Choose Agent Type

Before you start building, it helps to know what kind of agent fits your goal. Zia Agents supports a few distinct configurations, each designed for a different kind of work. Picking the right one upfront makes setup faster and results better.

That said, this is not a one time choice, and you can always change it later.

What defines an agent's type

When you create an agent in Zia Agents, the type is determined by what you give the agent to work with. Specifically, two things: a knowledge base and tools.

A knowledge base is a collection of your own documents (product docs, FAQs, policies, internal guides) that the agent can search through before responding. This is what keeps its answers grounded in your business context rather than relying on general knowledge.

Tools are API connections that let the agent take action in your systems. Fetching CRM records, updating deal stages, routing tickets, sending emails. Without tools, an agent can talk but it can't do anything.

Depending on what combination of these two you provide, your agent falls into one of four types.

Type	Knowledge Base	Tools	Use when
Agent without tools or knowledge base	No	No	You need a controlled AI assistant for general tasks like drafting, brainstorming, or Q&A
Agent with knowledge base only	Yes	No	You want responses grounded in your own documents and content
Agent with tools only	No	Yes	You want the agent to perform actions in your systems via API calls
Agent with tools and knowledge base	Yes	Yes	You need both business context and the ability to act on it

The four agent types

Agent without tools or knowledge base

This is the simplest configuration. The agent relies entirely on the language model's built-in knowledge to respond. It can answer general questions, help draft content, summarize topics, or generate ideas, but it has no awareness of your specific business data and no ability to interact with your systems.

You might wonder why you'd create an agent for this instead of just using an LLM directly. The difference is that an agent gives you a controlled environment around the model.

You define the agent's role,
Set its behavioral instructions,
And configure guardrails that limit what it can and can't say.

A raw LLM has no memory of your rules from one session to the next. An agent built on Zia Agents carries those boundaries consistently, and you can refine them over time. You also get observability into how it's performing, which you don't get from a standalone LLM prompt.

This works well for quick internal utilities.

A writing assistant that helps your team draft emails within a consistent tone
A brainstorming tool that generates campaign ideas while staying within your brand guidelines
A general Q&A agent for factual lookups where you want to control how the responses are framed.

Note: Zia Agents also supports Zoho's in-house LLM. For organizations already within the Zoho ecosystem who prefer not to route queries through external models, this means your data stays within Zoho's infrastructure. The agent works the same way, but with the added assurance that no third-party LLM provider processes your inputs or responses.

Agent with knowledge base only

Here, the agent can search through documents you've provided before generating a response. This means its answers are grounded in your actual content rather than the model's general training data.

This is a good fit when the agent's primary job is to answer questions accurately based on information that's specific to your business.

Think of a support agent embedded in your help center that resolves common queries using your documentation.
A product FAQ assistant that customers can chat with, or an internal policy bot that helps employees find the right information without filing a ticket.

The agent understands your world but doesn't take any actions in it.

Agent with tools only

This configuration gives the agent the ability to perform actions through API calls but without any proprietary documents to reference. It uses the language model's general reasoning to decide when and how to use the tools you've connected.

This is useful when the agent's job is primarily operational.

An agent that watches for new leads in CRM, scores them based on your criteria, assigns them to the right rep, and sends an intro email automatically.
An agent that monitors a support queue, routes tickets based on urgency, and updates statuses.

The work is about executing tasks, not answering questions from your knowledge base.

Agent with tools and knowledge base

The full configuration. The agent has both your business context from uploaded documents and the ability to act through connected tools. It can understand your processes, reference your documentation, and follow through with actions in your systems.

This is what you'd reach for when building agents that need to both know your business and operate within it.

A sales assistant that understands your product catalog and can update deal records.
A support agent that references your troubleshooting guides and escalates tickets when it can't resolve something.
An operations coordinator that reads your internal playbooks and triggers the right workflows based on what's happening.

Most teams that are building production-ready agents end up here.

Multi-agent workflows

There's a fifth option that sits outside the four types above. Instead of building one agent that does everything, you can connect multiple specialized agents into a workflow where each one handles a specific part of a larger process.

Each agent in the chain completes its piece and passes context to the next one.

A hiring workflow, for example, might have one agent screening applications, another scheduling interviews, and a third generating offer letters. They coordinate end-to-end without manual handoffs between steps.

This isn't something you need to start with. Most teams begin with a single agent and evolve into multi-agent setups as their processes become more complex. But it's worth knowing it's available when a single agent starts feeling stretched.

How to decide which one you need

If you're still unsure, think about what your agent will actually be doing day to day.

If it needs to talk to users and answer questions using your business content, start with a knowledge base agent.
If it needs to perform actions in your systems without much back-and-forth conversation, go with tools.
If it needs to do both, set it up with the full configuration.
And if the task is simple enough that general AI knowledge is sufficient, the basic agent without KB or tools will do the job.

Most teams start with a conversational agent backed by a knowledge base because it's the most intuitive to set up and test. From there, you can layer in tools or connect it to a larger workflow as your needs evolve.

You can always refine later

Choosing an agent type is a starting point, not a commitment. You can adjust your agent's configuration, swap out its tools, update its knowledge base, or even restructure it as a different type down the line. The platform is designed to let you start simple and grow at your own pace.

The important thing is to get started. You'll learn more from building your first agent than from planning the perfect one.