Platform

Connections & skills

Two ways to make the agent yours. Connections give it reach into the tools you already run your business on. Skills give it your playbook.

Connections

A connection lets the agent work inside an outside tool, with your permission. You grant each one yourself, and you can revoke it at any time.

  • Gmail — read and send email from a Gmail inbox.
  • Google Calendar — read and manage calendar events.
  • Google Sheets — read and update spreadsheets.
  • Slack — post and read messages in Slack channels.
  • Notion — read and update Notion pages and databases.
  • Linear — track and update Linear issues and projects.
  • GitHub — work with repositories, issues, and pull requests.
  • HubSpot — work with HubSpot contacts, deals, and tickets.

Connections are personal. Each workspace member connects their own accounts, so the agent never acts through a teammate’s inbox on your behalf.

Connections are optional. The agent covers the full property workflow without any of them.

What connections unlock

With connections in place, one ask can cross tools:

  • Post tomorrow’s check-ins to the team Slack channel.
  • Log every completed turnover in a Google Sheet your cleaner already reads.
  • Draft the review-request email in Gmail when a stay ends.
  • Open a Linear issue when a guest reports something broken.

Skills

A skill is a written playbook the agent follows: your message tone, your escalation rules, how you price shoulder season. Install skills from the public skills.sh registry, then switch each one on or off.

Every member starts with five hosting playbooks already installed. Keep the ones you use and remove the rest — and read every word of them in the Skills library:

Skills are personal too. Each member installs and configures their own, so your playbook never rewrites a teammate’s.

You can also write your own. “New skill” on the Skills page opens an editor for a custom SKILL.md — your house rules, in your words. Writing skills is owner-managed: teammates can only create, edit, or delete custom skills when an owner delegates the Agent skills permission through a custom role.

Automatic and manual skills

Each installed skill runs in one of two modes. An automatic skill is available on every message, and the agent decides when it applies. A manual skill stays out of the conversation until you name it for a specific ask.

Use manual mode for a skill you have not fully vetted. It only speaks when called.

Safety boundaries

Skills come from a public registry, so the agent treats their text as reference material, never as instructions with authority. A skill cannot authorize a tool call, request credentials or private data, send data anywhere, or speak as you.

The approval contract does not bend for either feature: every change to a live listing still stops and asks you first, whatever a skill suggests.