Scrydon
Ontology

Packs

Reusable ontology bundles — object types, link types, action types, and bindings — distributed through the marketplace.

A pack is how Scrydon ships organisation defaults — a reusable, opinionated preset that an org installs once and everyone inherits. Today a pack bundles ontology (object types, link types, action types, identity rules, and bindings); the same mechanism increasingly carries the defaults for the surfaces those Objects bind to:

  • Knowledge base — the knowledge-base pack wires uploaded pages and managed tables to typed Objects.
  • Process flows — the process-flows pack ships typed Objects for ISO 27001 / 42001 review and audit cycles.
  • Data sources (soon) — default bindings against the managed tables an org ingests.

Think of a pack as "the typed world an organisation starts from", distributed through the platform marketplace and installable in one click. First-party packs and customer-authored packs use the exact same artefact and install path.

What's in a pack

ElementPurpose
Object typesThe typed entities the pack defines.
Link typesThe typed relationships between them.
Action typesThe typed mutations the pack exposes.
BindingsDefault bindings (with logical source names that the installer maps to local tables).
ManifestPack metadata — id, version, description, dependencies, classification, license.

When you install a pack, the platform creates everything on a fresh main branch in your organisation. Bindings that ship with logical source names need to be mapped to your workspace's managed tables on first use.

Built-in packs

Scrydon ships a small number of packs out of the box. Common ones:

  • knowledge-base — links knowledge-base pages and managed tables to typed Objects.
  • process-flows — typed Objects for ISO 27001 / ISO 42001 process tasks (review cycles, audit cycles).

These are installed automatically the first time you open the workbench.

Domain packs

Domain packs are pre-built for specific verticals:

  • fraude-intelligence — financial-services AML/fraud-detection objects: Transaction, Customer, SuspiciousActivityReport, RegulatedEntity, Supervisor. Six pre-built bindings against typical silver tables.
  • nato-maritime — maritime-defence objects: Vessel, AISContact, Incident, Region.
  • iso-quarterly-review — quarterly compliance review tasks.
  • iso-yearly-review — yearly compliance review tasks.

See Examples for end-to-end walkthroughs of these packs.

Installing a pack — two-stage model

Pack install splits in two: an org admin admits the pack to the org's catalog (Stage 1), and workspace users pick what to materialize in their environment (Stage 2). Ontology content is org-wide and lands at Stage 1; data sources are per-environment and land at Stage 2.

Stage 1 — Admin uploads the pack (org-wide)

From Settings → Packs:

  1. Drag-drop your .bundle.tar.gz.
  2. Review the bundled ontology + data-source contributions.
  3. Click Publish.

The platform admits the pack to the catalog. Any ontology contributions are installed immediately org-wide (object types, link types, bindings appear on main). Any data-source contributions are catalogued but not yet materialized — no data_source row, no ingest schedule, no Dapr tick.

Each catalog row summarizes what the pack contributes as colored badges — N ontologies, N workflows, N process flows, N data sources, N KB domains. Click a badge to see the detail behind the count (workflow names and block counts, data-source vendors and columns, KB domain slugs); click the row itself to open the full pack detail sheet.

Stage 2 — Workspace user instantiates a data source

From the Marketplace in apps/analytics, while in a workspace + environment:

  1. Pick a data source from a catalogued pack.
  2. Click Install in this environment.
  3. The data source materializes in your env and starts ingesting on schedule. Operate it from Analytics → Data Sources.

Data-source-only packs (no ontology) are a no-op at Stage 1 beyond catalog admission — the upload completes silently and the data source appears in the marketplace for users to pick.

Packs with only workflow / process-flow content still install via the agentic Process Templates / Workflows pages.

Mapping bindings to your data

Most packs ship bindings with logical source names — e.g. "silver_table:regulated_entities". After install, you map each logical source to a real managed table in your workspace:

  1. Open the Bindings tab.
  2. Find a binding showing Not ready — table not registered.
  3. Click Pick table and choose the matching managed table.

The platform highlights exact-name matches so the picker is nearly one-click for the common case.

Authoring a pack

Packs are authored with the Ontology Authoring SDK. The SDK lets you:

  • Define object / link / action types with TypeScript + Zod.
  • Bundle them with bindings, manifest, and assets.
  • Build a .bundle.tar.gz that's uploadable through the marketplace.

The same artefact ships first-party packs and customer-authored packs — there is no separate path.

Pack lifecycle

End-to-end, a pack travels from a TypeScript project to typed entries on main in your workspace. The platform validates at every hop; bindings that need workspace-specific table mappings remain pending until an admin completes them.

The split between Apply ontology, Apply bindings, and Apply workflows is real — each kind has its own applier inside the platform, so a pack can ship any subset (ontology-only, bindings-only, or a full vertical preset).

On this page

On this page