Dashboard

The Dashboard tab is a Grafana-style space where you build grids of chart widgets. Each widget is rendered from the output of its own hidden Heym workflow, so any data you can produce in a workflow — database queries, API calls, RAG lookups, LLM output — can become a chart. You can keep several dashboards and share each one with people or teams.

Dashboards

Everyone starts with one dashboard, created the first time they open the tab. Use the dashboard selector at the top of the tab to switch between dashboards, and + next to it to create a new one. The selector lists your own dashboards first, then the ones shared with you, labelled with the name of the person who shared them.

The tab reopens the dashboard you used last, and the address bar carries the open dashboard (?tab=dashboard&dashboard=<id>), so a link opens that exact dashboard for anyone who has access to it.

The settings button (owners only) renames the dashboard, sets which records it accepts as a detail page, manages sharing, and deletes it. Deleting a dashboard removes its widgets and their workflows. You always keep at least one dashboard of your own, so the last one cannot be deleted.

Widgets

A widget is a single chart on the grid. Supported chart types:

• Bar (vertical or horizontal)
• Line
• Area (filled trend, supports multiple series)
• Pie
• Table (scrollable; click a column heading to sort ascending, and click it again for descending. One column can render as status chips, one column can hold link buttons that open a URL in a new tab, and rows can open a detail page)
• Numeric (a single KPI value with an optional unit)
• Gauge (a single value against a min–max range, e.g. a percentage)
• Scatter (X/Y points for correlation plots)
• Proportion (one bar split into shares with a percentage legend, e.g. a language breakdown)
• Gantt (task dates, completion and dependencies)
• Entity relationships (DB) (tables, fields, PK/FK keys and connections)
• Bar gauge (one horizontal gauge per row with a red→green gradient and a value, e.g. free disk space)
• Text (a markdown message, e.g. a status note like "Last execution at 19:47"; supports interactive checkboxes when the markdown is static, and explicit numbered lists for custom or descending numbering)
• HITL (your pending human reviews. The header shows 1/n pending. The workflow name opens that run on the canvas. Approve, Request changes, and Reject resolve it. The history icon beside the widget title opens that run in the history dialog on this page. The widget does not use upstream rows, and each person sees their own queue. It can show one workflow's reviews instead, for everyone who can open that workflow; see HITL widgets)

Besides charts, a Run a workflow widget runs one of your workflows from the dashboard: on a dropped file, with its input fields, or with a Run button; see Run widgets.

Each widget loads its data asynchronously when you open the tab, so the page stays responsive while charts populate.

Adding a widget

1. Click Add widget, give it a title, and pick a chart type. Below the picker, the dialog draws an example of that chart type with sample data and names the rows it expects, so you can compare types before you build anything.
2. The widget opens in the workflow editor with a starter graph: a Set node that produces the rows, connected to a Chart Output node. Replace the Set node with any data source. A HITL widget is only the Chart Output node; it reads your pending reviews and does not need a data source.
3. Build the workflow so the node feeding Chart Output produces an array of rows, then configure the Chart Output node's field mapping (label field, value field, etc.).
4. Save, return to the Dashboard tab, and the widget renders.

For a Run a workflow widget, pick the workflow it runs instead; the picker says how the widget will run each one. The widget is added to the grid directly; there is no new workflow to build.

Double-click a widget (or use its edit button) to reopen its workflow at any time. Use Clone widget in the widget header to duplicate both the widget and its complete workflow. The copy is placed below the original and labeled with (Copy) so you can edit it independently.

HITL widgets

A new HITL widget shows every review waiting in your own queue. To show only one workflow's, open the widget's Settings:

1. Under Workflow, pick the workflow. The list holds the workflows you can open, yours or shared with you or your team, that have a step waiting for a person: an Agent with human review, or a Slack approval, buttons or form step. The icon beside it opens that workflow in a new tab.
2. If the workflow has one such step, it is picked with it. If it has several, Select agent lists them by node name, with All agents first; Slack steps are marked (Slack).
3. Save. The header's 1/n pending counts only those reviews. Pick All workflows to show your own queue again.

The same choice is on the canvas, in the Chart Output node's properties. Fine-tune with AI can make it ("show only the invoice approvals"), and so can describing a new widget that names the workflow. Every change is a workflow edit, so it appears in Edit History.

