Quickstart

View as Markdown

Create a voice agent, test it in test agent, deploy it to a phone number, and make your first call.

Start building

You create and manage agents on the Agents page (under Build). Two ways to start a new one:

  • Describe what your agent should do in the prompt box, this starts a new Genie session that builds the agent for you.
  • Create from Scratch for a blank agent.

You can also reopen a recent agent to keep working.

Step 1: Create an agent

  1. Log in at platform.sarvam.ai and open Agents under Build.
  2. Either describe what your agent should do in the prompt box (this starts a Genie session that drafts the agent) or click Create from Scratch for a blank agent.
  3. You land on the Canvas, the visual agent builder where you configure everything.

From here, keep refining with Genie or write the prompt yourself. The steps below use the prompt-writing approach.

The Agents page with the prompt "What should your agent do?", a Create from Scratch button, and a Recents table of agents.

The Agents page: describe what your agent should do for Genie, or Create from Scratch.

Step 2: Configure a system prompt and initial message

  1. On the Canvas, open the agent’s Instructions tab.
  2. Set the Greeting (initial message) the agent speaks when the conversation starts. Insert variables as chips (for example {} developer_name).
  3. Write the system prompt: who the agent is, what it should accomplish, and how it should behave.
  4. Keep the first prompt short; you can refine it later in the Instruction guide.

The agent Instructions editor showing a Greeting with a developer_name variable chip and a system prompt, with the Genie panel open on the right.

Instructions on the Canvas: Greeting (initial message) and system prompt, with Genie open on the right.

Step 3: Select language and voice

  1. Open Settings in the agent.
  2. Under Language personalisation, set the starting language, turn on Switch language during call if needed, and choose the languages allowed.
  3. Configure voice under Speakers & voice.

See Conversation settings → Language personalisation for the full set of language controls.

Agent Settings showing Language personalisation with Switch language during call on, English and Hindi selected, and Starting language set to English.

Settings → Language personalisation: starting language, switch during call, and languages allowed.

Step 4: Test in test agent

  1. Open test agent and speak to your agent normally.
  2. Test how it handles the conversation: greeting, turn-taking, language, and flow.
  3. Once you’re satisfied, move on to the next step.

See Deploy for phone-number-based testing options.

The test agent panel with a conversation transcript showing the agent's initial greeting and a follow-up, next to language and settings controls.

test agent: chat with the agent and confirm the greeting and flow before deploying.

Step 5: Deploy to a phone number

  1. Go to Phone Numbers in the Deploy section.
  2. Get a phone number one of two ways:
    • Bring your own telephony: connect your existing provider and import your numbers. See Bring Your Own Telephony.
    • Rent from Sarvam: get a number through Sarvam’s Vobiz managed partner account. See Rent from Sarvam.
  3. Assign the number to your agent for inbound calls.

See Phone Numbers for provider-specific setup.

The Phone Numbers page for a Sarvam Vobiz connection, showing Total/Active/Archived counts and a table of phone numbers with status.

Phone Numbers: connections, number list, and Buy number.

Step 6: Make your first call

Make a call one of two ways:

  • Outbound: in Outbound Campaigns, configure a campaign from your agent to a cohort of contacts. See Campaigns.
  • Inbound: deploy the agent under Inbound Calls, then call the agent’s number from any phone.

Then open Monitor → Agent Analytics to review the call, transcript, and recording.

Unified View Details panel open on the Transcript tab, showing the agent and caller conversation with an audio player at the bottom.

Agent Analytics call detail: transcript and recording for a completed call.

Next: explore the full Build section to add tools, knowledge, states, and evaluation criteria.