Voice AI API, SDKs
built for developers
Ship voice AI agents in minutes with a REST API, Python and JavaScript SDKs, webhooks and a no-code flow editor. BYOK providers, real-time WebRTC, native SIP for phone calls.
60-second quickstart
Place your first outbound voice call with a single API call. Pick your language — curl, Python, or JavaScript.
curl -X POST https://api.thinnest.ai/v1/calls \
-H "Authorization: Bearer $THINNESTAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"agent_id": "agent_hindi_support",
"to": "+919876543210",
"metadata": {
"customer_id": "cust_123"
}
}'from thinnestai import Thinnest
client = Thinnest(api_key="YOUR_API_KEY")
call = client.calls.create(
agent_id="agent_hindi_support",
to="+919876543210",
metadata={"customer_id": "cust_123"},
)
print(f"Call started: {call.id}")import { Thinnest } from "@thinnestai/node";
const client = new Thinnest({ apiKey: process.env.THINNESTAI_API_KEY });
const call = await client.calls.create({
agentId: "agent_hindi_support",
to: "+919876543210",
metadata: { customerId: "cust_123" },
});
console.log(`Call started: ${call.id}`);Developer capabilities
REST API
Create agents, start calls, query transcripts and outcomes — everything available over HTTPS. OpenAPI spec published.
Python & JavaScript SDKs
First-class Python and TypeScript SDKs for the server-side. Web client SDKs for browser-based voice widgets.
Webhooks everywhere
Every agent event — call started, transcript updated, tool called, outcome captured, call ended — is a webhook you can consume.
Real-time WebRTC
Low-latency voice over WebRTC for browser widgets. Native SIP for phone calls. Both share the same agent runtime.
Flow versioning
Agent flows are versioned like code. Publish, revert, A/B test, diff between versions — all safe for production.
Full observability
Every call produces a transcript, recording URL, latency breakdown, provider cost breakdown and outcome. Export to your own stack.
What developers build on ThinnestAI
Outbound campaign triggered by a webhook
Your CRM fires an event → ThinnestAI places a Hindi outbound call within seconds → outcome written back to your CRM via webhook.
Inbound support widget on your website
Drop our JS widget into your React / Next app. Users click to talk to a voice agent in their language, over WebRTC, with zero phone number required.
Programmatic agent creation
Create and configure agents via REST — good for multi-tenant apps where each customer gets their own agent.
Custom tool calling
Expose any REST endpoint as a tool the agent can call — database lookups, CRM updates, third-party APIs, anything.
Developer FAQ
Is there a free tier for developers?
+
Yes. Sign up for free with welcome credits — no card required. Free tier includes platform-managed Sarvam LLM and STT, so you can build your first agent without bringing any API keys.
What languages do the SDKs support?
+
Python and TypeScript/JavaScript are first-class server SDKs. Web client SDKs for browser-based voice widgets. REST API is language-agnostic — use curl, Go, Ruby, PHP or anything that can make HTTPS calls.
How do I connect my own provider API keys (BYOK)?
+
Bring your own keys in the dashboard or via the REST API. Each agent can be configured to use specific providers for LLM, STT and TTS. We support OpenAI, Anthropic, Groq, Gemini, Sarvam, Mistral, DeepSeek (LLM), Deepgram, AssemblyAI, Sarvam Saaras (STT), ElevenLabs, Cartesia, Sarvam Bulbul (TTS), and Twilio + Vobiz for phone.
Is there an OpenAPI spec I can generate clients from?
+
Yes. The full OpenAPI 3.1 spec is published at docs.thinnest.ai/api-reference. You can generate clients for any language using the standard OpenAPI generator tools.
How do webhooks work?
+
Every agent event (call started, transcript updated, tool called, outcome captured, call ended) fires a webhook to your configured endpoint with a signed payload. You can subscribe to all events or a specific subset. Standard HMAC signature validation.
Can I use ThinnestAI from a serverless function?
+
Yes. The REST API and SDKs work from any serverless environment — AWS Lambda, Vercel Functions, Cloudflare Workers, Google Cloud Functions. WebRTC widgets use a short-lived token you generate server-side, so your API key never reaches the browser.
What's the rate limit?
+
Free tier is 10 requests/second and 5 concurrent calls. Paid tiers scale to 100+ requests/second and hundreds of concurrent calls. Enterprise customers get custom quotas.
Do you have a sandbox / test mode?
+
Yes. Test mode calls use the same agent logic but route to a simulated phone with no real telephony cost. Good for CI/CD pipelines and integration tests.
Start building with free credits
No card required. Platform-managed Sarvam on the free tier — you can build your first Hindi voice agent without bringing a single API key.