Skip to main content

Capability

A discrete unit of functionality the platform can grant, gate or expose to agents.

type: capability · status: published · abstraction: entity

What it is​

A Capability is a self-contained unit of business logic. Rather than just being an isolated action, a Capability is a complete package containing the specific data it owns (Facts), the operations allowed on that data (Actions), and mechanisms for handling heavy background data (Resolvers).

For example, consider a 'File Preflight' capability. It provides an Action to Run Preflight against a given File, accepting specific configuration settings. When triggered, the capability creates a Preflight Run record (a Fact) linked to that File, updating its status and a high-level summary as the check progresses. The rest of the workspace simply monitors this Fact.

To the rest of the workspace, a Capability's facts are strictly read-only. The only way to change them is by triggering one of its actions. This ensures no user, workflow, or AI agent can bypass business rules by directly editing a record.

Capabilities do not wrap ordinary Object and Record CRUD or the intrinsic mechanics of a Value Type. A Capability is warranted when behaviour adds business rules, specialist processing, provider integration or Capability-owned state — for example, File Preflight or Deep-Zoom Viewing.

Why it exists​

An operating system cannot just passively store data; it must execute governed work. Capabilities are the platform's toolkit for substantial business logic, specialist processing and provider integrations. When functionality is implemented as a Capability, human users, Business Flows and AI Agents invoke the same Actions and observe the same Facts under identical validation, permission and audit rules.

Relationships​

Outgoing

  • acts on → Record (many-to-many) — Capabilities read and mutate Records under permission.

Incoming

  • is exposed by ← Module — Modules make their functionality available as Capabilities.

Neighbourhood​

Stakeholder summaries​

  • Product — Define what the system can do and how it's exposed; who can do it is granted through Permissions roles, and agent-safe tiers declare what may run unsupervised.
  • Investor — The safety and monetisation layer: Capabilities are how functionality is licensed, and how AI is kept inside the lines.
  • Marketing — Automate with confidence. Your AI assistants operate under exactly the same permissions as your people.
  • Engineering — Registry of invokable actions with permission checks, audit logging and tool schemas for agent exposure.