Build a workflow
You build a workflow visually on the canvas. This page covers creating one, adding and connecting nodes, and configuring each step — or you can simply describe what you want and let Ellie build it for you.
Create a workflow
From the Workflows list page you have three ways to start:
- Create Empty Workflow — an empty canvas with just a Start node.
- Create from Template — copy a ready-made workflow and adjust it. See Templates.
- Import — load a workflow someone exported as a file. See Export & import.
Build with chat
Prefer to describe what you want rather than place every node yourself? Click Chat on the canvas to open a split view where you tell Ellie, in plain language, the workflow you need. Ellie plans the steps, adds and connects the nodes, and writes each node's instructions for you — showing its progress as it goes so you can follow the reasoning. While it's building, the canvas is read-only; when it's done, you can run the workflow or keep refining it by hand.
Keep the conversation going to shape the result. Small tweaks apply quickly — "rename this node," "connect the diagnostic to the planning step," "add a summary step at the end." You can also just ask questions about the workflow, or ask Ellie to review the graph and say what's missing.
When a change would alter or remove a node that's already there, Ellie asks you to approve it first, so nothing you've written gets overwritten without your say-so. The approval sticks around until you answer it — you can leave and come back.
Build or edit
The split view has two modes, and you switch between them at the top:
- Build with Chat — change the workflow itself: add, connect, rename, and configure nodes.
- Edit Artifacts with Chat — change the reports a run produced, in plain language. See Artifacts & exports.
Both live in the same conversation, so you can build a step, run it, and refine its output without switching context.
Chat sessions
Each workflow keeps a history of chat sessions. Use New Chat to start a fresh one, or the session switcher to reopen an earlier conversation; you can delete any session you're done with.
Sessions are private to you. If you share a workflow with a colleague, they get their own chat — neither of you sees the other's conversation.
Describing a workflow in chat is often the fastest way to a first draft. Once the shape is right, open the Inspector to fine-tune each step's inputs and Knowledge.
Add nodes
Click Add node to open the Node Library. It's a browser: the list of available steps on the left, a full preview of whichever one you've selected on the right.
The preview tells you what you're about to add before you commit — what the step is for, an example of how to prompt it, and badges for what it can use (AI Chat, Files, Web links, Artifacts, Framework). Page through neighbouring nodes with the arrows to compare. When you've found the right one, click Add to canvas.
Keep the library open
If you're placing several nodes at once, click Dock to side to turn the library into a slim rail down the left of the canvas. From the rail you can drag nodes straight onto the canvas and drop them where you want them, which is faster than adding one at a time. Collapse the rail to icons when you need the space, and JobKred remembers whichever arrangement you left it in.
Every workflow starts with a Start node. From there, add the specialists you need — for example a Workforce Diagnostic node, then a Skills Diagnostic node, then a Career Pathing node. Each node corresponds to an agent in the Agent Catalog, so you can read exactly what each one does and what it produces.
Connect nodes with edges
Drag from one node's handle to another to create an edge. The edge sets the order and passes the first node's output into the next as input — so a diagnostic's findings can flow straight into a planning step.
- Parallel branches are allowed — two nodes can both read from the same upstream node and run side by side.
- Loops are not — a workflow must always flow forward. JobKred will reject a graph that loops back on itself when you save.
Once the graph is wired up, use Reorganize layout on the canvas to tidy it — nodes are rearranged into clean left-to-right layers so the flow is easy to follow.
Configure a node in the Inspector
Click any node to open the Inspector, where you set it up. It has two tabs:
- Inputs — the text inputs the node needs, which Framework it should measure against, and which of the workflow's Knowledge this step should use. See Add Knowledge.
- Outputs — the reports this node will produce, and the version history of each one once it's run more than once.
Below the tabs, Connections always shows which nodes feed into and out of this one. At the top, Run Node executes just this one step to test it, without running the whole workflow — it turns into Cancel while the step is working.
Drag the divider between the canvas and the Inspector to give either side more room.
Guided Inputs
Not sure what to type into a node's inputs? Use Guided Inputs — a conversational helper that asks you questions and fills in the node's inputs for you. Your answers are saved with the node, so you can pick the conversation back up and refine them later.
Node Settings
For finer control, open a node's Settings to adjust how that step behaves — the instructions its assistant works from, and which artifacts it produces.
Use Run Node to check a single step produces what you expect before you wire up the rest of the graph and run the whole thing.
Saving
Your work autosaves as you edit — you'll see a "Last saved" indicator. Node titles and descriptions are editable inline, so you can rename steps to match your process.
Autosave pauses while you're on the Activities tab, so browsing your version history can never overwrite it. Switch back to the Workflow tab to resume editing.
Next step
Give your workflow the data it needs → Add Knowledge.