Agents

An agent is the AI assistant that talks to your customers. You configure how it sounds, what it knows, which actions it can take, and when it hands off to your team - then publish it when it is ready to go live.

What an agent is in Lyro

Each agent has its own personality, knowledge, and skills. You can create several agents for different jobs (for example one for order questions and one for technical support), and Lyro routes each conversation to the right one based on the agent's description.

You manage agents from the Agents page, where each agent shows its current state. Use New Agent to create one, or open an existing agent to edit it.

Just getting started? See Create your first agent for a guided walkthrough.

General settings

The General tab holds the core configuration of the agent.

Visitor context

Under Visitor context you choose which contact properties the agent can see in its system prompt. Name and email (plus other system properties) are always sent. You can add custom properties on top so the agent personalizes its answers. Manage the available properties from Settings.

Knowledge

The Knowledge tab controls what the agent can look up. Toggle on the knowledge bases this agent is allowed to search. If you leave every base off, the agent has no knowledge base and will not search at all.

You also pick a knowledge mode:

• Full - use the knowledge base when relevant, and fall back to general knowledge otherwise.
• Strict - answer only from the knowledge base, and refuse when the answer is not there.

Use Strict mode when accuracy matters more than coverage, for example for policy or compliance answers.

Skills

Skills bundle instructions and tools the agent can call - for example looking up an order or checking delivery status. Each skill belongs to the agent, so you create, edit, and remove them right on the Skills tab.

• Use New skill to add one, then give it a name, a trigger, instructions, and any tools.
• A skill set to Always on is available in every conversation; otherwise it activates when its trigger matches.
• The tool count on each skill row shows how many actions it can perform.

For reusable, account-level connections and actions, see Tools and integrations.

Escalation

The Escalation tab decides where the agent hands off when it cannot resolve a conversation - to a human, a ticket queue, or another channel. Toggle on the escalation procedures this agent should use. Each procedure shows the condition that triggers it.

Escalation procedures are shared across your workspace, so you configure them once and link them to as many agents as you like. To create or edit procedures, open Escalation. For a full setup, see Set up escalation to your team.

Versioning: draft and published

Every change you make to an agent is saved as a draft first. Your live agent keeps serving the published version until you publish your changes, so you can edit safely without affecting customers.

• Unpublished-changes indicator - when a draft differs from the live version, the header shows an amber Draft badge, and amber dots appear on the tabs and fields that changed. A matching live agent shows a green Published badge.
• Save draft stores your edits without going live.
• Publish opens a change diff so you can review exactly what changed, add an optional note about it, and confirm. Publishing creates a new numbered version (v1, v2, and so on).
• Discard draft throws away all unpublished changes and returns the editor to the live version. The live version is never touched.

Version history

Open the clock icon in the header to see every published version and the open draft. Selecting any entry shows its diff compared with the current live version, including the note you added at publish time.

A brand-new agent starts as a draft and only goes live on its first Publish.

Activating and deactivating

The Active / Inactive switch in the header is the operational toggle - it controls whether the agent is serving customers, independently of the draft and publish state. An agent cannot be activated until it has been published at least once. Activating happens automatically on that first publish.

To remove an agent entirely, use the menu on its row in the Agents list and choose Delete. A workspace must always keep at least one active agent.

Roles

A role defines the boundaries of an agent - what it is permitted to do and what it refuses. Lyro ships with built-in roles, and you can create your own. Assign a role on the General tab and manage the full list from Settings. The system prompt then layers your specific instructions on top of the role.