI spec edits. Buckets organize related outputs and influence display modes such as cards, tables, charts, or text sections.

Bucket configuration with display modes
Buckets provide structure so complex dashboards remain readable.
5

What Datasources are

Datasources tell the dashboard where information comes from. Common kinds include integrations, project documents, API-connected sources, and crawl URLs.

Use the name-based selector first. It maps readable names to the stored reference ID. Manual reference input is still available for advanced or edge-case values.

Datasource rows with kind selector and named reference dropdown
Prefer selecting by name over raw hashes whenever possible.
6

What Draft Thread is for

Draft Thread is draft-scoped collaboration context. Keep iterative requests there while you tune the draft. It is ideal for "next change" instructions and rationale notes.

Use it when you want continuity in refinements without losing context between edits.

Draft thread conversation panel in dashboard editor
Thread messages stay tied to this draft while you iterate.
7

How Assistants work and how to add each type

Open Add assistant, choose a type, then configure name, prompt, allowed tools, and access mode. Assistant type determines behavior:

  • Chat: answers dashboard questions in real time.
  • Scheduled task: runs on cron to refresh/process data.
  • Trigger: reacts to dashboard events.
  • Notifier: sends Slack/email alerts when conditions match.
  • Scraper: fetches and extracts web content on demand.
  • Webhook listener: receives inbound HTTP POST payloads.
Add assistant dialog showing assistant type cards
Pick the assistant type based on trigger mode: user chat, time schedule, event, or inbound webhook.
Tip:A reliable workflow is: planning prompt for broad shape, AI spec edits for precise adjustments, manual review of tasks/buckets/datasources, then publish with a clear change summary.