• Who sees the reviews. A widget filtered on a workflow shows that workflow's reviews to everyone who can open the workflow, and they can approve, request changes or reject from it, as they could run the workflow. An unfiltered widget shows each person their own queue.
• Who sees the widget. Someone who cannot open the chosen workflow does not see the widget at all, even on a dashboard shared with them. When the dashboard's owner loses access to it, or it is deleted, the owner sees "You no longer have access to the workflow this widget shows" and can pick another one in Settings.
• Only reachable choices. A workflow you cannot open is refused when the widget is saved, whether from Settings, the canvas or the AI.

Detail pages

A table widget can link its rows to another dashboard, which then works as a detail page: one dashboard that serves every customer, invoice, or ticket.

1. Open the table widget's Settings and pick the dashboard under Row link.
2. Enter the Record column, the column whose value identifies the row (for example id), and optionally a Label column that names it (for example name).
3. Save. Clicking a row (or focusing it and pressing Enter) opens the linked dashboard with ?record=<value>&label=<name>.

On the detail dashboard, every widget workflow reads the value as $page.record (see Chart Output), and the page shows Dashboard / <label> above the grid with a button to clear the record. Back returns to the table you came from.

• Who can open it. Rows are clickable only for viewers who can open the linked dashboard. Others see the same table without links.
• Record format. The record comes from the URL, so the detail dashboard decides what it accepts. In its settings, Detail page records is one of: IDs (letters, digits, -, _ and ., up to 128 characters; the default), numbers, UUIDs, or email addresses. Any other value is refused before a widget runs, and the widgets show "This page does not accept that record".
• Still the owner's run. Widgets run as the dashboard owner on a detail page too. Builders should pass $page.record only where a node escapes it, never into query text.
• Caching. Each widget caches its chart per record, for the same time-to-live as the widget, and keeps the 50 most recently used records.

Run widgets

A Run a workflow widget runs one of your workflows and shows the result: status, run time, and outputs. What it shows depends on what the workflow takes:

• A file. For a workflow that starts with a File Upload Trigger, drop a file on it (or click to pick one) and the widget uploads it through the trigger's single-use upload link. The widget shows the file types and size the trigger accepts and refuses others before uploading. When the workflow also starts with a Text Input node, the widget shows its fields next to the drop zone: choose a file, fill the fields, and Run sends both. The values travel with the upload link the widget mints, so a shared upload link cannot set them.
• Input fields. For a workflow that starts with a Text Input node, the widget shows its fields with their defaults; Run runs the workflow with your values. Values for fields the workflow does not have are dropped.
• Nothing. A workflow without inputs gets a Run button.

The result's Wrap button wraps long output lines inside the box; turn it off to read the output's JSON layout.

• Anyone who can open the dashboard can run the widget, as anyone can refresh a chart. Access to the workflow is the dashboard owner's: adding the widget needs a workflow that both you and the owner can run, and the run itself uses the workflow owner's credentials, like any upload link.
• The result stays on your screen until you leave the page; other viewers do not see your files, values or results. The run appears in execution history with the trigger dashboard.
• A dashboard share never grants access to the workflow itself. Edit workflow opens it for people who already have access, and Fine-tune with AI is not offered.
• File runs execute on the main instance, where uploaded files are stored, as every File Upload Trigger run does.

Generating a widget with AI

Click AI, describe the metric you want (for example, "workflow success rate over the last 30 days as a bar chart"), and pick an LLM credential and model. Heym generates a complete widget workflow ending in a Chart Output node and adds it to the grid. See Chart Output for example prompts per chart type.

The widget gets a short title of one or two words in the language of your request, such as "Open cases". A title never carries a data table's id; when the model names the widget with a sentence, Heym keeps its first few words.

The widget API can also make a count card without AI: a text widget on a data table (data_table_id) with count_column and count_value shows how many rows have that value; the widget's title names the value. Heym Work leads pages made from a data table with one count card per status.

Use a widget's Fine-tune with AI button to revise an existing widget with a new instruction. Each AI fine-tune snapshots the previous workflow into the widget's Edit History, so you can review or roll back changes from the workflow editor.

From Chat and MCP

Chat, and MCP clients through the heym_chat tool, can read your dashboards, create new ones, build and change several widgets at once, add widgets one at a time, rename or delete widgets, and tidy a dashboard's layout. Heym Work shows each dashboard as a page, so the same answers cover its pages.

