What your workflow knows.
A knowledge set is a container of related things the workflow can reference. Each item inside is one piece: a taxonomy entry, a prompt, an example, a lookup row. The types are yours to define.
US Invoice taxonomy
124 entries · org-scoped · v3
Approved vendors
4,820 rows · project-scoped · live
Features bring the right knowledge to the right content.
Content picks up features as it moves through the workflow: "US invoice", "vendor: ACME", "high value". Each knowledge set has an expression saying when it applies. When the expression matches, the set's items attach to the content object.
AND country == "US"
INV-2026-04812
EXPRESSIONSAND / OR / NOT, over feature names
Set an expression on a knowledge set in the UI or via API. The evaluator walks the tree against the content object's current features and decides whether the set applies.
FEATURESContent-addressable, shared
A feature is a typed fact about a content object: same value, same feature, every time. They accumulate as the workflow learns more about the content.
Before every step, the right knowledge is there.
When an activity plan runs, an assessment evaluates the content object's current features against every knowledge set. Matching items get attached so the step can see them. As steps finish, new features may emerge, and the next assessment picks up new knowledge.
Parse PDF
Classify
Extract attributes
Risk review
LOOPThe loop closes itself
A script step can set_feature() on the content object. The next assessment picks it up, more knowledge sets match, and the step after sees the richer context. Knowledge responds to what the workflow learns.
AUDITSnapshots, not surprises
Every knowledge set is versioned. The audit trail records which snapshot a content object actually saw, and when. When the knowledge evolves, old applications are superseded, never silently overwritten.