Timbre
Start building
Live demo — click below to talk to a Timbre agent

A voice agent on your website.One line of script.

Timbre turns a single <script>tag into a full voice conversation in your visitor’s browser. Bring your own prompt, pick a voice, plug in MCP servers and HTTP tools — your agent answers in <500ms.

Start building free

No credit card · First 1,000 minutes on us

acme.com
Live web call
Visitor · Bella (Cartesia)
WebRTC · en
Visitor
Hey, what does Timbre actually do?
Bella
We turn one script tag into a live voice agent on your site — speech, tools, MCP, the works. Want me to show you the embed snippet?
Visitor
Yeah, drop it.
sonic-3 · 320ms
Embed snippet
<script
  src="https://app.timbre.app/embed.js"
  data-key="pk_live_..."
  async
></script>
On-call tools
  • lookup_order(id)200 OK
  • mcp:zapier.send_email200 OK
  • book_meeting(time)…
50+
Voices
<500ms
Latency
native
MCP
<500ms
voice-to-voice
50+
Cartesia voices
Multi
lingual STT
MCP
native tools
What ships today

Voice that talks to your stack.

Embed on any site, attach your tools, ship in an afternoon. Phone number routing comes next.

Embed widget
One script tag. That’s it.

Drop a single script onto any page and visitors get a floating call bubble. Per-agent public key, allowed-origin allowlist, theming you control from the dashboard.

  • Custom button shape, color, icon
  • Per-key concurrent + daily quotas
  • Live transcripts streamed back to you
MCP-native
Plug in your stack.

Attach any MCP server — Zapier, Linear, your own — and its tools appear automatically inside the agent. Per-agent scoping, encrypted headers at rest.

  • HTTP & SSE transports, auto-detected
  • Encrypted Authorization headers
  • Tools resolve at dispatch, not at build time
Lifecycle hooks
Pre-call. On-call. Post-call.

Define HTTP tools that fire at each phase. Pull caller context before the agent speaks, expose tools for the LLM to call mid-conversation, hand off transcripts to your CRM after hangup.

  • Mustache-style {{var}} templating
  • JSON-Schema params for on-call tools
  • Headers + body templates encrypted at rest
Voices & emotion
Sound human, on purpose.

50+ Cartesia voices in dozens of languages, dynamic SSML emotion based on caller sentiment, multilingual auto-detection via Deepgram nova-3.

  • Cartesia sonic-3 TTS
  • Deepgram nova-3 multilingual STT
  • Emotion / speed / volume picked per turn

SoonInbound & outbound phone calling via your own SIP / Twilio trunk — join the waitlist.

How it works

From prompt to live call in minutes.

  1. 01
    Create an agent

    Pick a Cartesia voice, write a system prompt and first message, choose a language (or let multi auto-detect).

  2. 02
    Attach tools

    Connect MCP servers and define HTTP tools for pre-call context, on-call functions, and post-call webhooks.

  3. 03
    Embed

    Toggle Embed on the agent, allowlist your domains, and drop the snippet onto your site. Visitors talk to your agent in their browser.

Hear it before you build it.

Click below to talk to a real Timbre agent running this exact stack. Then go ship one yourself.

Create an account