Builder Guides

The builder guides are hands-on, task-oriented walkthroughs for building AI Employees (AIEs) on Ema. Where Core concepts explains what the building blocks are, this section shows you how to assemble them in the AI Employee builder — from your first workflow through advanced features like voice and custom code.

Every guide is verified against the product. You should be able to follow any one of them end to end and ship a working AI Employee.

Before you start

An AI Employee is the top-level unit you build. Its behavior is defined by a workflow — a directed acyclic graph (DAG) of typed nodes that you wire together in the AI Employee builder. The five node types are:

• Start — defines the workflow's input schema (the fields it accepts at run time) and, optionally, a schedule.
• Agent — a reasoning step. You pick one of the agent types and configure its Instructions, Knowledge base, Tools, and human-in-the-loop behavior.
• Transform — a deterministic, non-LLM step that reshapes data from an upstream node using a list of field-mapping rules.
• Publish — declares the output fields the workflow exposes (used by dashboards and CSV export).
• End — terminates a branch.

You connect nodes with edges. An edge can be unconditional, or it can carry a condition so the run only follows that branch when the condition holds. You move data between nodes with {{...}} references in each agent's input mapping.

Guides in this section

• Build your first workflow — add nodes, connect edges, map inputs with {{...}} references, and run a workflow end to end.
• Writing effective Instructions — how Instructions and the system prompt drive an agent, how to reference inputs, and how to write Instructions that produce reliable results.
• Setting up a knowledge base — upload or connect a source, ingest documents, and attach a knowledge base to an agent for retrieval-augmented answers.
• Designing human-in-the-loop forms — pause a run to collect structured input with an inline form, an external form, or a free-form conversation.
• Creating a Voice AI Employee — provision a phone number, choose a voice, configure conversation and call settings, and route calls to people.
• Creating a Dashboard AI Employee — build a table-driven AI Employee where every row is one workflow run, inputs are columns, and published outputs fill in as runs finish.
• Building custom code agents — write a TypeScript function that runs in a sandbox, calls the LLM, and reaches your integrations.
• Worked examples — three complete end-to-end builds (support triage, document extraction, knowledge-base helpdesk) that wire real agent types and workflow nodes together.

How a build typically goes

1. Create the AI Employee and open the AI Employee builder.
2. Define inputs on the Start node — the fields your workflow accepts.
3. Add agents from the agent library and configure each one's Instructions, Knowledge base, and Tools.
4. Connect the nodes with edges, adding conditions where the path should branch.
5. Map inputs on each agent so it receives the data it needs from the Start node or upstream agents.
6. Test the workflow with a dry run, then iterate.
7. Publish to make the version live for real runs.

You don't have to build by hand. Autopilot can create and edit workflows, agents, and knowledge bases for you from a plain-language description — these guides explain the same building blocks Autopilot manipulates, so you can review and refine whatever it produces.