Skip to content

Knowledge & Capabilities

Knowledge & Capabilities define what a job can safely use. Knowledge contains approved source material. Capabilities are approved actions, integrations, and tools that can fetch live data, save a record, call an approved service, or operate a connected website. Accesses are the connected accounts those capabilities may need.

The Job owns inputs, outputs, handoffs, and versioned work-step instructions. Knowledge & Capabilities define the safe boundary.

Functions and capabilities can be used directly by a Job or from an AI work step. A Job does not need an AI step to call a deterministic function, send an output, or use a website publish destination.

Use Accesses for connected accounts such as Google Drive, Gmail, Calendar, bexio, websites, or approved API accounts. A function can require an Access, and the Job chooses which approved Access it may use.

Start with the job boundary

Before adding a capability, decide:

  • What the job is allowed to do.
  • Whether it is read-only or writes data.
  • Which fields the job can send.
  • What confirmation is needed before a change.
  • What error message should be shown to a user.

The capability definition should be narrower than the underlying system. If the job only needs to create leads, expose a create-lead function instead of a general database endpoint.

Common capability patterns

Pattern Example First-pass safety rule
Lookup Check appointment slots or order status. Read-only and scoped to one record type.
Connected account lookup Read contact summaries or approved records from a connected business account. Start read-only and expose only the fields the job needs.
Capture Save lead details from a chat. Require the user to provide contact details first.
Draft Create unpublished content. Do not publish automatically.
Update Change a website content field. Confirm record and field before writing.
Connected Data Read site content, create a draft, update approved fields, or publish reviewed content. Validate and sync the website first, then require review before publish.

Functions and integrations are reusable. Add the capability to the job step that needs it. If the capability needs a connected account, choose the matching Access. When one already exists, TROPIKAL should prefill it. If none exists, add the Access from the setup flow and return to the job after connection.

If a capability should be available to an AI step, grant the capability and needed Access to its Job.

Test with one request that should use the capability and one request that should not.

Write clear tool descriptions

The model chooses capabilities from names, descriptions, and parameters. Use plain descriptions:

{
"name": "create_lead",
"description": "Save a new sales lead after the visitor provides name, contact method, and interest.",
"parameters": {
"name": "string",
"contact": "string",
"interest": "string"
}
}

Avoid generic names like call_api or run_action. They make capability choice less predictable.

Handle tool failures

Tell the AI work step how to recover:

  • If a required field is missing, ask a follow-up question.
  • If a resource is unavailable, explain what failed in user language.
  • If a connected account needs reconnecting, reconnect the Access instead of broadening the job’s permissions.
  • If a write action is denied, do not retry with broader permissions.
  • If a destructive action is requested, ask for explicit confirmation.

Connected Data

When a website is connected, the Website dashboard can sync approved business objects into jobs. A job can then read approved website information, create drafts, or publish reviewed content when the website exposes those actions.

If Connected Data is not available:

  1. Open the Website dashboard.
  2. Validate the website connection.
  3. Sync Connected Data.
  4. Confirm the needed action is listed.

Update the website connection or contact TROPIKAL support rather than broadening the job beyond what it needs.