Skip to content

Capabilities in the console

The Capabilities page (in the agent's primary navigation) is the roster of everything your agent can do beyond conversing: one card per capability with its status — Ready, Needs attention, or Off — and links to its configuration. The concept is covered in Capabilities; this guide covers the console mechanics: connections, secrets, custom integrations, and readiness.

Each capability configures on its own page, grouped under Settings · Capabilities: Scheduling, Representatives, People fields, Outreach, Tools & connections, Custom integrations. The dedicated guides cover the first four (Scheduling, People, Reaching Out); this page covers the rest.

Connections

Capabilities that act through an external account — Google Calendar for scheduling, Slack for notifications — need a connection. Settings · Tools & connections lists each item with its connection state:

State Meaning
Connected Working.
Expired The provider's token lapsed. Reconnect when you get a chance; the agent recovers automatically in most cases.
Revoked / Error The connection is broken and won't work until you reconnect. The page shows the provider's last error.
Not connected Never connected.

Connect starts the provider's sign-in flow; the resulting access belongs to this agent alone. Disconnect revokes access at the provider and deletes the stored credential; the item stays configured for a later reconnect.

For notification-style items you also choose the destination service — where Notify Operator lands (Slack, email), which calendar meetings write to. The agent's behavior never changes with the service; you can swap it later without touching any Playbook.

Secrets are write-only

API keys, webhook URLs, and auth headers are stored encrypted and never shown back. The console displays a saved secret as set, with a Replace action. Secrets never appear in an exported document.

Custom integrations

A custom integration is an ability backed by your own HTTP endpoint. Define it at Settings · Custom integrations:

  1. Name and description. The description teaches the agent when to use the ability; write it the way you'd brief a new teammate ("Looks up an order by ID. Use when a customer asks about order status, shipping, or returns.").
  2. Inputs. What the agent should gather from the conversation, as a small schema (orderId: string).
  3. Endpoint. Method, URL, and an optional authentication header (write-only).

At runtime, when the agent decides the ability applies, Anychat calls your endpoint with the inputs, hands the response back, and the agent continues the turn with the result.

Readiness

A capability with incomplete configuration is not offered to the agent — it fails visibly in setup rather than silently in conversation. Two surfaces tell you what's missing:

  • The Capabilities page's status per card ("no representatives — calls book with 'someone from our team'").
  • The readiness card on the agent's Overview, which flags breaks in the behavior graph — a Playbook referencing a Skill that doesn't exist, a required ability that isn't configured, no Default Playbook — with a link to the page that fixes each. A clean agent shows no card.