Canvas
The canvas is the visual surface your agent builds and updates. It lives in the right panel of your agent project and displays whatever your agent finds useful — KPI metrics, data tables, charts, status indicators, forms, and more.
The canvas is primarily a reporting surface — the agent writes to it on each heartbeat and on chat, and you read it. It can also include interactive controls (buttons, inputs, forms) when the agent decides they're useful, and those controls can either run locally in the canvas or send a request back to the agent.
What the canvas can show
The canvas is built from components. The AI can combine them freely to create any layout.
Data components
Metric (KPI card)
Displays a single number with an optional label and trend indicator. Used for at-a-glance measurements.
MRR Open PRs API Uptime
$48,200 12 99.97%
▲ 8% vs last month +3 this week ▼ 0.02%
Chart
Visualizes data over time or by category. Supports line, bar, horizontal bar, area, pie, donut, and progress charts.
Table
A grid of rows and columns for structured data — open tickets, recent PRs, invoices, repository activity.
DataList
A vertical list of items with labels and values. Good for key-value pairs and settings summaries.
Display components
Card — A container with a title and optional border. Used to group related information together.
Text — Headings, body text, captions, code snippets, muted labels.
Badge — A small label with color variants. Used for status indicators (e.g., Active, Failed, Pending).
Alert — A highlighted callout for warnings, errors, or important notices.
Progress — A horizontal progress bar, useful for completion percentages.
Separator — A dividing line between sections.
Icon — One of 40+ icons (arrows, checks, alerts, domain icons like dollar, calendar, globe, etc.)
Layout components
Row / Column — Arrange components horizontally or vertically.
Grid — Multi-column grid layout.
Tabs / TabPanel — Tabbed sections for organizing multiple views.
Accordion / AccordionItem — Collapsible sections.
ScrollArea — A scrollable container for long content.
Interactive components
Button — A clickable button. Buttons can update canvas state locally (e.g., toggle a section, filter a table) or send a request back to the agent (e.g., "Refresh data", "Mark resolved") — the agent decides which is right for the situation.
Input / Textarea — Text entry fields for simple forms.
Select — A dropdown of options to choose from.
Checkbox / Switch — Boolean toggles.
Interactive controls in the canvas are real React components — they can manage their own local state for things like filters, accordions, or unsaved form input. When an action genuinely belongs to the agent (calling a tool, mutating workspace state, sending a message), the button sends a request back to the agent and the canvas updates from its response.
How the canvas is updated
The agent writes to the canvas by calling canvas tools during a chat session or heartbeat run. This happens automatically — you don't need to trigger it manually.
On heartbeat: The agent runs its checklist, fetches fresh data, and updates the canvas with new values.
On chat: When you ask for a dashboard or describe a layout change, the agent builds or updates it immediately.
Via a Button: Clicking a canvas button sends a request to the agent, which takes action and updates the canvas in response.
Prompting for a canvas
You don't need to know the component names to describe what you want. Describe it the way you'd describe a report or dashboard to a colleague.
Examples:
"Build a dashboard with our key metrics at the top — MRR, churn rate, and active subscriptions — and a table of recent invoices below."
"I want three tabs: Overview, GitHub Activity, and Support. Overview should show API status and uptime. GitHub should show open PRs in a table. Support should show ticket volume as a bar chart."
"Add a section showing CI status for each of our repos. Use green badges for passing, red for failing."
"Create a status card at the top showing whether the service is healthy or degraded, with a timestamp of the last check."
Tips for better canvas prompts
Name your sections. "A section called Incidents, a section called System Health" gives the AI clear structure to work with.
Describe the data, not just the look. "A table of open GitHub PRs with columns for repo, title, author, and age" is better than "a table of PRs."
Include trend information if you want it. "Show MRR with a week-over-week percentage change" explicitly asks for trend arrows — they're not added by default.
Reference your connected tools. "Using our Stripe data" or "using the GitHub repos we configured" tells the agent where to pull the data from.
The canvas updates automatically
Once configured, the canvas refreshes on every heartbeat tick without you doing anything. Each tick:
- Agent runs the heartbeat checklist
- Fetches fresh data from connected tools (GitHub, Stripe, etc.)
- Updates the canvas components with the latest values
- Sends an alert to your channel if anything needs attention
You can also ask for a manual refresh at any time:
"Refresh the dashboard with the latest data."
Canvas examples by template
| Template | Canvas includes |
|---|---|
| Research Assistant | Key takeaways card, article table with source links, topic breakdown |
| GitHub Ops | Open PR queue table, CI status badges per repo, issues table |
| Support Desk | Ticket volume chart, SLA status, priority breakdown, open tickets table |
| Revenue Tracker | MRR metric with trend, failed payments table, revenue chart (30 days) |
| Incident Commander | Service health status grid, recent incidents table, uptime metrics |
| Project Board | Sprint progress, velocity chart, tasks table with assignees |
Related
- Heartbeat — the canvas is updated on each heartbeat tick
- Chat with AI — how to describe canvas changes through chat
- Templates — each template includes a pre-built canvas dashboard