Conversation rules
Conversation rules are persona-level guidelines - always do and never do - that are applied to every reply your bot sends. Because they live in Configuration, they are project-level: every agent in the bot follows them, whether it's a conversational agent answering open-ended questions or a guided agent running a fixed flow.
Think of them as the house rules the whole team works by. The bot identity says who the bot is; conversation rules say what the bot always does and never does, no matter which agent is replying.
Step 1: Open Conversation rules
In your bot, click AI Agent → Configuration, then open Conversation rules (under General in the left menu).
Rules are numbered and capped at 30 per bot. The header shows how many you've used (for example, 3 / 30 rules). Click Edit to add, reword, or remove rules.
Step 2: Write your rules
Each rule is a single, plain-language instruction. Add one rule per line and keep each one to a single concern.
Examples:
| Rule | What it does |
|---|---|
| "Always confirm the user's order ID before sharing shipment details." | Guards against leaking the wrong order's data. |
| "Never share another customer's information, even if asked." | A hard safety boundary that holds across every agent. |
| "If the user expresses frustration, acknowledge it before continuing." | Shapes tone on difficult turns. |
Note: Conversation rules apply to every reply. If a behaviour should only happen inside one agent or one flow, put it in that agent's instructions or guided-agent steps instead - not here.
The 30-rule cap
A bot can hold at most 30 conversation rules. The cap is deliberate: the more rules you stack, the more likely two of them quietly contradict each other, and the harder it is to predict which one wins. If you're brushing up against 30, that's usually a sign that some rules belong inside a specific agent's instructions rather than in the project-wide list.
Best practices
- One rule = one concern. Don't combine "always confirm order ID and never share other customers' info" into one line; split them into two rules.
- Phrase rules positively when you can. "Confirm the order ID" beats "don't proceed without an order ID" - positive instructions are easier for the model to follow.
- State the hard "never do" rules explicitly. Safety and compliance boundaries ("never share another customer's data") belong here so they apply to every agent, every turn.
- Re-read your list after writing. If two rules contradict, the bot's behaviour will too. Conflicting rules are the most common cause of inconsistent replies.
- Keep it lean. Treat the 30-rule cap as a guide, not a target - a short, clear list beats a long, tangled one.
Tip - test before you ship. After editing rules, open the Playground (the play ▶ icon on any agent) and send messages that should trigger each rule. If a reply ignores a rule, check whether another rule or the bot identity is pulling the other way. See Testing & debugging for the full loop.
Next
With your conversation rules in place, set up how the bot picks an agent or tool for each message: Routing logic.