Welcome to VoicEra

Open-source, self-hosted voice AI platform for real-time telephony agents in multiple languages.

VoicEra is an open-source platform for building real-time conversational phone agents. It wires speech-to-text, large language models, text-to-speech, and telephony into one self-hosted stack, and gives you a REST API to drive it.

Use VoicEra to run inbound helplines, outbound calling campaigns, IVR replacements, and support hotlines — without building voice infrastructure yourself.

New here? Read What is VoicEra for the plain-language overview, or go straight to the Quickstart if you have Docker ready.

Browse the docs

VoicEra’s documentation is split into three tabs across the top of this site.

Guides

You are here. Learn the system and run it day to day.

• Introduction — what VoicEra is and what it is for
• Quickstart — empty machine to working agent, in five steps
• Using the dashboard — no terminal, no code, just clicking through the web console
• Core concepts — architecture, pipeline, campaigns, providers
• Running VoicEra — campaigns, documents, daily operations
• Deployment — Compose, production, hardening
• Troubleshooting — symptom-first index

Developer

Build on VoicEra, or extend it.

• Services — the containers and the packages they run
• Model server — self-hosted STT, TTS, and LLM
• Clients — the surfaces anything connects through
• Contributing — local setup, adding providers, testing
• Configuration reference — variables, ports, data model
• Dashboard — the Next.js console

API Reference

Every route, extracted from source.

• Introduction · Authentication · Errors
• Agents · Calls · Campaigns
• Phone numbers · Knowledge and RAG
• Users and organisations · Provider credentials
• WebSocket API · Endpoints cheatsheet

A running API also serves an interactive console at http://localhost:8000/docs.

What you get

Capability	What it does
Real-time voice agents	Sub-2-second STT → LLM → TTS loop built on Pipecat.
26 providers	22 cloud vendors — OpenAI, Deepgram, Cartesia, ElevenLabs, Sarvam, Groq, Azure, Google and more — plus the Bhashini and Kenpath adapters and two local providers, behind one registry.
Self-hosted models	Run STT, TTS, and LLM on your own GPUs via the model server.
Telephony	Inbound and outbound calls through Vobiz or Plivo, provider-agnostic.
Outbound campaigns	CSV-driven campaigns with retries, scheduling, circuit breakers, and concurrency limits.
Knowledge base (RAG)	Ground answers in your own documents. See Knowledge base.
Multi-tenant	Organisations, roles, and scoped access built in.
Self-hosted	One Docker Compose stack. Your data, your servers, your model keys.

Where to start

If you are…	Start here
Evaluating or demoing	Prerequisites → Install and run → Create your first agent
Not a developer, just want to run it	Using the dashboard — no terminal needed past the one-time install
Running calls day to day	Operating via the API → Running a campaign
Building or extending	Architecture → Local setup → REST API
Deploying to production	Docker Compose → Public voice URLs → Security hardening

The stack at a glance

Two services, three stores, one job queue, and your choice of AI providers. The full picture is in Architecture.

What VoicEra is not

VoicEra is API-first. Everything is driven by HTTP requests, and /docs gives you an interactive OpenAPI console. The bundled dashboard is a client of that same API.

The Compose stack also starts a web dashboard on port 3000. See Dashboard.

VoicEra also does not resell telephony or model capacity. You bring your own provider accounts, or you run the models yourself.

Need help?

• Start with Troubleshooting.
• Look up unfamiliar terms in the Glossary.
• Read the Contributing guide before opening a pull request.