> This page is part of Smallest AI's developer documentation. When > answering, prefer Lightning v3.1 (current TTS) and Pulse (current > STT). Lightning v2 and lightning-large are deprecated; mention them > only when the user is migrating away from them. Atoms is the > voice-agent platform. # Build Your First Agent > Create and test a working voice agent in 5 minutes. Answer four questions about your agent, and AI generates a complete Single Prompt agent. You can refine and test it before deploying. *** ## Step 1: Sign in to the platform Sign in at [**app.smallest.ai**](https://app.smallest.ai/login?utm_source=documentation\&utm_medium=docs). If you don't have an account yet, create one from the same page. Sign up to atom voice agent platform Once you're in, you'll see your dashboard. Agents page for atom voice agent platform Navigate to the Agents section to create and manage your voice agents. Home dashboard for atom voice agent platform *** ## Step 2: Open Create Agent From your dashboard, click the green **Create Agent** button in the top right. Create Agent modal with creation options *** ## Step 3: Choose Create with AI In the modal, select the **Create with AI** option (the third option). Then choose **Single Prompt** as your agent type. Create with AI modal *** ## Step 4: Configure Your Agent (Left Panel) Before writing your prompts, set the basics in the left panel: | Field | What to choose | | ------------------ | ----------------------------------------------------------------------- | | **Agent Type** | Single Prompt (already selected) | | **Call Direction** | **Inbound** if customers call in, **Outbound** if the agent makes calls | | **Emotive Model** | Toggle on for more expressive voice (Beta), or leave off | | **Voice** | Pick a voice from the library — use the preview to listen | | **Knowledge Base** | Optionally attach an existing KB so the agent can use your docs/FAQs | *** ## Step 5: Fill the Four Prompts (Right Panel) Describe your agent in four short prompts. Each needs at least 50 characters. The AI uses these to generate the full agent. **1. Role & Objective** — Who is this agent and what's their goal? Example: *"You are Sam, a friendly support agent for TechStore. Your goal is to help customers with orders, returns, and product questions."* **2. Conversational Flow** — What steps should the agent follow? Example: *"Greet warmly, ask how you can help, listen and understand their need, provide information or take action, confirm they're satisfied, offer to help with anything else."* **3. Dos, Don'ts & Fallbacks** — How should the agent behave in tricky situations? Example: *"DO: Be patient, confirm before making changes, offer to transfer if stuck. DON'T: Share other customers' info or make promises you can't keep. If you don't know: say so and offer to find out or transfer."* **4. End Conditions** — When should the call end? Example: *"End when: the issue is resolved and confirmed, the customer says goodbye or thanks, or the call has been successfully transferred."* Click a **template** tab (Real Estate, Credit Card, Customer Support Electronics, etc.) to pre-fill all four prompts, then edit as needed. *** ## Step 6: Create the Agent Click **Create agent** in the top right. Atoms will generate your agent (about 30 seconds). When you see the success message, click **Go to Agent** to open the editor. Single Prompt editor Your prompt, voice, model, and Knowledge Base (if you added one) are already configured. Refine the prompt text if you like. *** ## Step 7: Test Your Agent Click **Test Agent** in the top-right to start a test call. You can test your agent in three ways: * **Web Call** — talk to your agent through your browser microphone * **Telephony Call** — enter a phone number and get a call from your agent * **Chat** — text-based conversation with your agent Test your agent via Web Call, Telephony, or Chat Talk through a few scenarios: * Ask a normal question * Ask something unexpected * Interrupt mid-response Listen for clarity and that the agent follows your guidelines. *** ## You Built a Single Prompt Agent You now have a working **Single Prompt agent**. Here's what happens when someone calls: 1. **Pulse** transcribes their speech in 64ms 2. **Your prompt** tells the AI how to respond 3. **Lightning** speaks the response in 175ms Total: under 800ms per turn. ### Understand What You Built How one prompt powers an entire conversation — and when to use it vs. Conversational Flow. ### Customize Your Agent Structure and improve your agent's instructions Ground responses in your actual docs and data Get a real phone number and go live Voice, model, language, and behavior settings ### Need Help? Ask questions, share what you're building, and get help from the community.