runs

A run is one pass of a process: a full copy of the process's canvas, worked by its resident pixie with one set of inputs. The Runs table on the process node is where you start them, find them and reopen them.

the runs table

The Runs panel hangs under the process node. Each row is a run: its name, one column per input, a Results cell and three small buttons.

• Search. A search box (Search runs…) and page controls appear once the table outgrows one page of 5 runs.
• Rename. Double-click a run's name, or use the pencil that shows on hover. Names stop at 40 characters. The pixie names a run itself as its first move, so most runs arrive already named.
• Run again. The play button at the end of a row starts a new run with that row's inputs. Hover it to see "Run again with these inputs". Only one run can be starting at a time.
• Delete. The bin on hover deletes the run and its workspace canvas, pixie chat included. Results it already published stay where they landed.
• Collapse. The header chevron folds the whole table away.

drafts: add row

Add row puts an empty draft at the top of the table, with an input cell per field. Fill the cells, then press that row's Run. A draft costs nothing until you run it, and you can keep several open to line up a batch. Discard this draft removes one. Add row is greyed out when the process declares no inputs.

enter a run

Click a run's name to open its workspace. It is a canvas of its own: the cloned work, the pixie in the dock, and the full chat. A spinner beside the name means the pixie is working; the tooltip says so.

You can talk to the pixie there: ask for a change, answer a question it raised, or tell it to stop. The speech-bubble button on a row opens the same chat without leaving the host canvas.

what finished means

A run has only two states: working and idle. There is no "completed" badge, because a pixie can always take another message. A run is finished when the pixie is idle and has published results.

So an idle run with a dash in Results is not done: it may be waiting for you, or it may have stopped without delivering. Open it and look.

the attention dot

When a pixie wants you, for a question or for results waiting on your approval, a dot appears on the process node and on that run's speech-bubble button. Hovering shows the pixie's note, such as "ready for review". The dot clears when you read the chat, or when you approve the staged results.

where results land

• The run's Results cell. Up to three thumbnails and a +N count. Click the cell to expand the run's results under the row, with the request it was given.
• The node's tray. Published results from every run collect in the tray under the process node, grouped by the output they were filed under. Hover a result for its model and inputs, or pop it onto the canvas with Add to canvas.
• Downstream. Wire the node's output handle into another node or a pipeline and the published results flow on.

Nothing reaches these places until the run publishes. With the default human review, that means you clicking approve: see review and improve.

closing the tab

A run keeps going if you close the tab, lose the connection or log off. Come back later and the row shows where it got to. If it needs you, the dot is waiting.

limits & good to know

• Model runs are capped at 500 per user per rolling 24 hours. The cap counts model runs, including those a pixie starts for a process run. Failed runs and the pixie's own chat turns do not count. At the cap, new model runs are refused until older ones age out.
• Run names: 40 characters.
• Deleting is permanent. There is no undo.

related

• make a process — build the thing you are running.
• review and improve — approve, reject with notes, fold a run back.
• sessions — runs started in bulk by a pipeline.