TrainAR Docs
Try Free
Skills & Knowledge · Last updated 18 May 2026 · 3 min read

What are skills?

Skills are the actions the AI assistant can take during a live training session.

Skills are the actions the AI assistant can take during a live training session. When an engineer asks a question or requests something from the glasses, the AI decides which skill to call, calls it, posts to the skill's webhook, and waits for the response — then speaks or displays the result.

Skills tab

What a skill is

A skill is a registered, named, described webhook the AI can invoke at runtime. Each skill row in your tenant has:

  • A name the AI uses to identify it (snake_case, e.g. look_up_warranty_status).
  • A display name shown in the Dashboard.
  • A description that tells the AI when it should call the skill — this is the most important field, written in plain language.
  • A parameters schema (JSON Schema) that defines the arguments the AI fills in from the conversation.
  • A webhook URL — your HTTPS endpoint that receives the call.
  • An HTTP method (POST or GET).
  • An auth header (optional) for your endpoint to verify the call came from TrainAR.
  • A timeout in milliseconds.
  • Response handling — how the AI uses what comes back (see Skill response fields).

How a skill is called during a session

  1. Engineer says something to the AI — "What's the warranty on this boiler?"
  2. The AI scans the available skills for ones whose description matches the intent.
  3. The AI fills in the parameters_schema from the conversation context (e.g. extracts a serial number).
  4. The AI POSTs the JSON payload to the skill's webhook_url and waits up to timeout_ms.
  5. Your endpoint does whatever it needs to (lookup, calculation, write to your CRM, etc.) and returns a JSON response.
  6. The AI interprets the response — speaks it, displays it on the glasses, or both, per the skill's response configuration.

Where you manage skills

Dashboard → Skills & Knowledge → Skills tab. Admin role required. You'll see your tenant's skills listed, plus any platform skills entitled via Marketplace bundles (those are read-only).

From here you can:

  • Create a new skill — see Creating a custom skill.
  • Edit an existing skill — change description, parameters, response handling.
  • Test a skill against your endpoint — see Testing a skill.
  • Toggle active/inactive without deleting.
  • Assign to specific seats or org-wide (the assignment_scope column).

What skills are NOT

  • Skills aren't knowledge. Knowledge is documents the AI searches for content. Skills are actions the AI takes. A skill might use your knowledge inside its implementation, but they're separate concepts. See Knowledge overview.
  • Skills aren't MCP tools. Skills are TrainAR's webhook-based action protocol. MCP Tools are tools the AI gets by connecting to third-party MCP servers. See MCP Tools overview.
  • Skills aren't bundled with the platform automatically — except for a small set of system skills (search_knowledge, capture_camera) that every tenant has. Everything else is either a platform skill entitled via a bundle, or a tenant skill you author.

Where to next