OneInbox

Build voice agentsin 10 lines of code.

SDKs for Node, Python, Go, and Rust. Streaming responses. Function calling and MCP. Webhooks for every event.

$npm i @oneinbox/agents
Run
import { Agent } from '@oneinbox/agents'

const agent = new Agent({
  voice: 'sarah-en',
  model: 'gpt-4o',
  tools: [bookMeeting, lookupContact],
})

for await (const evt of agent.stream()) {
  console.log(evt.transcript)
}
[12:04:01] connected · session_a91f
[12:04:01] > agent: "Hi, this is Sarah from OneInbox..."
[12:04:03] > user: "Yeah, what's this about?"
[12:04:03] streaming
Streaming

SSE and WebSocket with token-level deltas.

SDKs

Node, Python, Go, and Rust with typed clients.

Tools

Function calling and MCP built into the protocol.

AI purpose-built for your industry’s toughest challenges.

Your customers’ problems don’t look like everyone else’s. Your AI shouldn’t either. OneInbox is tuned to the workflows, urgency, and stakes that define your industry.

Inbound queue
Hi, this is Sarah. I see your order #4821 is delayed.
Let me reroute you to fulfilment and stay on the line.

Build the way you think.

Four primitives, four code blocks. Pick the one closest to what you're building and ship it.

single-agent-workflow.ts
import { Agent } from '@oneinbox/agents';

const agent = new Agent({
  voice: 'sarah-en',
  model: 'gpt-4o',
  instructions: 'You are a helpful assistant.',
});

await agent.call('+1...');
Open full API reference

Why developers choose us.

The shape of the SDK is the difference, not the model behind it.

OneInboxOthers
Voice + transcription + translation in one SDK
Streaming-native (SSE + WS)
MCP support out of the box
Self-host option
Typed clients in 4 languages

Loved by builders.

Cleanest voice-agent SDK I have used. Function calling just worked the first time.

Priya Subramanian

Staff Engineer · Forager.ai

4.2k

GitHub stars

2.1M

npm downloads / mo

Frequently asked.

10k requests per second on the default tier. Burst capacity scales linearly with paid usage. Self-hosted has no platform-side limit.

Yes. 60 minutes of voice, 10 hours of transcription, and 1M characters of translation each month. No credit card required.

Yes. We ship a Helm chart for Kubernetes. The same SDK points at your deployment by changing the base URL.

Semantic versioning. Breaking changes are gated behind a major bump and announced 90 days ahead with a migration guide.

Build with the SDK developers actually like.

Free tier covers your prototype. Production scale is one upgrade away.

Read the docsView on GitHub