Workflows and plans

What it is

A workflow is a versioned graph that turns a request into results: read from sources, shape the data, optionally structure it into ontology, and write to destinations. The graph is the plan — there is no hidden representation that matters more than what you see in Pipeline Studio.

What it holds

• Nodes of distinct kinds: reading from sources, transforming data, extracting structure from unstructured content, mapping between shapes, applying skills, predicting, writing to destinations, and writing into ontology. Kinds are not interchangeable — each declares what it needs and produces.
• Edges that carry data between nodes. An edge is valid only when the two sides' interfaces agree — structured versus unstructured, single record versus collection, known schemas, ontology compatibility.
• Validation state: issues scoped to the graph, a node, or an edge, at info, warning, or error severity. Errors block; warnings deserve a reading before you proceed.
• Policy: approvals and runtime gates authored on nodes and aggregated up — closest to the effect they guard, especially external writes.
• Versions: drafts, commits, branches, and compare. The working graph is an uncommitted draft until you commit it.

How it relates

• Views project the same graph per question: topology, data flow, permissions, health, outputs, AI reasoning. Switch views instead of guessing.
• Skills attach at workflow or node level and steer generation, refinement, and validation — always visibly, never silently.
• Runs execute committed versions; evidence links every output back through the graph to sources.
• Missing capabilities (a source, skill, or reference the draft needs) resolve inline from the proposal — install where the work happens.

What this means for you

• Challenge the AI's assumptions before editing nodes; misreadings fixed early never become structural defects.
• Treat validation errors as blockers and warnings as questions.
• Commit versions you can name; branch experiments you might regret.

Image: annotated plan graph labeling nodes, edges, validation state, policy, and versions.

Next

• Run one: Run your first pipeline.
• Every node and field: Workflow nodes.