Workflows overview

Video walkthrough: authoring pipelines from templates, configuring per-node models and knowledge layers, and inspecting run results.

Workflow library — Each row is a saved pipeline. Draft, Published, Not paused, and Succeeded answer different questions.

Pieces

A workflow is a saved definition. You can author it, run it when your role includes workflow:run and workers are available, and read the result in the workspace. Output stays on the run unless you also queue a destination export. A failed run stays failed. It does not roll back documents already stored.

Templates

Workflow template library — Browse templates shows ready-made recipes. Create workflow on a card saves a workflow. It does not run the recipe.

Performance review evidence template — A template detail lists the canonical stages and the configuration still required, such as a collection, a schema version, and a result destination. Human review on a recipe is decision support, not an automatic decision.

The canvas guide shows the graph and the Add Node menu. Index a collection, RAG chat, and summarize a collection continue each recipe onto a canvas with a collection bound. Open Workflows, then Browse templates. Recommended recipes name their stages and the connectors they expect. Creating the workflow copies that recipe into your library as a definition you still configure and run.

Library

Back on the library, filter by type, status, trigger, and access. Owned by me and Favorites narrow the list. They do not change whether a run happened. Open workflow is how you inspect one definition and edit it on the canvas.

Runs

Last run, last run status, and success rate are the execution record. Manual trigger means you start it. Not paused means admission is open. A succeeded status on an older date is that run, not a new one. Non-retaining workspaces have no workflows.

Workflows overview

Pieces

Templates

Library

Runs

Related documentation