Skip to main content
The Customer Agent is the customer facing AI agent built into Plain. It answers your customers from your knowledge sources, asks a question when it needs a detail, and hands the thread to your team when it can’t help. There is no flow builder and no decision tree to configure: you turn it on, give it your knowledge, and shape its behavior with custom instructions in your own words.

Customer Agent and Sidekick Agent

Plain has two AI agents. The Customer Agent responds autonomously to your customers without a human-in-the-loop; the Sidekick Agent works for your team and helps you automate your support process.

Display name

Customers never see the name Customer Agent. Wherever the agent acts, such as in replies and on the thread timeline, Plain shows its display name instead, with an AI agent badge on timeline entries. New setups default to your workspace name followed by “Support”, for example “Acme Support”. Change it in Agents → Customer Agent → Preferences. Workspaces that set up the agent before it was renamed keep their existing display name. The Customer Agent doesn’t introduce itself unless your custom instructions name it or ask it to. When it introduces or describes itself, it says it’s an AI.

How the Customer Agent handles a message

Each time a customer sends a message on a thread assigned to the Customer Agent, it reads the thread and its context and decides what to do:
  1. Checks it should act: it stops before answering if someone on your team has replied, your workspace is out of credits, or a customer tagged a user on Slack. See when it stops before answering
  2. Searches your knowledge: as many times as it needs, rewording a search that finds nothing useful. It skips the search for a greeting or a thanks
  3. Ends with one action: it replies, asks the customer one question, hands off, answers the part it can and hands off the rest, or skips a message that needs no reply, such as spam
  4. Checks the message before sending it: every sentence and code block must be backed by the thread, the customer’s attachments, its searches, or your instructions, and every link must come from a search result or your instructions. A message that fails goes back to the Customer Agent to rewrite, or it hands off instead. Customers never see a message that failed
After it replies or asks a question, the Customer Agent snoozes the thread as Waiting for customer and picks it up again when the customer writes back. A message gets up to 6 rounds of searching and writing, and 45s. If the Customer Agent can’t finish within either limit, it hands off to your team with a note saying so. Your custom instructions change most of this behavior, including when it hands off and how it writes. A short list of fixed rules, such as never inventing facts or links, applies whatever they say. Set up, test, and deploy the Customer Agent in five steps.

1. Enable setting

The Customer Agent is part of Plain AI, which is enabled by default on new workspaces. If you’ve previously turned it off, turn it back on in Settings → Plain AI.

2. Add knowledge

The Customer Agent automatically reads your customer facing Help Centers. Add anything outside Plain as a knowledge source. The Customer Agent only states facts from this knowledge, the thread, and your instructions, so without knowledge it hands off most threads. If nothing is indexed at all, it hands off every thread with a note asking your team to add knowledge. When the Customer Agent can’t answer a question from this knowledge, Plain records a knowledge gap. Once a second thread asks the same question, Plain raises a task for your team, so you know what to document next.

3. Route threads

The Customer Agent works on threads assigned to it, except threads from customers marked as spam, which it skips without replying. There are two ways to assign threads.

Assign via workflow

Go to Workflows, create a rule, and use the “Assign to user” action. Under Customer Agent, pick your agent’s display name with (Live) or (Test). The conditions on the rule decide which threads the Customer Agent picks up, so you can scope it by tier, label, channel, or business hours. Example: “When a new thread is created on the chat channel with tier Free, assign to the Customer Agent.” Use this Workflow Template to get started.

Manually

Assign the Customer Agent to any thread yourself from the assignee picker, exactly like assigning a user. Use it to test the Customer Agent on a specific thread.

4. Test

Run the Customer Agent in test mode to see how it behaves on your threads without giving it permission to send messages to your customers. Try it in the Playground to test specific scenarios and see each search and decision it makes. While you test, write custom instructions to fit the Customer Agent to your support process and tone.

5. Go live

When you’re happy with the Customer Agent, switch its workflow assignments from Test to Live. You can do this one workflow at a time, see test mode. The Customer Agent starts replying to customers on the threads your workflows assign to it.

Credits

Each thread the Customer Agent resolves uses 100 credits from your workspace’s credit allowance. A thread counts as resolved when all of these are true:
  • The Customer Agent replied on the thread at least once
  • The Customer Agent resolved the thread on its own, without a team member getting involved
  • The thread moved to Done and stayed in Done for 7 days
Internal notes your team leaves on the thread don’t count as getting involved. If the customer replies before the 7 days are up, the thread leaves Done. The Customer Agent can reply again, and the 7 days start over when the thread moves back to Done. When the Customer Agent snoozes a thread as Waiting for customer, it marks the thread Done after 7 days without a reply. See when the Customer Agent uses credits for a thread.

How do I know it’s working?

Head over to either of the following pages to see which threads the Customer Agent is working on: Handled by Customer Agent lists Todo threads the Customer Agent is working on or has answered. The Customer Agent snoozes threads as Waiting for customer after it replies, so turn on Show all statuses to include snoozed and done threads. The count next to Handled by Customer Agent only includes Todo threads. Handed off threads sit in your queue like any other. Moving a thread back to Todo counts as your team taking over; marking it Done or snoozing it doesn’t. To see what the Customer Agent could not answer, go to Knowledge → Knowledge gaps. Each open gap is a task for your team, and the Sidekick Agent can draft the missing article for you. See Knowledge gaps.

What can Customer Agent see?

Each time a customer message arrives on a thread that Customer Agent is assigned to, it assembles a fresh snapshot of the thread to use as context: the conversation, the customer’s name and company, the channel, attached files, labels, and any thread fields you share with it. See Context for a full breakdown of what is included.

Custom instructions

Custom instructions tell the Customer Agent how to handle support for your workspace, in your own words. They win over its default behavior, so you can change when it hands off, what it asks, and how it writes. For example, you can have the Customer Agent:
  1. Introduce itself by a name you choose
  2. Hand every billing question to your team
  3. Ask for the exact error message before troubleshooting, then pick the thread back up when the customer answers
  4. Search using your documentation’s terminology when customers use different words
See Custom Instructions for examples and guidance on writing them.