Open Voice Agent: AI voice agents for phone calls and websites.

Self-hosted, tool-capable, and built for realtime speech. Bring your own provider keys and keep client billing isolated per workspace.

Open source
BYO keys
Realtime voice
Telnyx/Twilio-ready
Product

Configure, deploy, and operate voice agents — without glue code.

Open Voice Agent ships a dashboard for agent setup, a web embed widget, and phone call handling via telephony providers. It’s built for a “done-for-you” operator model: clients bring their own keys, and you keep billing isolated per workspace.

Dashboard-first agent configuration: prompt, tools, and deployment settings.

Website widget for realtime web voice (microphone consent gated).

Inbound calling: connect Telnyx (Twilio optional) and route to agents.

Dashboard preview

Replace this placeholder with a screenshot once you’ve finalized the homepage copy.

Agents
Create / test / deploy
Workspaces
BYO keys per client
Calls
Analytics + transcripts
Embed
Widget + allowlist
How it works

From zero to first agent in three steps.

The workflow is designed for operators: set up a workspace, add keys, then create and deploy an agent.

1
Create a workspace

Create a workspace for a client so data and keys are isolated.

2
Add provider keys

Add keys in Settings → Workspace API Keys for Grok/OpenAI and telephony verification.

3
Create and deploy

Create an agent (prompt + tools), then deploy to web embed or inbound calling.

Features

Everything you need to run voice agents for clients.

Built for a production operator workflow: multi-tenant, key-isolated, and focused on realtime speech.

Realtime providers

Grok realtime primary + OpenAI fallback (workspace-scoped).

Tool calling

CRM, calendar, SMS, and call controls via built-in tools.

Workspace key isolation

Bring your own keys. Keep client billing isolated per workspace.

Website widget

Embeddable voice widget with domain allowlisting.

Phone calls

Inbound calling via Telnyx (Twilio optional).

Analytics + exports

Call logs, transcripts, and export workflows for operators.

Security

Operator-friendly defaults, without sacrificing control.

This is not legal advice. It’s a technical posture designed to make responsible operation easier across regions.

Invite-only by default (public registration disabled).

Keys are workspace-scoped; clients can rotate keys without reading existing secrets (member role).

Retention tooling exists to purge old call media on a schedule.

FAQ

Common questions.

Short answers grounded in how this product is built today.

Is Open Voice Agent open source?+

Yes. This project is designed to be self-hosted so you can own your infrastructure and iterate fast.

Do you store my provider keys?+

Provider keys are managed per workspace so each client can bring their own keys. The UI only shows whether a key is set, not the secret value.

Can I run one instance for many clients?+

Yes. The product is multi-tenant via organizations + workspaces, with workspace-scoped keys and admin/member roles.

Does it work for phone calls and websites?+

Yes. You can embed an agent on a website and also connect inbound calling via supported telephony providers.

What do I need to deploy?+

A VPS with Docker, Postgres + Redis, and your chosen provider accounts (realtime + telephony).

Ready to build your first voice agent?

Invite-only by default. Ask your admin for access.