Self-host VoicEra
v3.1Open-source voice AI for self-hosted phone agents
- Stars
- 316
- CPU
- 2 cores
- RAM
- 4 GB
- Storage
- 20 GB
- Ports
- 8000 (API), 7860 (Voice Runtime), 3000 (Dashboard)
- Requires
- MinIO (S3-compatible object storage), PostgreSQL (via FerretDB), Redis
What is VoicEra?
VoicEra lets operators run phone helplines, calling campaigns, and support lines without building voice infrastructure. The stack separates a control plane (API, dashboard, workers) from a real-time voice runtime. It uses FerretDB on PostgreSQL for data, MinIO for recordings, and Redis for queuing. The platform integrates 22 cloud AI vendors like OpenAI and Deepgram, plus Bhashini and Kenpath adapters for Indian languages, and two local providers for self-hosted models. Telephony works with Vobiz and Plivo. Multi-tenancy with roles is built in. Deployment uses a single Docker Compose file for evaluation, with production guidance for hardening, TLS, external stores, and scaling. Licensed Apache 2.0 by COSS-India.
Should you self-host VoicEra?
VoicEra scores 47/100 on selfhost.directory (how scoring works) and ranks #36 of 69 in AI Chat. It has 316 GitHub stars.
There was no stable release in the past 12 months, so check the repository's recent activity before you rely on it.
Setup is advanced: plan for several services and some ongoing maintenance. It is Apache-2.0-licensed and written mostly in Python.
Its closest alternative on selfhost.directory is ChatGPT Next Web (85/100, 88 905 stars). It scores higher overall, so compare the two before you pick. See VoicEra vs ChatGPT Next Web side by side.
Features
- Data sovereignty
- Indian language support
- Knowledge base (RAG)
- Multi-Provider AI Support
- Multi-tenancy
- Outbound campaigns
- Provider-neutral architecture
- Real-time voice agents
- REST API with OpenAPI
- Self-hosted models
- Telephony Integration
- Web Dashboard
VoicEra README
FAQ
How do I back up?
Back up all three durable stores together: PostgreSQL (via pg_dump), MinIO, and the Chroma volume. Redis is ephemeral. Store PROVIDER_AUTH_ENCRYPTION_KEY with the backup.
Where are recordings stored?
Recordings are stored in MinIO under voicera-calls/{org_id}/{call_id}/. They should be fetched through the authenticated API rather than directly from the bucket.
How many calls can run at once?
DEFAULT_ORG_CONCURRENCY_LIMIT caps simultaneous calls per organisation, default 10. Campaigns can set a lower max_concurrency. Telephony account channel limits or model vendor rate limits usually bind first.
Is there a default login?
No. The first POST /users/signup creates the user, an organisation, and a super_admin membership. Whoever signs up first owns the deployment.
Do I need a public URL?
Only for real phone calls. Your telephony provider fetches /answer over HTTPS and opens a WSS connection for audio. For evaluation, a websocket agent needs no public URL and no telephony account.
Can I use only OpenAI?
Yes. OpenAI registers speech-to-text, text-to-speech, and a language model, so one credential covers all three. Mixing providers is also common.
Do I need a GPU?
Not with cloud model providers. The core stack runs on 2 CPU cores and 4 GB of RAM. A GPU is only needed to self-host models with the model server.
Where is the dashboard?
The dashboard runs at http://localhost:3000 and covers agents, numbers, campaigns, knowledge documents, call history, and per-call latency. The core stack also provides an interactive OpenAPI console at http://localhost:8000/docs.
More in AI Chat
vortexSelf-hosted RAG application for document chat with multiple LLM providers. Combines vector and keyword search, re-ranks results, and shows source citations.
webwhizSelf-hosted, open-source AI chatbot that trains on your website data to provide instant customer support with 100+ language support and GDPR compliance.
WhirlWhirl is a self-hosted AI chat app routing conversations through OpenRouter for every major model, with long-term memory, living documents, and MCP integrations.
WhoDBOpen-source, self-hosted database workspace: browse and edit data, query in a scratchpad, visualise schema graphs, with optional Ollama, OpenAI or Anthropic AI chat.
WorkWitRun your own AI assistant for office work — handles contract review and resume screening, with 40+ skills, MCP connectors, and team access control.
YakGPTYakGPT is a self-hosted, open-source ChatGPT web UI with speech-to-text and text-to-speech, connecting directly to the OpenAI API using your own key.
YALSYALS is a self-hosted, open-source OpenAI-compatible API server for GGUF models via llama.cpp, with continuous batching and JSON schema support.
ZimaOS-BlueZimaOS Blue is an open-source, self-hosted AI agent runtime in Go that runs locally on any device with voice, browser automation, deep research, and 20+ IM integrations.
zolaSelf-hosted open-source AI chat for OpenAI, Mistral, Claude, Gemini, and local Ollama models. BYOK, file uploads, dark mode, Supabase auth.
小女仆机器人Self-hosted Rust AI agent bot for QQ, OneBot 11, WeChat with memory, RAG, todos, RSS, web search, tool calling, dice, multi-model routing — ~25 MiB, SQLite.
AChatSelf-hosted, local-first AI chat platform for teams: shared conversations, roles, billing, moderation and plugins, on Node.js with PostgreSQL and Redis.
agentsInkeep Agents lets teams build AI assistants and workflows via drag-and-drop builder or TypeScript code with full sync. Self-host via Docker or Vercel.
0 comments