The persona prompt
Write the instructions that define your agent's personality, tone, and rules, and reference tools directly from the prompt to lock them on.
The persona is your agent's core instructions — its personality, tone, the rules it follows, and what it should and shouldn't do. You write it in the Prompt tab on the agent detail page.
Think of the persona as the agent's character and standing orders. For precise, repeatable workflows like refunds, pair the persona with a procedure instead of cramming every step into the prompt.
Writing the prompt
The Prompt tab is a rich editor. You can use paragraphs, lists, and formatting to structure your instructions, and expand it into a full-screen dialog for longer prompts. A character count is shown as you write.
Your changes autosave as you type — there's no save button. The prompt doesn't go live until you publish a version.
A good persona usually covers:
- Role and scope — what the agent handles, and what it should hand off.
- Tone — how it should sound (friendly, concise, formal).
- Rules — hard limits, like "never promise a refund without checking eligibility."
- Escalation — when to bring in a human.
Mentioning tools and knowledge
Type @ in the prompt to reference a specific tool, toolset, field, or record directly inside your instructions. For example, you might write: "When a customer asks about delivery, use @find order to look up their order."
Mentioning a capability inline makes your instructions concrete and keeps the agent's behavior tied to the tools you actually want it to use.
Auto-locking
When you mention a tool or toolset in the prompt, Auxx automatically enables and locks it on. This prevents you from accidentally turning off a tool your instructions depend on. Locked tools show a lock badge in the Tools tab.
To remove a locked tool
Delete the mention from the prompt first. Once nothing references the tool, it unlocks and you can disable it normally.
Chat agents
For chat agents that answer customers in the chat widget, the persona is where you set tone, escalation rules, and guardrails. Their tools are configured separately and filtered to a customer-safe set.