Botmaker 3.0 is an agentic version of the platform. It’s an AI that manages and resolves complete business processes, able to make decisions, take actions, and execute concrete tasks in external systems. Unlike previous versions, it isn’t programmed with rigid scripts or decision trees: you describe the goal in natural language and the agent navigates the necessary steps on its own to resolve it. 

You can: 

• Create autonomous AI agents that manage complete business processes. 

• Coordinate several agents with an orchestrator that decides which one responds based on the context. 

• Add tools to your agents such as knowledge bases, MCPs, code actions, and integrations. • Design automations that run on their own by time or by event. 

• Add human validation (Human in the Loop) at the critical points of the process. • Work collaboratively across multiple users on the same agents and orchestrators. 

Key concepts 

Before moving on to the answers, it’s useful to be clear on these definitions, which you’ll come across throughout the article and inside the platform. 

Agent: an AI-powered bot that understands a specific business goal and decides what to do step by step to resolve it. One agent equals one goal. If you need to resolve several goals, you use several agents coordinated with each other.

Orchestrator: the director of the agent team in a channel. It receives the customer’s message, consults the relevant agents, and chooses the most suitable one to respond. 

Channel: the messaging medium the customer arrives through. For example: WhatsApp, Webchat, Instagram, Telegram, Slack, or Google Business Messages. 

Workflow: each logical step the agent goes through to resolve its goal. For example: “collecting information,” “confirming appointment,” or “taking payment.” 

Resource: a tool the agent can use to resolve a workflow. There are several types: logics, knowledge bases, integrations (MCP), code actions, and external apps. 

Logic: an ordered sequence of steps. It’s a type of resource, useful when the sequential order matters. 

Automation: a trigger that runs on its own by time or by event, without the customer having to message the bot. It lives outside the agent. 

Human in the Loop (HITL): a working mode in which the agent knows to ask a human operator to validate, confirm, or resolve an exception. 

Multimodal: the agent understands and responds with text, audio, or image within the same conversation. 

MCP (Model Context Protocol): a standardized connector to an external system such as a CRM, a calendar, or a database. 

Ticket: a record the agent generates for each workflow started within a conversation. It lets you see the process status and follow up on it. 

Costs and billing 

How is the cost of using Botmaker 3.0 calculated? 

Billing is based on token consumption. You pay for what your agent consumes: the more it resolves, the more is billed. If it isn’t used, you don’t pay. The configuration tool has no extra costs: creating agents, orchestrators, automations, or entities doesn’t generate a charge by itself. The cost appears when the agent runs and starts handling requests. 

Is billing by tokens, conversations, or messages? 

By tokens, and the cost depends on the model. You’ll find the consumption details on the Account screen of the platform. The MAU or DAU session-based plans continue to work the same way. Only generative consumption is added. 

Do orchestrators generate additional costs?

The orchestrator doesn’t have an extra cost as a piece, nor is it billed separately. It’s included in the plans. Its execution does consume tokens, just like any agent. 

Will I be able to see the cost per agent? 

Yes. Each feature reports its own tokens. You’ll be able to see how much each agent, each orchestrator, and each feature consumes, and attribute the spend by business area. The generative technologies screen shows the token details used by each agent, orchestrator, and by Human in the Loop. 

AI models 

Which AI model does Botmaker 3.0 use? 

In this first release, the platform runs on the default model Open AI 5.4. The model details and its consumption cost are available on the Account screen. 

Will I be able to choose the model (LLM) I use? 

We’re working on letting you select the model at the agent level, so that a single orchestrator can coordinate agents using different models depending on the case (by token consumption or by reasoning capability). 

Agents, orchestrator, and workflows 

What is an agent and what does it do? 

An agent resolves a specific business process. Think of it as a virtual employee with a clearly defined job. For example, at a dental practice you could have one agent to book appointments, another to answer frequently asked questions, another to handle dental emergencies, and another to send satisfaction surveys. Each agent has a goal, its own workflow, and a set of tools to resolve it. 

What is the orchestrator? 

It’s the new piece in version 3.0. It works as a layer that receives the customer’s message, looks at which agents are available on that channel, and picks the most suitable one to respond. It asks the relevant agents (not all of them, to save tokens), evaluates which response fits the context best, synthesizes a single response, and sends it. 

How do channels, orchestrators, and agents relate to each other? The relationships are simple: 

• 1 channel = 1 orchestrator. Each channel has a single orchestrator. There can’t be two directors in the same conversation. 

• 1 orchestrator = N agents. An orchestrator manages a team of several specialized agents.

• 1 agent = N orchestrators. A well-configured agent can be reused across several channels and orchestrators without needing to duplicate it. 

What is a workflow? 

It’s the smallest unit of the agent’s behavior. Each agent goes through a sequence of workflows based on the conversation. Workflows can branch: the agent chooses the next one based on context, not on a fixed tree. For example, the “Book appointment” agent could have the workflows: greeting + intent, request basic information, show availability, confirm appointment, and wrap-up. 

Can AI agents be combined with traditional scripted flows? 

Yes. The paradigm has been inverted: you now work with orchestrated AI agents and, when you need tighter control or a rule-based flow, you create it as a logic inside the agent. The logic works as an ordered sequence of steps the agent follows to resolve a specific task. 

How do you give the bot a personality? Where is the tone configured? 

