Build a Custom AI Employee

IntermediateLabCustom EmployeesCustom toolsAI Workforce

Estimated time · about 18 minutes|Required · Partner Center access on a Professional plan or higher, A simple API to test against, or dummyjson.com's free public order-lookup endpoint, used as the example in this step

Outcomes

✓Start a custom AI Employee from an outcome statement, not a feature list

✓Write a role and connect the knowledge and capabilities that outcome needs

✓Shape what it will and won't do through its role and capabilities, not a dial

✓Add one custom capability with a simple tool

✓Test against a pass/fail script before calling it done

Start from the outcome

You are here: Partner Center, on a Professional plan or higher. A Custom AI Employee starts from a blank slate. You choose the outcome, then give it the knowledge to answer from and one real action it can take.

Every employee so far came pre-built. This one starts blank, so the first decision is the outcome, not the employee: "qualify inbound leads before booking" is a real brief, "build a sales bot" is not. Building one costs nothing extra, and there is no limit on how many you create, so the only real constraint is a clear outcome. If the outcome resembles a data analyst, an inside sales representative, or a support role, the custom AI employee guide has a worked example close to it, worth stealing from rather than starting from nothing.

Try it now

Write your outcome in one sentence before you open the configuration panel: what this employee does, and for whom.

Create the employee

Labcreate the employee and its profile

Go to AI → Workforce, then Create.

1. 1.Give it a name and avatar that fit the role.
2. 2.Note what it can do on its own. A custom employee runs at the same autonomy the platform sets for you: it manages full conversations, captures leads, and books, and every conversation stays reviewable. There is no level to choose here, so what shapes its behaviour is the role, the knowledge and the capabilities you give it next.
3. 3.In Purpose, write the behaviour prompt: who it is, its tone, and the one outcome it exists for.

The employee now exists in your AI Workforce roster, with a profile you can open and edit.

Give it knowledge and capabilities

Labconnect knowledge and turn on capabilities

Stay on the employee's Configure panel.

1. 1.In Knowledge sources, use Add knowledge then + Add new knowledge to add the website pages or documents the outcome needs.
2. 2.In Capabilities, turn on the built-in capabilities that match the outcome, such as lead capture or booking.

The employee can now answer from real knowledge and take the actions its outcome needs, using only built-in pieces.

Add one custom capability

Most outcomes need one more thing a built-in capability does not cover: a real action in another system, a price check, an order status, an inventory count. Knowledge only answers from what has been uploaded to it, so an outcome that depends on live information hits a ceiling fast without a live connection. That connection becomes a tool, added inside a custom capability.

This step uses a real, working endpoint so you are testing against an actual response, not a hypothetical one: dummyjson.com, a free public test API that needs no authentication. Its /carts endpoint stands in for an order-lookup system. Give it an order number and it returns that order's total, its discounted total, how many products and total items it contains, and the products themselves (name, quantity, and line total for each). It is not a real Vendasta or client order system: it stands in so the tool you configure here behaves exactly like one pointed at a client's real order API.

Labadd a capability with a tool

Go to your employee's Configure page → Capabilities → + Add a capability.

1. 1.Select + New capability at the bottom of the picker, then name it CheckOrderStatus and describe what it does in detail: "Given an order number, looks up that order's status: its total, discounted total, item count, and the products in it, so the employee can read back a real summary instead of guessing." A specific description here is what lets the AI decide correctly when to call this tool versus a different one.
2. 2.Select + Add a tool, then + New tool. Method: GET. URL: https://dummyjson.com/carts. Add one parameter named order_number, set as a path parameter — the platform appends it to the end of the URL, so asking about order 7 calls https://dummyjson.com/carts/7.
3. 3.Save.

The tool now appears on the capability, ready for the AI to call once its instructions say when.

Path parameters, query parameters, headers, and every field in detail: Creating Custom Capabilities walks through the rest.

Write the capability's instructions

Brief the capability the way you would brief a new hire's one task: exactly when to use it, what it needs before it can act, and how to talk about what comes back. "Only call this when the customer asks about an existing order, and get the order number first" holds up in every conversation; "help with orders when it makes sense" does not.

Test before you call it done

Labrun the pass/fail script

Go to the employee's Chat, or its assigned channel.

1. 1.Ask a question that should trigger the capability, and confirm it calls the tool and answers correctly.
2. 2.Ask a related question that should not trigger it, and confirm it does not.
3. 3.Ask the triggering question again without the information the tool needs, and confirm the employee asks for it instead of guessing.

Three questions, three different situations, and the employee handled all three the way its outcome intended.

Chain more tools when a job needs it

One tool answering one outcome is a complete employee. Chaining several tools together, or connecting through middleware, is a deeper build for a job that genuinely needs it; the Tools & Integrations Overview is where to go next when that day comes.

What you now have

• A custom AI Employee built around one outcome you named, not a feature list
• A role, knowledge, and capabilities, all set from a blank start
• One custom capability with a working tool, briefed like a new hire's task
• A pass/fail test result you can trust before anyone else talks to it

Knowledge Check

Three quick questions on starting from an outcome, sorting knowledge from tools, and testing a new capability.

Congratulations, you completed this step!

Keep learning›