Quickstart

Five minutes. One agent. Voice, chat, and WhatsApp. No call center, no plumbing — just a prompt and a key.

Prerequisites

Node 18+ (or Python 3.11+, your call), and a Saaya workspace. If you don't have a workspace yet, request access first.

1. Create your workspace

Sign in at app.saaya.ai. The first thing you'll see is a workspace picker — pick one or create a new one. Workspaces are how Saaya isolates billing, RBAC, and data.

2. Install the SDK

The TypeScript SDK is the canonical client. A Python SDK ships with the same surface area.

terminal
npm install @saaya/sdk
# or
pnpm add @saaya/sdk
# or
bun add @saaya/sdk

3. Set your API key

Generate a key from Settings → API Keys in the dashboard. Keys are scoped to a workspace and respect RBAC — never commit them.

.env
# .env
SAAYA_API_KEY=sk_live_...

4. Define your first agent

Saaya agents are versioned configs. The minimum viable agent has a name, a prompt, and a voice — Saaya picks reasonable defaults for everything else.

create-agent.ts
import { Saaya } from "@saaya/sdk";

const saaya = new Saaya({ apiKey: process.env.SAAYA_API_KEY });

const agent = await saaya.agents.create({
  name: "Maya — Pipeline Concierge",
  prompt: `
    You're Maya. You qualify inbound interest for Saaya.
    Voice is warm, direct, and a little dry.
    Always finish a discovery call by booking the next step.
  `,
  voice: { provider: "elevenlabs", voiceId: "rachel" },
  llm:   { provider: "anthropic", model: "claude-opus-4-7" },
  channels: ["voice", "chat", "whatsapp"],
});

console.log(`Agent ready: ${agent.id}`);

Heads up

The first create call also publishes a draft version. Pull the agent into the dashboard to tune voice, knowledge bases, and tools visually — or stay in code.

5. Test it

You can test the same agent on voice, chat, or WhatsApp without deploying.

On voice:

test-voice.ts
// Place a test call to your phone — Saaya will route the call
// to your agent and dial back when ready.
await saaya.sessions.testCall({
  agentId: agent.id,
  toNumber: "+919876543210",
});

On chat:

test-chat.ts
const session = await saaya.sessions.create({ agentId: agent.id, channel: "chat" });
const reply   = await saaya.sessions.send(session.id, "Hi, what does Saaya do?");
console.log(reply.text);

6. Deploy it

Publishing an agent attaches it to a phone number, a chat widget, or a WhatsApp business number. Saaya handles routing on every channel.

deploy.ts
await saaya.agents.publish(agent.id, {
  // Attach the production phone number you configured in the dashboard.
  phoneNumberId: "ph_2N3rH...",
  // (Optional) bind a chat widget to your domain.
  webChat: { allowedOrigins: ["https://acme.com"] },
});

Coming back to a published agent?

Edits roll out as draft versions. Promote to production with saaya.agents.promote(versionId) — and roll back with saaya.agents.rollback(versionId) if a release misbehaves.

7. Where to next

That's it — you have a live agent. From here, the path branches based on what you're building:

  • API reference — drop the SDK, talk to the API directly, get every endpoint.
  • MCP endpoint — let Claude or Cursor query your sessions and KBs.
  • Knowledge bases — ground every answer in your sources, with strict citations.
  • Campaigns — bulk dispatch, retries, fallbacks across channels.
PreviousIntroductionNextCore concepts
Was this page helpful?