Tone, style, restrictions, and the use of buttons are configured at the orchestrator level, not on each agent. Each agent inherits that personality when responding, which prevents you from ending up with contradictory instructions across agents. On the orchestrator you can define: 

• Tone: formal, informal, friendly, or technical. 

• Length of responses: short or detailed. 

• Personality: brand voice (friendly, neutral, playful). 

• Restrictions: what it can’t say or promise. 

• Whether or not buttons are allowed in the response. 

How are guardrails configured and how do you keep the agent from going off track? 

Restrictions are defined at the orchestrator level. In addition, internally Botmaker works as a set of agents that check each other, so it’s rare for an agent to stray outside the territory it was created for. Agents only respond using the tools they have available: if no tool is added for a given topic, they don’t engage in conversation about that topic. 

Agent resources and tools

What tools (resources) can an agent use? 

There are five types of resources available: 

• Logics: an ordered sequence of steps. For when the sequential order matters. For example, the step-by-step of booking an appointment. 

• Knowledge bases: PDFs, websites, FAQs. The agent reads and responds with that information. For example, fees and accepted medical coverages. 

• Integrations (MCP): turnkey connectors to external systems such as Google Sheets, Google Calendar, CRMs, and databases. 

• Code actions: code integrations the agent can run. Useful for custom integrations.

• External apps: other tools integrated into the platform. 

How does the agent decide when to use each resource? 

The decision is made by the model based on the “When it will be used by the agent” field. If the resource’s description is clear and specific, the agent knows when to invoke it. If it’s ambiguous, it may get confused or call it when it shouldn’t. 

Note: an ambiguous description like “For appointment topics” usually causes errors. A clear description like “When the patient requests a new appointment and has already provided their information” works much better. 

How do logics work? 

Logics are built with three types of blocks: 

• Instruction: a generic directive to the agent (“Ask the patient for their ID”). It’s the most used. 

• Conditional: branches based on conditions (“If the coverage is Swiss Medical, show premium time slots”). The most important in practice. 

• Loop: repeat N times. Useful for processing lists; rarely used in conversation. 

A good practice when building logics is to use several small boxes with one instruction each, rather than one giant box with multiple chained instructions. The model handles separated instructions better. 

Can I upload documentation or files as a knowledge base? 

Yes. Knowledge bases accept PDFs, websites, and FAQ content. The agent reads that content and responds based on it.

Can code actions still be used? 

Yes. Code actions remain available as one more resource the agent can fall back on whenever it needs to. The code actions you already had configured in previous versions can still be used in Botmaker 3.0. 

What are automations? 

They are processes that run on their own, without the customer messaging the bot. There are two types: 

• By time (schedule): they run at defined moments (every hour, every Monday at 9 AM, on the 1st of every month). 

• By state (event): they react to something that happened in a conversation, such as a workflow change or a completion. 

Some examples: send an appointment reminder 24 hours in advance, push a data point to the CRM when an appointment is confirmed, or send a satisfaction survey one day after the appointment. 

Available channels and integrations 

Which channels does Botmaker 3.0 support? 

Botmaker 3.0 supports the main messaging channels integrated into the platform. These include WhatsApp, Webchat, Telegram, Instagram, Facebook, and Google Business Messages. 

Is it useful for automating internal processes, not just conversations with customers? 

Yes. Botmaker 3.0 stands out as a conversational process platform, but it can also be used for non-conversational processes. You can design agents and automations that connect your ERP, your CRM, spreadsheets, and other tools. For example, invoice uploads, internal notifications, or administrative routines. 

Multimodality 

Do agents understand images, audio, or video? 

Yes. The models are multimodal: they can understand text, images, audio, and video. Not only what an image says, but also its essence. For example, a customer can send a photo of a promotion over WhatsApp and the agent can understand the content to take the order. 

Human in the Loop, tickets, and traceability 

What is Human in the Loop (HITL) and when is it used?

Human in the Loop adds a way of working in which the AI agent sits in the middle of the conversation at the service of the human operator. The human asks the agent to confirm, execute, or develop a response, and the agent communicates with the end user. At the same time, the agent presents the customer’s communications to the human with analysis, suggestions, and shortcut buttons so all the human has to do is validate. 

Any workflow state can be marked as “requires human confirmation.” When a case reaches that point, the AI stops and hands control off to an operator, who sees the case in the dedicated Human in the Loop inbox. According to internal testing, this mode can reduce manual work by up to 82%. 

How can I see the conversations each agent is handling? 

Every time an agent starts a workflow, a ticket associated with that conversation is generated. Tickets appear in a new section below Chats and can also be accessed directly from the agent. There you’ll see the workflow status, the ticket details, internal comments, the associated conversations, and the activity. You can also route the ticket to a support team or add followers so they stay in the loop. 

Is there a Kanban-style view for tickets? 

Yes. Tickets can be viewed as a list or as a Kanban-style board where the columns are the workflow states and the tickets are arranged according to which stage they’re in. You can move a ticket from one state to another manually, as long as the conditions defined in the workflow are respected. 

Can I see previous versions of an agent and revert changes? 

Yes. Agents have a versioning system: each adjustment you publish is recorded in a version you can view and revert if a recent change didn’t produce the expected result. 

Can multiple users work collaboratively? 

Yes. The platform lets several people work simultaneously on the same agents. Each user is identified by a color, you can see who’s online and where they’re working, and you can split tasks or work in parallel on different agents and orchestrators. 

Remember to visit our Help Center for further information.