Playbooks
House rules your AI follows per situation — matched by intent, not scripted like a bot.
Playbooks let you steer how your AI agent handles specific situations — what to do, what to ask for, and when to open a ticket — without building a chatbot flow. You write a short rule in plain language and give a few example customer messages. When a real message looks like one of your examples, the agent follows that rule, in its own words.
A Playbook is guidance, not a script. The agent still understands the question, blends rules that overlap, and handles wording you didn't anticipate. There are no branches, buttons, or dead-ends to maintain.
Playbooks are available on the Growth plan and higher.
How it's different from a bot#
A bot needs you to map every path with if/else rules, and breaks the moment a customer phrases something unexpectedly. A Playbook just describes the outcome you want for a kind of request. The agent recognises the situation, applies your rule, and keeps reasoning normally for everything else.
Create a Playbook#
Go to Playbooks in the sidebar and select New playbook.
Name it
A short label for the situation, e.g. Order status or Damaged item return.
Write the instruction
Describe, in plain language, how the agent should handle this situation — what to explain, what to collect, what not to promise. This is the heart of the Playbook.
Add trigger examples
A few example customer messages, one per line. The agent matches similar wording, not exact text, so two or three natural phrasings are enough.
Set the ticket policy and access level
Choose when a ticket should be opened and what verification the situation needs (see below).
Ticket policy#
Each Playbook decides what happens with support tickets for its situation:
| Policy | What the agent does |
|---|---|
| Never | Resolves it in chat and never opens a ticket. |
| Offer | Suggests a ticket and opens it only if the customer agrees. |
| Always | Opens a ticket once it understands the problem. |
| Only if a condition is met | Opens a ticket only when the condition you describe is true. |
Use Never for situations you want handled entirely in chat (for example, explaining a return policy), and Always for things that always need a person (for example, a request to speak to a human).
Access level#
The access level tells the agent what the customer must prove before it shares sensitive details or takes action for that situation:
| Level | Requires | Good for |
|---|---|---|
| S0 | Nothing | General questions — catalog, policies, hours. |
| S1 | An identifier (order number, phone, email) | Read-only status, e.g. where an order is. |
| S2 | Login or a one-time code | Personal data and actions — address, payments, cancellations. |
Test before you rely on it#
Open the Test a message box at the top of the Playbooks page, type a sample customer message, and you'll see which Playbook would fire and how strong the match is. Use it to check your trigger examples catch the phrasings your customers actually use.
Import many at once#
To set up a lot of Playbooks quickly, select Download template, fill it in (one row per
Playbook; put trigger examples in one cell separated by |), and Import it. Any rows
with problems are reported so you can fix just those.
What stays in your control vs. ours#
You control the behaviour: the instruction, the examples, the ticket policy, the access level. Chatasist enforces the safety — a Playbook can never make the agent skip identity verification, invent facts, leak another customer's data, or claim it did something it didn't. Your rules steer the agent inside those guardrails.