• Which dashboards, and what is on them. Ask "which dashboards do I have?", "what is on the Sales dashboard?" or "which dashboard is the revenue workflow on?". Chat lists your own dashboards and the ones shared with you, with your access and a link to each, and every widget on them: its title, chart type, and the workflow behind it. A HITL widget filtered on a workflow you cannot open stays hidden, as it does on the dashboard.
• A new dashboard. Ask "create a Support dashboard with open tickets by priority, weekly ticket volume and the backlog". Chat creates the dashboard, then builds its widgets in the same answer. When you did not say what it should show and no data table fits, it asks first.
• Several widgets at once. For a new dashboard, or when you ask for more than one widget, Chat builds them in parallel with the same AI widget generator as AI generate and Heym Work's page wizard, reading the data tables they use. Each widget appears as soon as it is ready, in the order Chat planned. Its first data is checked, and a widget that shows nothing (no rows, or every value 0) goes back to the generator once with what it returned. Chat then lists what it built and anything that failed, with the reason. Building widgets this way needs a chat model on an OpenAI, Google or Custom credential; with another model, Chat builds each widget as described in the next item.
• Change several widgets. Ask "make the volume chart a line chart and show only the last 30 days in the backlog". Chat sends every change in one go, the same way a widget's Fine-tune with AI does, so each change is saved to the widget's Edit History and can be reverted. Heym Work's page view sends the notes you leave on a page's widgets this way.
• A new widget. Ask "add a chart of open cases by status to the Sales dashboard". Chat first puts an empty widget on the dashboard, then writes the widget's own workflow, test-runs it and fixes it until a run passes, the same way it builds workflows. No separate workflow is left behind, and each save appears in the widget's Edit History. The widget's title and chart type follow what Chat saves.
• An existing workflow. Ask "put the revenue workflow on the Sales dashboard". A workflow that ends in a Chart Output node becomes a chart widget with its own copy of the workflow, so later changes to the widget never touch the original. A workflow with a trigger or input node cannot become a chart widget. Any other workflow becomes a run widget, which also needs a workflow the dashboard owner can run.
• Rename or delete a widget. Ask "rename the Revenue widget to Sales" or "delete the Open cases widget". A chart widget's workflow takes the new name too; a run widget's workflow keeps its own. Deleting a chart widget deletes its workflow with it, as the widget's delete button does; a run widget's workflow stays. Chat deletes only a widget you asked it to, and asks which one first when your words match several.
• Tidy up. Ask "tidy up the Sales dashboard", or "add these three widgets and tidy up". Chat arranges the widgets you see in their order, as the Tidy up button does: rows of two or three at random, so asking again gives another arrangement. Ask for a number per row ("three per row") to get rows of that size. This arranges the dashboard in Heym; Heym Work keeps its own arrangement of a page, which the page's own Tidy up button changes.

New widgets go below the dashboard's existing widgets and, like every widget, run as the dashboard owner. Adding, changing, renaming, deleting and tidying need write access to the dashboard. Creating a dashboard and building or changing widgets need a chat that can build workflows: the Chat tab, Heym Work's Chat for editors, and heym_chat.

Editing the layout

Toggle Edit to enter edit mode, where you can drag widgets and resize them on a 12-column grid. Layout changes are saved automatically. Widget titles are editable inline from the widget header.

Caching and refresh

Each widget caches its last computed data on the server for a configurable time-to-live (TTL). While the cache is fresh, reopening the dashboard serves the stored result instead of re-running the workflow. The cache is replaced in place every time the data is recomputed (one cache per widget, plus one per record on a detail page), and it is automatically invalidated when you edit the widget's workflow.

Use a widget's Refresh button (or Refresh in the toolbar) to bypass the cache and recompute immediately.

Sharing

Owners share a dashboard from its settings, with individual users (by email) or with any team they belong to. Each share is Read or Write; when someone has both a personal and a team share, the stronger one applies.

Widgets always run as the dashboard owner. Whoever is looking, a widget uses the owner's credentials and global variables, and its runs appear in the owner's execution history and traces. Teammates therefore see the same data without access to the accounts behind it, and every viewer shares the one cached result per widget. Widgets that a write collaborator adds or clones belong to the owner as well.

Because of this, Write carries the same trust as sharing a workflow for editing: a collaborator with write access can change what runs with your credentials. Give it only to people you would let edit your workflows. A test run that a collaborator starts from the workflow editor uses their own credentials, as with any shared workflow; only the dashboard itself runs as the owner. Read grants no access to the widget workflows, and read-only viewers receive the chart itself but not the per-node execution details behind it.

Removing a share, or lowering it from write to read, takes effect immediately.

Related

• Chart Output node
• Analytics tab — built-in execution metrics (distinct from user-built dashboards)

Schedules and database diagrams

Choose Gantt in Add widget to show tasks over time with completion and dependency links, or Entity relationships (DB) to map tables, fields and their relationships. Both types use the Chart Output node, and are also available on Heym Work pages. When Work suggests widgets for a schedule or database schema, it draws previews from the proposed tasks or entities.