When a Mind works with workflows, every step appears in the chat as a card, and the workflow itself opens in a window over the conversation. This page covers those cards, the window's Jobs and Diagram views, and the states and notices you may read. For the Workflows area itself, see The Workflows screen.
Workflow cards in chat
A workflow step renders as a card like any other tool step: click the card body to fold its details, and use the card's chip or corner button to open the step in Tool History (Tool cards and Tool History). Workflow steps come in three shapes: the workflow list, a workflow's plan, and a run, described below.
The list card collects the workflows the step found. Its title counts them, for example Workflows · 12, and it badges {n} running and {n} failed when something needs attention; while it loads, it reads Listing Warehouse workflows….
The list table covers Workflow, Id, Enabled, Schedule, Finalized / total, Last run and Last run at; the full view adds Duration and Can. Chips read Enabled or Disabled, and Scheduled or Not scheduled; a workflow that never ran shows never run.
The body sums the list up, for example "12 workflows · 2 running · 1 with a failed last run", with chips for {n} workflow(s), {n} running and {n} failed last run.
Workflows cannot be enabled or disabled from the chat: "Enabling or disabling a workflow happens in the Workflows page, never from the chat."
The card shows up to 12 workflows; the rest stay in the full view behind "+{n} more - open the tool view for the full list." With nothing to list it reads No Warehouse workflows.
A workflow's plan card
The plan card describes one workflow. Its title names it, for example Workflow · "Sales load", or shows the id when the name is not known; while it loads, it reads Reading the workflow plan….
Chips summarise the settings: Enabled or Disabled, Schedule on or Schedule off (with the schedule, and when the schedule's time zone differs from the server's, also its times on the server clock, as in 05:00 on the server clock, UTC), batch {n}, retries {n}, ignores errors; a live job adds run #{id} live and a status chip that reads Running.
A summary line counts the plan, for example "6 steps · 4/6 finalized".
Steps in execution order lists #, Step, Kind, Pattern, Finalized, Schedule and Depends on; the full view adds Table, Source system and Transformation id.
Only finalized transformations take part in the plan. The rest appear under Not in plan with the reason "not finalized - not in the plan".
When nothing qualifies, the steps table reads No finalized steps - nothing is in the plan.
A Lake workflow's plan card, titled for example Lake workflow · "Sales DB", counts tables instead, for example "914 tables · 905/914 deployed". Its steps are the workflow's tables, Pattern shows each table's sync type, and the Finalized column is Deployed, reading Deployed or Not deployed on each row. A table that is not deployed stays in the plan: under Not deployed, a line such as "9 of 914 tables are not deployed - a run plans them and skips them." counts them. With nothing to list, the steps table reads No tables to show.
A large Lake workflow arrives one page of tables at a time. A note above the steps, such as "Showing tables 1-50 of 914.", says which part you see, while the badge on Steps in execution order counts every table. The Diagram section appears only when one page holds the whole workflow.
Recent runs lists Run, Status, Started, Duration, Operation date, Trigger and Restarts (the full view adds Finished), and reads No runs recorded. when the workflow has no runs yet.
A collapsible Diagram section holds the workflow graph, with Copy diagram source to copy it out. On the card it stays collapsed until you open it; in the step's full view it opens expanded.
A run card
The run card follows one run. Its title ends with the run, for example Workflow run · run 2317, and while it runs it reads Running… then Running for {duration}.
When it settles, the subtitle reads "{name}" · {status} · {progress or next}, and a chip counts {n} failed step(s) across the whole run, including failed steps the card does not list.
The status chip reads Running or Running Error while the run is live (with a spinner), Failed, Error or Running Error when it went wrong, and Unknown when no status was read.
Progress shows as a fraction with counters: {finished}/{total} steps done, plus {n} running, {n} failed, {n} skipped or {n} aborted where they apply, for example "3/7 steps done · 1 running". The progress bar shows the success share, with failed or aborted steps in red.
The run facts list Started, Finished, Duration, Operation date, Restarts and Workflow id.
Failed steps lists up to 10 failed steps with their error text, and the first stack opens expanded; a failure without text reads No error text recorded. Its badge counts every failed step of the run, and when more failed than are listed, a note such as "Showing failed steps 1-10 of 23." says so.
Skipped steps by reason counts the skipped steps per reason, such as Table Is Not Deployed or Synchronize Option Is Disabled. It shows up to 10 reasons and gathers the rest under "Other reasons ({n})"; a skip with no recorded reason reads "no reason recorded". Its badge counts every skipped step.
Steps lists Status, Step, Kind, Duration, Rows and ETL log id, plus a Note column with the skip or failure reason when a listed step has one; the full view adds Started, Finished and Step id, and says No steps recorded for this run. when the run has none. The badge on Steps counts every step of the run, not only the listed ones.
A run of more than 50 steps lists only the steps that did not finish, 50 at a time, with a note such as "Showing unfinished steps 1-50 of 123 (finished steps are not listed; more on later pages)." When the Mind reads one kind of step, the note names it, for example "Showing failed steps 21-23 of 23."
