Categories
Self-Hosted AI Agents Telephony Voice AI Conversational AI Speech to Text Call Center AI Phone AgentsStars
Forks
Watchers
Developer links
OpenRing
Phone calls answered by AI agents cost only what your providers charge, because OpenRing never adds a per-minute platform fee. It speaks Vapi's assistant JSON, tool calls, and webhooks, so a Vapi client moves over by changing its base URL, and Import copies assistants, squads, tools, and Twilio numbers in one click. Calls run over Twilio, Telnyx, Plivo, Vonage, or any SIP trunk, and campaigns redial numbers that were busy or unanswered. Build each agent from a transcriber, a model, and a voice, choosing among Deepgram, OpenAI, Groq, and ElevenLabs or a free local trio of Whisper, Ollama, and Kokoro, then press Test to talk to it from the browser with no phone number. Model routes (Fastest, Cheapest, Reliable, Custom) switch to a backup when an answer is slow or fails. Tools call your server, post to n8n, Make, or Zapier webhooks, reach MCP servers, transfer calls, or hang up, while squads hand callers between assistants and knowledge bases answer from PDFs. Evals test an assistant against mock conversations before real callers do, escalations ring an on-call list until someone confirms, and an MCP server lets Claude Code or Cursor place calls. The dashboard shows every call with its transcript, latency per turn, and cost per provider, tallies the fees avoided against hosted pricing, and sends an alert when calls fail. Running on a dedicated VPS on RepoCloud with guaranteed CPU, RAM, and SSD, full root SSH access, and a browser serial console. MIT licensed.
Benefits
- Pay providers, not a platform
- Deepgram, OpenAI, Twilio, and every other provider bill you directly at their own prices, while the dashboard tallies the fees you avoided against hosted per-minute pricing.
- Move over from Vapi quickly
- Assistants, tools, provider names, and webhooks share Vapi's JSON, so clients only change a base URL, and the Import button copies assistants, squads, tools, and Twilio numbers across.
- Try voice agents for free
- Whisper transcribes, Ollama thinks, and Kokoro speaks on your own hardware at no per-minute cost, and browser test calls need no phone number or carrier account.
- Own your keys and transcripts
- Keys, recordings, and transcripts stay in a local SQLite database with no telemetry, and the dashboard sits behind its own password with slowed-down retries.
Features
- Vapi-compatible REST API
- Assistants, squads, tools, phone numbers, and server messages use Vapi's JSON, and Vapi's official Python SDK passes most of the project's own compatibility checks.
- Multi-carrier phone lines
- Handles calls over Twilio, Telnyx, Plivo, Vonage, and SIP trunks, plus inbound Genesys Cloud, with campaigns that redial busy or unanswered numbers up to five attempts.
- Pluggable voice pipeline
- Each call chains a transcriber, model, and voice from Deepgram, AssemblyAI, Whisper, OpenAI, Anthropic, Gemini, Groq, Ollama, ElevenLabs, Cartesia, or Kokoro, orchestrated by Pipecat.
- Model routes with failover
- Fastest, Cheapest, Reliable, and Custom presets order backup models so the next one takes over when a model fails or answers slowly, and transcribers and voices get routes too.
- Tools, MCP, and squads
- Function and API-request tools reach your server or n8n, Make, and Zapier webhooks, the mcp tool uses MCP servers over Streamable HTTP, and squads hand callers between assistants.
- Call analytics and alerts
- Every call records its transcript, latency per step, and cost per provider, with optional AI-written summaries, an alert webhook for failed calls, and a Needs attention panel on Home.