Templates and sharing

Start from a template

A template is a ready flow for a common job: importing data, labeling with review, training a detector, evaluating or comparing models, checking dataset quality. It contains the nodes, their settings and the wiring, but none of your data — you point it at your own datasets when you create a flow from it.

On the Home page, click Start from template. The library has two tabs:

• Templates — flows to start from. Filter by task, action, model or dataset, or check Recently used and My templates. The OvalBee logo marks an official template from the OvalBee library, and a globe marks one that another workspace of your installation has published.
• Case studies — write-ups of complete projects built on these flows, with results and the reasoning behind each setting.

Select a template to read what it does and what it needs, then click Create flow from template.

Some templates bring a short step-by-step setup that asks only what the template can’t decide for you, such as the ontology to label with and who labels and reviews. It also lists every dataset the flow will create: where items come from, the working datasets the flow moves them through, and where finished items end up. You can rename any of them, and pick a dataset you already have for the input and the result. Fields you haven’t filled in are highlighted when you press Next. Your answers are kept, but nothing in the flow or your workspace changes until you press Start: it creates the ontology or saves your changes to it, writes the answers to the nodes, creates the datasets with the names you gave them (each dataset node takes its dataset’s name), then runs the nodes straight away instead of waiting for their schedule. A new flow stays paused until then. Later closes the setup and leaves a Finish setup bar on the canvas to come back to. Saving such a flow as a template keeps its setup.

Inside an open flow, Templates in the toolbar adds a template to the current flow instead: it arrives as a group you move into place and click to drop. A template with its own setup opens it once you drop the group.

Once the group is dropped, OvalBee looks for datasets and outputs the flow already has that fit the template’s empty inputs, and offers to connect them. For example, a live training template gets your existing labeling pool and next-batch dataset. This works before any dataset is picked, too: in a flow built only from templates, a Dataset node counts by its name, and an empty “To dataset” already in the flow can point at the template’s first Dataset node, so a labeling block dropped next to a data intake block is fed by its sampling step. Each suggestion also adds the edges that follow from it, such as an edge from the node that fills a dataset to the template node that reads it. Rows whose names clearly agree come ticked. Models and reports are only suggested, never ticked for you, because a name can’t tell which model a template needs. The Auto-bind inputs flow setting applies here too: Ask shows the list, Always connects the ticked rows without asking, and Never turns the suggestions off.

The Use cases section walks through the templates by goal.

Save your own templates

Turn a flow you’ve refined into a starting point for the next project:

• More actions → Save as template… saves the whole flow.
• Select some nodes and click Save selection as template in the toolbar to save just that part — a labeling block, an evaluation branch.

Templates are portable by design. They keep nodes, settings and wiring, and leave out anything tied to one workspace: selected datasets, secrets, models, people and run results. A reference from a saved node to a node outside the selection would break the template, so OvalBee refuses to save it and tells you which reference to fix. Your templates appear under My templates in the library.

The save dialog also takes:

• Save to — keep New template, or pick one of your workspace’s templates to replace its content with this flow. Its name, description, tags and README are filled in so you can adjust them.
• README — an optional Markdown page shown next to the template in the library.
• Publish to every workspace — a published template appears in the library of every workspace on the installation, and anyone can read it without signing in. On a self-hosted installation anyone who can edit flows can publish; on OvalBee Cloud only an instance administrator can publish, or change and delete a published template.

Edit details in a template card’s menu changes the same fields later, without touching the flow inside.

Export and import

More actions → Export flow… offers two kinds of file:

You can also export the canvas as a PNG image for slides and documents.

To bring a flow in, use Import from file on the Home page. For a backup, the import dialog lets you name the new flow and decide what happens to its references: clear the datasets that don’t exist in this workspace, keep or clear them all, or reset every input to its default. You can also reset the run state and re-link secrets by name. A template file becomes a new flow directly.

Duplicate in a flow card’s menu on the Home page makes a copy of a flow in the same workspace.

Share a flow with people

Everyone in a workspace sees its flows according to their role. To give someone access to one particular flow — a collaborator who shouldn’t see anything else, or an operator who should be able to edit this one — use Share… in the flow’s card menu or in the editor’s More actions menu. Choose Can view, Can run or Can edit. Sharing only ever adds access on top of the person’s role, and you can grant at most the access you have yourself.

Per-flow sharing is available on plans that include workspace roles.

Public examples

An instance administrator can publish a flow as a read-only public example with its own link. Visitors can explore its nodes, results, reports and data without signing in, but can’t edit or run anything. Secret values are never published, but outputs are — datasets, reports and models the flow produced are part of what visitors see, so publish only flows whose data is meant to be public. Labeling and Review cards show their live counts, with people listed as Labeler 1, Reviewer 1 and so on rather than by name. Publishing is limited to flows from one workspace chosen by the administrator.

Related pages

• Use cases — templates organized by goal.
• Flows — flow setup and versions.
• Members and roles — who can see and change what.