Quick Start
Create your first AI voice agent in 5 minutes. Step by step.
By the end of this page you'll have a working AI agent you can talk to in your browser.
What you need
- An email address
- 5 minutes
- (Optional) ₹100 of free credits — added automatically when you sign up
Step 1: Sign up
Go to app.osmtalk.com and create your account. You'll land in a dashboard.
The left sidebar has six things:
- Agents — the bots you build
- Calls — history of every conversation
- Phone Numbers — real numbers people can dial
- Campaigns — bulk outbound calling
- Settings — account, billing, storage, webhooks
- Team — invite coworkers
That's it. No setup wizards, no email confirmation loops.
Step 2: Create your first agent
Click Agents in the sidebar, then click the + New Agent button (top right).
You'll see four ways to start:
| Option | What it does | Best when |
|---|---|---|
| Pick a template | Fills in a tested prompt for a common role | First time — just pick one |
| Blank | Empty page; you write the prompt | You know exactly what you want |
| AI Generated | Describe in one sentence; AI writes the prompt | "I want a bot that..." in plain English |
| From an existing agent | Clone a previous agent | You're tweaking, not creating |
Pick a template. The five templates:
- Maya — Customer support
- Arjun — Sales agent
- Priya — Receptionist
- Nisha — Survey caller
- Kavi — Appointment booking
For this walkthrough, pick Priya (Receptionist). Click Create.
Step 3: Talk to your agent
You're now on the agent's page. Look at the right side of the screen — there's a "Test Call" button. Click it.
Your browser will ask for microphone permission. Click Allow.
You'll hear Priya say hello. Try saying:
"Hi, I'd like to book an appointment with Dr. Smith next Tuesday afternoon."
Priya will respond naturally — ask for your name, confirm the time, etc. When you're done, click End Call.
You just had your first AI conversation. That conversation was real — the same tech handles a phone call.
Step 4: See what happened
Click Calls in the sidebar. Your test call is at the top of the list. Click it.
You'll see:
| Duration | How long the call was |
| Cost | Exact ₹ amount (around ₹0.50–1 for a 1-minute test) |
| Transcript | Every word spoken, who said it |
| Recording | A playable audio file (you can download it) |
| Latency | How fast the AI was responding |
This is everything you'd see for a real phone call too.
Step 5: Change the agent's behavior (optional)
Want Priya to speak Hindi? Want her to mention specific business hours? Go back to Agents → Priya → Configure.
The most useful field is System Prompt (top of the page). It's a paragraph in plain English that tells the agent who it is. Change it to whatever you want:
You are Priya, the receptionist at Greenleaf Dental Clinic.
We're open Monday-Friday 9 AM to 7 PM IST. We do general dentistry,
cleanings, and emergency tooth extractions.
If someone wants an appointment, ask for:
1. Their name
2. Their phone number
3. The reason for the visit
4. Preferred date and time
If it's an emergency, ask them to come in within 30 minutes.
Keep responses short and warm. Don't repeat the same phrase twice.Click Save. Test again — Priya now knows about your specific clinic.
What's next?
Pick the next step based on what you want to do:
| You want to... | Read |
|---|---|
| Put this on your real phone number | Phone Numbers |
| Embed a chat/voice widget on your website | Widget |
| Let the bot save leads to your CRM | Actions & Tools |
| Call 1,000 people from a list | Campaigns |
| Change the voice or accent | Voice settings |
| Understand the AI model choices | Creating agents |
| Track all your costs | Billing |
| Place calls from your code | TypeScript SDK + runnable examples |
| Iterate on a prompt without spending credits | Use client.eval.simulate() — see Example 04 in the SDK repo |
| Tune VAD / turn-taking | VAD & Turn Detection deep dive |
| Every advanced setting explained | Advanced Settings reference |
Common confusions when starting out
"I keep getting cut off mid-sentence."
The AI thinks you're done talking after a short pause. Go to Configure → Advanced → VAD & Turn Detection, raise Stop seconds to 0.5. Full guide: VAD troubleshooting.
"It speaks too fast / slow." Configure → Voice → Speed. Range 0.7 (slow) to 1.2 (fast). 1.0 is normal.
"The voice sounds robotic." The default voice (Helena) is the most natural for English. For Hindi/Tamil/Telugu, try ElevenLabs voices in Voice settings.
"I tested twice and used ₹2 of credits." Test calls cost real money — they go through the same AI providers. A 1-minute test is typically ₹0.50-1. Your free trial credit lasts ~3 hours of testing.
"My agent gives weird answers." Almost always a system-prompt issue. Be more specific in the prompt. Tell it what NOT to do, not just what to do. Example: instead of "be helpful", write "if the caller asks for prices, say 'I don't have pricing details — let me transfer you to sales'."
You'll learn the rest by trying. Go play.
Skipping the dashboard — for developers
If you'd rather drive everything from code instead of clicking around:
git clone https://github.com/osm-API/osmtalk-examples.git
cd osmtalk-examples/01-personalized-call
cp .env.example .env # fill in OSMTALK_API_KEY + AGENT_ID + PHONE_NUMBER_ID
npm install
npm run check # preflight — verifies your config before any real call
npm start # ← places a real outbound callnpm run check is the killer feature — it validates your env, authenticates with the API, confirms the agent + phone-number IDs exist in your org, checks provider health, and runs CSV validation. If it passes, the real run almost certainly works. Saves the "why doesn't this work" debugging loop.
Full SDK reference: TypeScript SDK · Examples (4 scenarios).