# Your first agent. Step by step.

> Learn HeyRik from first principles. Plan, create, configure, prompt, voice, ground, test, launch, and improve an AI voice agent through a complete beginner path.

Source: https://www.heyrik.com/docs/academy

Eight practical lessons to take you from a clear brief to a voice agent you can put to work. Learn at your pace and pick up where you left off.

8 focused lessons No prior voice-AI experience needed

Your Academy progress

## Loading your saved progress

Your course progress will appear in a moment.

0 of 8 lessons

[Start lesson 1](https://www.heyrik.com/docs/academy/plan-agent)Checking saved progress…

The Academy teaches one complete workflow: decide what the agent must accomplish, build it in the dashboard, give it facts and a voice, test the cases most likely to fail, send controlled traffic, and improve it from actual call evidence.

**One course, one usable outcome**

Do not begin by writing a giant prompt. Begin with one job and a visible definition of success—for example, “capture a qualified property enquiry and obtain permission for a human follow-up.”

## Build one agent, lesson by lesson

Every lesson has one outcome, exact dashboard actions, a complete example, a practical completion step, and the matching API path. Complete them in order the first time.

[01 — Plan your agent — A one-page agent brief — Loading progress](https://www.heyrik.com/docs/academy/plan-agent)

[02 — Create the agent — A saved agent in Draft — Loading progress](https://www.heyrik.com/docs/academy/create-agent)

[03 — Configure the builder — All six builder areas reviewed — Loading progress](https://www.heyrik.com/docs/academy/configure-agent)

[04 — Write instructions — A complete, testable prompt — Loading progress](https://www.heyrik.com/docs/academy/write-instructions)

[05 — Choose or clone a voice — A voice tested for the job — Loading progress](https://www.heyrik.com/docs/academy/voice-cloning)

[06 — Add trusted knowledge — Grounded PDF or website facts — Loading progress](https://www.heyrik.com/docs/academy/knowledge)

[07 — Test and launch — Controlled live traffic — Loading progress](https://www.heyrik.com/docs/academy/test-and-launch)

[08 — Improve from calls — A repeatable improvement loop — Loading progress](https://www.heyrik.com/docs/academy/improve)

## What you will build

The course uses a property-enquiry agent because it demonstrates inputs, outputs, knowledge, human follow-up, and a measurable result without requiring a complex integration.

**First-agent result**

_ready for controlled testing_

Job**Qualify one property enquiry**

Languages**English and Telugu**

Knowledge**Approved project and pricing PDF or website**

Captured details**property\_type, budget, location, timeline**

Completion**Caller confirms details and follow-up permission**

Fallback**Transfer or arrange a human response when facts are unavailable**

## Choose the right path

Both paths create the same kind of agent. The Academy starts with the dashboard because it makes every setting visible; developers can open the linked API equivalents afterward.

Start here

### Build in the dashboard

Use AI Agents to generate a draft, then work through Assistant Details, Variables, Call Settings, Integrations, Knowledge Base, and Recent Calls. Use Test before switching the agent to Live.

For developers

### Connect your application

Use POST /agents/generate for an unsaved draft, POST /agents to create it, then attach knowledge, choose a voice, assign a number, or dispatch calls through the public API.

**The two Generate actions behave differently**

Dashboard generation creates a saved draft and enriches it. The API endpoint `POST /agents/generate` returns a draft only; your application must review it and call `POST /agents` to save it.

## Know the whole journey before you begin

Each lesson produces something the next lesson needs. Skipping a step usually creates a problem that appears only during a call.

1

### Plan

Choose one audience, one call direction, one primary job, a human fallback, and the data that proves success.

2

### Create

Describe the job, choose a template or start from scratch, and keep the new agent in Draft.

3

### Configure

Review all six builder areas instead of treating the generated draft as finished.

4

### Instruct

Write identity, language, sequence, knowledge, guardrails, recovery, and closing sections.

5

### Equip

Add input and output variables, a matching voice, approved knowledge, and only the integrations the job needs.

6

### Test

Test the happy path, confusion, correction, silence, refusal, missing knowledge, transfer, and closing.

7

### Launch

Switch to Live only after testing, then assign a number or send a small controlled outbound batch.

8

### Improve

Read transcripts and outcomes, change one cause at a time, and repeat the same test set.

## Prepare before opening the builder

Having these decisions and source materials ready turns agent creation into a review exercise instead of guesswork.

**0 of 8 complete**Loading saved progress…

**01**Write the agent’s one-sentence job and definition of success.**02**Decide whether calls are inbound, outbound, or both.**03**Collect an approved PDF or public website containing facts the agent may state.**04**List the information known before a call and the information to capture during it.**05**Choose the human transfer or follow-up rule.**06**Prepare at least eight test scenarios, including difficult and incomplete conversations.**07**If cloning a voice, obtain the speaker’s consent and prepare a clean 20–30 second sample.**08**For API work, create and secure an API key on a backend service—not in browser code.

**Ready?**

Continue to the complete build course. It includes the exact builder order, a full prompt, variables, testing, and a safe launch plan.