Step kinds are Transformation, History, Gate, Init and Other. A History step is a platform-generated history load: it has no SQL and cannot be edited.
A step that loaded data chips its log as etl log {id}; a run that is still going carries a live chip.
A run with failures opens its card by itself so you can read what went wrong, and a Next: line carries what the Mind will do next.
If a started run has no id yet, the card chips run id not resolved yet.
When a wait gives up, the card says Still running: the agent will keep waiting: the run continues even though the Mind stopped waiting. Watch it from the card, or stop it if you may (Stop and abort runs in chat).
The card keeps up to 12 steps; the rest sit in the full view behind the same "+{n} more - open the tool view for the full list." line.
A Workflow or Workflow run chip in a message opens the workflow in a window over the chat, so the conversation stays where it is; the chips themselves are covered in References in chat.
Chip
What opens
Workflow
The window on Jobs, scoped to that workflow.
Workflow run
The window on Diagram, over all runs you can read.
The window offers two views as pill tabs: Jobs and Diagram.
Jobs embeds the same jobs table as the Workflows screen. Double-click a row to open that job's nodes. The window keeps its own column layout, so resizing or hiding columns here does not change the Workflows screen table (Job runs).
Diagram shows the workflow as a graph and stays current: while a run is active it refreshes about every 5 seconds. The diagram is built the first time you open it and stays ready while the window is open. For the diagram on its own screen, with run switching and export, see Workflow diagram.
One window serves every chip: opening another chip switches the window to that workflow or run and starts from that chip's view.
The window closes by itself when a popup opens over it, so nothing opens hidden behind it. Open the chip again to bring it back.
Who can open what
The window follows the access rules of the Workflows area (Workflow permissions): it lists the workflows you can read, and only those open.
Diagrams need a desktop screen, and a Warehouse workflow's diagram also needs read access to Warehouse tables. Without diagram access the window shows a single view: no tabs, only Jobs.
A Workflow run chip opens the run over all runs you can read; a Workflow chip scopes the jobs to that workflow; a chip that names a workflow but carries no id opens all jobs.
A workflow that was deleted, or that you cannot read, shows the same notice: Workflow not available, with "It may have been deleted, or you do not have access to it." Open The Workflows screen to check whether it is still there.
If the workflows cannot load, the window reads Workflows could not be loaded, with "Close this window and open the workflow again."
Questions, answered
Why did a workflow open on the Diagram instead of the Jobs?
The window opens on the view its chip points at. A Workflow run chip carries one run, so it opens Diagram with that run's graph; a Workflow chip carries a workflow, so it opens Jobs with its jobs list. Switch between the two with the Jobs and Diagram tabs.
Why is there no Diagram tab in the window?
Diagram access depends on your screen and your permissions: diagrams show on desktop screens only, and a Warehouse workflow's diagram also needs read access to Warehouse tables. Without it the window shows a single view, with no tabs.
Why does the window say "Workflow not available"?
The workflow behind the chip could not be opened. A deleted workflow and a workflow you cannot read show the same notice, so the window cannot tell them apart: "It may have been deleted, or you do not have access to it." Check The Workflows screen to see whether the workflow is still there and available to you.
Will resizing columns in the window change the Workflows screen?
No. The window's jobs table keeps its own column layout: resizing or hiding columns in the window does not change the table on the Workflows screen, and the other way around.
What do "3/7 steps done · 1 running" and the other counters mean?
They count the run's steps. {finished}/{total} steps done reports how many finished out of the whole plan, and counters such as {n} running, {n} failed, {n} skipped or {n} aborted are added for steps in those states. The progress bar shows the same picture, with failed or aborted steps in red.
What is a History step?
A step the platform adds to load history data. It carries no SQL and cannot be edited, and its kind chip reads History. The other kinds are Transformation, Gate, Init and Other.
What does "not finalized - not in the plan" mean?
The transformation in the Not in plan list is not finalized, so the workflow cannot include it in its steps. Only finalized transformations take part in the plan, so finalize the transformation first if it belongs in the run. A Lake workflow works differently: a table that is not deployed stays in the plan, its card lists it under Not deployed, and a run skips it until it is deployed.
Why can't I enable or disable a workflow from the chat?
Chat is read-only for workflow switches: "Enabling or disabling a workflow happens in the Workflows page, never from the chat." Open The Workflows screen and change it there.
Why did the workflow window close by itself?
The window closes when a popup opens over it, so nothing opens hidden behind it. Open the workflow or run chip again to bring the window back.
Why does a run card list only the unfinished steps?
The run has more than 50 steps, so the card lists only the steps that did not finish, 50 at a time, and the note above the table says so. The badges still count the whole run: every step, every failed step and every skipped step. Skipped steps by reason explains the skips, and the Mind reads further pages when it needs them.
Why does a run card say "run id not resolved yet"?
The step started a run, but the new run's id was not known when the card was drawn. The chip marks that gap; find the run on the Workflows screen (Job runs) and open it there.
What does "Workflows could not be loaded" mean?
The window could not load the workflows it needs for this view, so it cannot show the jobs or the diagram. Close it and open the workflow again from its chip, as the notice says.