pipelinesmini guide · 10 min

build a pipeline

chain processes you already have into one production line, fed by a single input form.

One process makes one kind of thing. A pipeline chains several, so the results of one become the inputs of the next, and a single form at the front starts the whole line. Here is the Bagelle line: an inspiration picture goes in, and six processes later a ring, its stack render and a showcase video come out.

six processes, one input form, one output.

before you start

  • At least two processes you have already built and run. A pipeline only references processes, it never creates them. Make them first: make a process.
  • Each process run once by hand, so you know what it publishes. The card learns how many results a run emits from the process's last run.

build it

  1. Create the pipeline. On the dashboard open the Pipelines tab and press New pipeline. Name it. You land on an empty canvas with a dock along the bottom that has pipeline items only.

  2. Add the first process. Open the + in the dock and choose Add process. Search across all your projects (Search processes across your projects…) and click a result. A card appears with the process's name, its resident pixie, a from project link back to where its runs live, and a port for each of its inputs.

  3. Add the front door. + then Input adds the pipeline input node. Its Add field button makes a form field: a label and a type of text, url or image. These fields are what you fill in at the start of every session.

  4. Wire the input into the card. Drag from a field's handle to the matching input port on the card. The port's chip changes from "not wired" to the field's name. If you would rather not wire by hand, the lightning-bolt on an unwired port creates a matching field and wires it.

    An input you don't wire can be filled by typing straight on the card; it then counts as set. Click an input's name on the card to flip it between required (*) and optional. A session won't start while a required input has nothing behind it.

  5. Chain the next process. Add a second card and drag from the first card's right-hand handle to one of the second's inputs. The first card's published results now feed that input. A card fed this way waits for its upstream card; it never starts on its own.

  6. Decide what happens with several results. If a process published more than one result in its last run, the card shows a split chip. Pick split (each result starts the next step separately) or combine (the next step receives them all as one set). That is the fan-out: one run's 5 results become 5 downstream runs, or one run of 5 inputs. sessions shows it running.

  7. Collect the deliverables. + then Output adds an output node. On each card, the chip reads internal or in output; click it to include that card's results in the output. The output node gathers those results for the selected session.

  8. Tidy and start. The grid button in the top bar lays the flow out left to right. Then press Start on the bar: that is a session.

declared output ports

A card lists each declared output as its own row with a handle on the right. The last row, all results (or other results), takes everything. Ask the process's resident pixie to declare outputs; you can't add them on the card.

what just happened

A pipeline is a graph of cards that each point at a real process. When you start a session, the input node's values go to the cards you wired it into. Each of those spawns an ordinary run of its process, with the same resident pixie and the same review rules, and shows up in that process's own Runs table too. When a run publishes, the pipeline hands its results to the cards downstream.

take it further

  • Talk to Pip, the pipeline's voice guide, to build all this by speaking: see sessions.
  • Put an approval where you want a person to look: approvals.
  • The same dock also has Sticky note, Postie and Table for notes, posting and rows.
  • See a real one end to end: the Bagelle pipeline.

checked against the product · 2026-10-23