> ## Documentation Index > Fetch the complete documentation index at: https://help.ciarem.ai/llms.txt > Use this file to discover all available pages before exploring further. # Your AI agent team: types and per-agent settings > The six predefined agent types, what you configure on every agent, and the extra options for qualifier, sales, and FAQ agents. **Control center > Agent team** organizes specialized agents so each one handles a specific type of need. **Create agent** starts from one of six predefined types: * **Qualifier agent**: asks key questions to understand the contact and gather information before passing them along. * **Sales agent**: shares product and service information: options, prices, availability, features. * **FAQ agent**: answers frequent questions from your business information: hours, prices, processes, policies. * **Appointment agent**: coordinates appointments and bookings against your [calendar availability](/ai-agent/booking-and-calendars). * **Support agent**: handles questions and problems from existing customers. * **Billing agent**: answers about payments, charges, invoices, receipts, and payment methods. You do not have to think about routing. The platform reads each agent's purpose and decides which one handles which conversation. Several agents can take part in the same conversation as the customer's need changes. ## How a message finds its agent For each message, Ciarem reads every active agent's type and its **When should this agent step into the conversation?** text, together with the conversation so far, and picks one agent with a confidence score. That text is the whole job description the router sees, so write it for the router. The agent that handled the previous message keeps short, ambiguous, or follow-up messages ("and the annual one?", "send them one by one"). The conversation only moves to another agent on a clearly new topic that belongs to it. Only agents with catalog access can send product photos or files. Only a sales agent with [checkout](/ai-agent/taking-orders) on can register an order. When a message asks for information that one agent's knowledge base covers, the router strongly prefers that agent. An agent cannot use what is not switched on for it. An agent with no knowledge base or catalog switched on is told so, and answers "I don't have that detail" to every specific question, then follows its handoff rules. The most common cause of an agent that "doesn't know anything" is a switch left off in its configuration, not a missing document. ## What you configure on every agent * **Name and icon**: how the agent appears in the inbox and reports. * **When should this agent step in**: a plain-language description of the moment it should act, the way you would brief a new hire: "Interest in plans, prices, or signing up for our services." Keep it specific; this is what routing reads. * **Knowledge**: toggle which [knowledge bases](/ai-agent/knowledge-bases) this agent can consult to answer. * **Follow-ups**: "Follow up when the customer goes quiet". Pick nudges at 15 minutes, 1 hour, and/or 2 hours. See [Proactive follow-ups](/ai-agent/follow-ups). * **On/off switch**: each agent can be activated or deactivated individually from its card. ## Type-specific configuration * **Qualifier agents** get **Questions the agent will ask**. Each question carries the question text, an optional **"Save the answer"** property (a built-in contact field such as name, email, phone, or city, or any custom property you have defined), and **"When to ask"**: *Only if there's no information* (skip it once the property is filled) or *Always ask*. Drag questions to set the order. Answers land on the contact automatically, so qualification fills your CRM as the conversation happens. * **Sales agents** get a **link to continue the purchase**, sent when the agent detects buying, booking, or quoting intent, plus catalog access (below). * **Qualifier, Sales, and FAQ agents** can be granted **catalogs**: toggle each [catalog](/using-ciarem/catalogs) the agent may consult, and optionally allow it to **Add content** or **Edit content**. For example, a sales agent can register a new order record while a FAQ agent stays read-only. Agents without access to a catalog cannot see it. Configuring a qualifier agent: questions with a property to save the answer to and when to ask, plus the catalogs it can consult ## Common mistakes Sales: "questions about our services". FAQ: "questions about the business". The router has nothing to separate them, so the conversation flips between two agents with different knowledge and different styles. | ❌ Overlapping | ✅ Distinct moments | | --------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Sales: "Questions about our services." FAQ: "Questions about the business." | Sales: "Interest in buying: plans, prices, what is included, how to sign up." FAQ: "Hours, location, policies, how a service works, questions from existing customers." | "Be friendly, answer quickly and always try to close the sale" is a briefing, not a trigger. No customer message matches it. It is the same confusion as writing instructions into the [knowledge base](/ai-agent/knowledge-bases): behavior lives in [Basic configuration](/ai-agent/basic-configuration) and in [corrections](/ai-agent/playground). This field is the *when*. | ❌ | ✅ | | ----------------------------------------- | ----------------------------------------------------------------------------------------------------- | | "Be friendly and always close the sale." | "The customer asks about plans, prices, or how to sign up." | | "Help customers with anything they need." | "Problems with an active service: something not working, a complaint, a change to an existing order." | A single all-purpose agent cannot ask qualifier questions *and* sell from a catalog *and* book appointments. The type unlocks the tools. Six agents with one-line descriptions and no knowledge attached do not help either: the router has more options and the customer gets the same answers. Start with two or three that each have a real job, a knowledge base, and the catalogs they need. Add another when the inbox shows a distinct need. The agent was created and the knowledge base exists, but the **Knowledge** switch for that agent was never turned on, or the catalog was never granted. Every specific question gets "I don't have that detail". | ❌ | ✅ | | -------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Sales agent with no catalog and no knowledge base. | Sales agent with the *Services* catalog granted and the *Policies* knowledge base switched on. The Control center's **Knowledge base** card warns when agents come up short. | Ten questions before the customer gets any help reads like a form, and people leave. Ask only what changes what you do next, save each answer to a contact property, and use **Only if there's no information** so returning contacts are not asked again. | ❌ | ✅ | | ----------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Name, email, phone, company, role, city, budget, timeline, how did you hear about us, best time to call, all set to **Always ask**. | Name (saved to name), city (saved to city), what they need (saved to a custom property), each set to **Only if there's no information**. A person asks the rest. | A Support agent asked to sell has no catalog to sell from: catalogs can only be granted to qualifier, sales, and FAQ agents. A FAQ agent asked to book has no calendar: booking comes with the Appointment agent and a [connected calendar](/ai-agent/booking-and-calendars). Pick the type by what the agent must *do*. The type decides which tools it can have. ## When something looks wrong | The agent… | Usually because | Fix | | ------------------------------------- | ------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------- | | Says it doesn't have that information | No knowledge base or catalog switched on for that agent, or the fact is not in the knowledge base | Turn the switch on in the agent's configuration, or add the fact | | Keeps switching between two agents | Their "when should this agent step in" texts overlap | Give each a distinct moment | | Never uses the agent you created | Its text doesn't match how customers write, or the agent is paused | Rewrite it in the customer's words; check the switch on its card | | Asks the same question every time | The question is set to **Always ask**, or its answer is not saved to a property | Save it to a property with **Only if there's no information** | | Says it can't send images or files | The agent has no catalog access | Grant the catalog to a sales, qualifier, or FAQ agent | The other tab on this screen, **Transfer rules**, decides when an agent should stop and hand the conversation to a person. See [Human handoff and transfer rules](/ai-agent/human-handoff-and-escalation).