Templates

​

Overview

A template is the reusable structure of a pipeline — the goal it pursues, the inputs it consumes, and the directed graph of source, transformation, and output nodes that turn data into derivations. Templates are authored once and run many times. Each run is a fresh execution of the same template against a specific set of inputs; see Runs. For example, a template might take weekly MES exports as input, compute per-station cycle times, and land them as simulation model parameters — and you can re-run it every week when fresh data arrives. Templates and runs are two views of the same pipeline. The template is the contract — the structure, the goal, the gate definitions. A run is where concrete values land against that contract.

​

Opening a Template

Click the Data icon in the left sidebar to open the Pipelines workspace. The canvas shows the currently selected pipeline. Use the mode picker in the top bar to switch between editing the template and viewing a specific run; it defaults to either Template or a run date (e.g. Apr 24, 2026). Selecting Template from the dropdown puts the canvas into template-edit mode. Use the pipeline picker to the left of the mode picker to switch between pipelines or to create a new one (+ Create new pipeline).

​

The Template Panel

When the canvas is in template mode, the left side panel shows the editable template metadata:

• Goal — a short paragraph stating what the pipeline is trying to derive. This isn’t just a label: Dexter actively uses the goal at run time to scope exploration and ground extraction decisions, so a clear goal steers your runs toward the parameters you can realistically extract from the inputs.
• Inputs — the sources the template consumes. Each row shows the source name and a hover-revealed delete control. Click + Add input to attach a new source: choose Upload a file (CSV, XLSX, PDF, or another file the agent should inspect) or Connect an integration (e.g., Salesforce, SAP). Connected integrations are managed via your ProDex team — see Data Sources.

​

The Graph

A template is a directed graph with three node types, connected by edges that flow from sources through transformations to outputs. Source — the entry point. References an input file or connected source and gives Dexter instructions on how to read it. Source nodes should always carry a gate that verifies input shape (columns present, values in range) before downstream work proceeds; Dexter adds one automatically when building a pipeline. Transformation — an intermediate step. Filters, aggregations, unit conversions, joins. Each transformation has a description that combines narrative instructions and (typically) a fenced Python code snippet showing how the transform was performed. Gates are optional on transforms — add one when the transform’s output drives a high-stakes downstream value. Output — the exit point. Output nodes collect derived results. Gates are not allowed on outputs; by the time a value reaches the output, the corrective lever is upstream. Each output node is one of two types:

• platform — the output feeds values back into another ProDex app page (Modeler, Planning, BOMs, BOM Configurator, Insights). The run detail surfaces an Open in [app] button to navigate to where the value lands.
• report — the output is a tangible downloadable file. The output node carries a required format (e.g. xlsx, pdf), and once the run completes the run detail surfaces a download link to the produced file.

​

Editing a Node

Click any node on the canvas to open its editor. Source nodes

• Name — short label that appears on the canvas and in run history.
• Details — source type (uploaded file vs. connected integration) and format (CSV, XLSX, integration name, etc.).
• Notes — free-text context Dexter and future viewers can reference; useful for recording quirks of the source data.
• Gate — Name, Description, and Instructions listing the checks Dexter must perform. Instructions render as Markdown in the chat panel, so formatting them as a list works well.
• Example data — a small preview of the bound input with copy / download / fullscreen controls so you can sanity-check what’s flowing in.

Transformation nodes

• Description — the narrative that explains what this step does. The fenced Python code block at the end is part of the same description field; the UI separates the prose and the snippet visually, but editing one edits the same field. Code blocks have download and copy controls.
• Gate (optional) — same Name / Description / Instructions structure as a source gate. Add one when the transform’s output drives a high-stakes downstream value.

Output nodes

• Name and narrative description — what this output represents.
• Output type pill — platform or report (see “The graph” above for what each does).
• Example value — a sample of the value this node will produce, so reviewers can see what to expect before running.
• Derivations — the produced value’s full lineage; click the node header to expand.

​

Running a Template

Click the Run Pipeline button in the top bar to kick off a new run against the template’s bound inputs. While a run is in flight:

• Dexter pauses at each gate and asks for confirmation in the chat panel. Confirm to unblock the downstream nodes; Reject if something’s off (wrong shape, bad values, missing context). A rejected gate blocks all downstream nodes — fix the input or the transform, then re-evaluate the gate or re-run the pipeline. If a gate stays rejected, the run finishes as completed_partial with whatever reachable values were still produced.
• The Run Information panel on the left tracks progress: inputs bound, gates resolved, results produced, last update.

The result is a Run: a record of every input consumed, every value produced, and every gate confirmation — an audit trail you can review and compare against future runs. See Runs for the run-detail surface and how derivations are corrected.

​

Re-Running

Templates are designed to run many times. When fresh data arrives, upload the new files (or refresh the connected source) and kick off another run from the same pipeline picker — the template doesn’t change, just the inputs. Past runs stay accessible from the mode picker for side-by-side comparison.

​

What’s Next

• Runs — reviewing what a run produced and how to correct values
• Gates — the confirmation flow in detail (states, what to do when you reject, when to add gates to transforms)
• Data Sources — the integration catalog and how connections are set up
• Insights — turn pipeline sources or outputs into chart dashboards your team can navigate to