What is Dograh?
Dograh is a platform for building phone and browser voice agents. You draw the call as a workflow of agent nodes, tools, transitions and end-of-call outcomes, test it by talking to it in the browser or by chatting in text, then attach it to a phone number. The backend is Python, the licence is BSD 2-Clause, and the whole stack starts from one Docker command.
It connects to Twilio, Vonage, Telnyx, Plivo, Vobiz, Cloudonix and Asterisk ARI for calls, and to whichever speech-to-text, language and text-to-speech providers you choose. Out of the box it runs on Dograh's own models with auto-generated keys, so a first agent needs no API keys at all. An MCP server lets Claude Code, Codex or Cursor inspect and edit agents, and Python and Node SDKs cover outbound calls from code.
Zansat Technologies, the company behind the project, also sells a hosted cloud at 1 cent a minute plus model costs, and an on-premise enterprise licence.
Is Dograh worth it?
Dograh's strongest argument is pace. The repository was created in September 2025, has about 6,156 GitHub stars and 1,589 forks, and shipped four releases between 2 October and 4 November 2026, adding WhatsApp inbound calling in 1.50.0 and Murf as a voice provider in 1.51.0. Release notes reference the Pipecat framework underneath, so the audio pipeline rests on a widely used base rather than a home-grown one.
The workflow builder is the part hosted rivals charge for. Agent nodes, global instructions, tool calls, a QA node and knowledge bases sit in one graph, and Test Chat lets you edit or replay a user turn and regenerate the agent's replies from that point, which shortens prompt iteration considerably. Bringing your own speech and language providers means the per-minute bill is whatever those providers charge plus your server, with no platform margin.
The defaults deserve a careful read. The local quickstart opens a public trycloudflare.com tunnel to the API so telephony webhooks can reach it, and leaves sign-up open until you set ENABLE_SIGNUP=false and re-run the start script. Anonymous telemetry is on unless ENABLE_TELEMETRY=false. A remote install runs a setup script as root, wants at least 8 GB of RAM and 4 vCPUs, and needs TCP 80, 443, 3478 and 5349 plus UDP 49152 to 49200 open for WebRTC; browsers only grant microphone access over HTTPS. The platform holds telephony and model credentials, so treat the server accordingly.
Compared with Vapi or Retell, you give up a managed service level and take on uptime, scaling and call quality yourself; 122 open issues suggest the edges are still being filed down. For a team that needs call recordings and transcripts to stay on its own infrastructure, that trade is usually worth making.
Where it earns its place
A healthcare clinic group with a data-residency requirement
Answer inbound calls and book or move appointments without audio leaving our network
Self-host Dograh on a server meeting the 8 GB RAM, 4 vCPU minimum, connect an Asterisk ARI trunk, and point speech and language at providers the clinic has approved. There is no platform fee; costs are the server and the model providers.
A two-person agency building outbound lead-qualification calls for clients
Prototype a calling script, test it and hand over a working number in a week
Build the flow in Dograh Cloud, iterate in Test Chat by replaying user turns, then connect a Twilio number. Billing is 1¢ a minute plus model usage, with 10 concurrent calls included before a sales conversation.
A developer who already writes code with Claude Code
Describe a voice agent in plain English and have it built inside the workspace
Connect Claude Code to Dograh's MCP server, share a call script with persona, rules and objection handling, and let the agent create the workflow as a draft for review before publishing.
Before you commit
- Setup
- One curl command fetches docker-compose.yaml and a start script; first start takes 2 to 3 minutes and serves the UI on localhost:3010. Remote servers use setup_remote.sh, run as root, which can fetch a Let's Encrypt certificate through sslip.io.
- Integrations
- Twilio, Vonage, Telnyx, Plivo, Vobiz, Cloudonix, Asterisk ARI and WhatsApp inbound; bring-your-own LLM, STT and TTS providers including Deepgram and Murf; MinIO or S3-compatible storage; MCP server; Python and Node SDKs; an OTEL endpoint for telemetry.
- Data and privacy
- Self-hosted deployments keep recordings and transcripts on your infrastructure, but audio and text still go to whichever model providers you configure. Anonymous usage telemetry is on by default and disabled with ENABLE_TELEMETRY=false. Cloud data handling terms are not published in the repository.
- Cost at scale
- Self-hosted cost is servers plus provider rates. Cloud adds 1¢ a minute; concurrency above 10 calls and volume discounts need a sales conversation.
- Limits
- WebRTC needs TCP 80, 443, 3478 and 5349 and UDP 3478, 5349 and 49152 to 49200 reachable; microphone testing requires HTTPS; default of 4 API workers; 122 open issues.
What’s good
- Visual call-flow builder with text and voice testing
- Any speech, language or telephony provider, or none to start
- MCP server and SDKs for building agents from code
- Frequent releases: four in about five weeks
What’s not
- Quickstart opens a public tunnel and leaves sign-up on
- Telemetry on by default
- Remote install needs root, 8 GB RAM and many open ports
- Call quality and uptime are yours to manage
Best for
Engineering teams that want Vapi-style voice agents on their own servers, with call data and provider choice under their control.
Dograh: common questions
- Is Dograh worth it?
- Dograh is the most practical open-source answer to Vapi so far, provided you are ready to run a WebRTC server with dozens of open ports and to read what its quickstart switches on. We rate it 3.9 out of 5.
- How much does Dograh cost?
- Free under BSD 2-Clause to self-host. Dograh Cloud charges a 1¢/min platform fee with 10 concurrent calls included and credits from $5, plus model costs unless you bring your own keys. Committed-volume and on-premise enterprise plans are priced on request.
- What are the downsides of Dograh?
- Quickstart opens a public tunnel and leaves sign-up on; telemetry on by default; remote install needs root, 8 GB RAM and many open ports; call quality and uptime are yours to manage.
- Who is Dograh best for?
- Engineering teams that want Vapi-style voice agents on their own servers, with call data and provider choice under their control.
- How much does Dograh cost for a team or in production?
- Self-hosted cost is servers plus provider rates. Cloud adds 1¢ a minute; concurrency above 10 calls and volume discounts need a sales conversation.
- Is Dograh safe for company data?
- Self-hosted deployments keep recordings and transcripts on your infrastructure, but audio and text still go to whichever model providers you configure. Anonymous usage telemetry is on by default and disabled with ENABLE_TELEMETRY=false. Cloud data handling terms are not published in the repository.
Answers checked Nov 16, 2026.