Integrations

Integrations connect your agent to outside systems so it can do real work mid-conversation — look up a customer, check live availability, book an appointment, create a CRM deal. Each integration adds a set of tools the agent can call.

How they work

Add an integration from the Integrations tab, give it credentials, and the agent gains that integration's tools. From then on the agent decides — based on the conversation — when to call them, and weaves the results into its reply. Toggle individual tools on or off per agent.

Credentials are stored once at the workspace level. Connect a system a single time and link it to as many agents as you like; rotating the credential updates every agent that uses it. Each agent link can also carry its own config overrides, per-tool settings, and link-level secrets — so two agents sharing one integration can, say, point at different locations or use different defaults.

Every integration has a test connectionaction that verifies the credentials and endpoint before an agent ever calls it. And Seny watches your site during crawls for systems it recognizes — when it spots one, it surfaces a recommendation on the agent that you can accept (pre-filled) or dismiss.

What you can connect

  • CRM — e.g. HubSpot: look up contacts, create deals, log notes live during a conversation.
  • Dealer DMS — e.g. PBS and XTime: look up customers, pull service menus, check appointment availability, and book service appointments against a live rooftop.
  • SMS / messaging — let the agent send a text mid-conversation.
  • Custom HTTP tools — point the agent at any endpoint of your own. Define the request shape on the Integrations tab and the agent gets a tool for it.

Write safety

Integrations that write into a live system (booking an appointment, submitting a lead form) support a mock-writestoggle — and on those adapters mock is the safe default on a fresh install. With it on, the agent goes through the whole flow and returns a synthetic confirmation without touching the live system — useful while you're tuning the agent's persona so a real backend isn't flooded with test bookings. Turn it off to go live.

Tool naming

Each integration's tools are prefixed with the integration's name, so two integrations on the same agent never collide — name them distinctly and the tools follow. The agent's system prompt is extended with a short playbook telling it when and how to use each tool.

MCP

Beyond these integrations, every agent is also reachable as an MCP server — so external MCP clients can talk to your agent directly. See the MCP guide for that.