Custom AI agents
The Support Agent answers your customers. Custom agents handle the work around it. They can react to events or run on a schedule, read the same things the Support Agent reads, and update the tools you already use, with the same safety checks as the Support Agent.
An agent can start from something in the inbox, a schedule, or a webhook from another tool. Whatever started it, like the conversation, message or payload, is passed in as context.
It can use your knowledge base, memory, past conversations and any connector with an MCP server. Pick a preset (read-only, internal writes or publish) to set the limit, then turn tools on one by one.
Segments control access to articles, files, playbooks and tools, and that’s enforced on the server. An agent working on a customer’s thread only sees what that customer is allowed to see.
Gallery
Watches your releases channel and drafts the help-centre changes before anyone asks.
When a thread closes, it saves anything new it learned about the customer or the product.
Files the request with product, checks for duplicates and links the customer to the issue.
When the linked issue closes, lists everyone who asked and drafts the outreach.
Anything that publishes, spends, or emails a customer waits for a person unless you say otherwise.
Every step is logged with what it read and what it wrote, in the same timeline you see on a Support Agent reply.
Anything an agent learns about a customer is there for the Support Agent and your team next time.
Results show up as drafts, notes and threads, right where your team already works.
Custom agents talk to your team and your tools. Only the Support Agent talks to customers.
Any tool with an MCP server can be used by an agent, and any webhook can start one.
Lloyd Humphreys
Founder, Buzzer
Tell us about something your team does by hand every week, and we’ll build the agent together on the call.
Book 30 minutes with Lloyd