Choosing Dograh vs Cognigy comes down to who runs the platform and how far it reaches. Cognigy is a closed suite owned by NiCE, sold to new customers as SaaS, with voice and messaging channels. Teams that want to own the platform get open-source Dograh, on their own servers or Dograh Cloud, with any model they choose.
Key Takeaways
- Cognigy is closed, owned by NiCE and sold only as SaaS to new customers.
- Cognigy wins on messaging channels, CXone fit and tools for human agents.
- Self-hosting an open platform moves servers, upgrades and call data to your team.
We are the team behind open-source Dograh, so this comparison tries hard to be fair. Cognigy, owned by NiCE since September 2025, is an enterprise suite for voice and messaging. We go row by row, then through building, channels, models, phone and cost, and say where Cognigy wins.
Dograh and Cognigy, row by row
NiCE closed its purchase of Cognigy on 8 September 2025 and sells it both inside NiCE CXone and as a product of its own. Each row below compares the two products on the same question.
| Row | Cognigy | Dograh |
|---|---|---|
| Who owns it | NiCE, since 8 September 2025 | Zansat Technologies, which builds Dograh in the open |
| Code | Closed and licensed | Open source under BSD 2-Clause, on GitHub |
| Who builds the agent | Your team, in AI Agent Studio or the new Agentic Experience Studio | Your team, by describing the agent in the Agent Builder or editing on the canvas |
| Voice and chat | Voice, webchat, WhatsApp and SMS, plus Live Chat and Agent Copilot for human agents | Phone and browser voice calls, a web chat widget and inbound WhatsApp voice calls |
| Deployment | Cognigy SaaS or inside NiCE CXone. On-prem closed to new customers | Your servers with Docker Compose or Helm, Dograh Cloud, or on-prem and private cloud on Enterprise |
| Data location | The Cognigy SaaS or CXone region you buy | Your own servers when self-hosted, Dograh's cloud on Dograh Cloud |
| Models | Six LLM providers on your keys, included models on CXone installs with a new SKU | Dograh's models, your own keys, or open-weight models you host |
| Speech-to-speech | One model in the docs, marked experimental | Seven options, from OpenAI Realtime to Gemini Live |
| Phone | SIP trunks from your carrier into the Voice Gateway | Twilio, Telnyx, Plivo, Vonage, Exotel, Vobiz, Cloudonix, Asterisk ARI and WhatsApp |
| Cost shape | By quote, per billable conversation, plus Voice Gateway line packages | Self-hosted, no per-minute platform fee. Dograh Cloud, 1¢/min platform fee |
| Data out | Encrypted snapshots that open only in Cognigy, conversation data through OData | CSV reports, agent JSON and your own database |
Who builds the agent and where it runs
On both products your own team builds the agent, and the tools differ more than the idea. Cognigy builders work in AI Agent Studio, and the next generation of NiCE Cognigy, launched on 12 November 2026, added the Agentic Experience Studio, agents built from reusable skills and AI Assist. Its Simulator runs test conversations written by an LLM, and an MCP server and a plugin let coding tools such as Claude Code work on agents.
Cognigy runs those agents for you. New customers get Cognigy SaaS, usually in a shared cloud environment, or Cognigy inside NiCE CXone. Cognigy no longer offers on-premises installs to new customers, while existing installs keep receiving updates. Our Cognigy review covers the rest of the product.
In Dograh you describe the agent in plain words in the Agent Builder, approve the "Your agent requirements" card and let it build, then edit on the canvas when you want to. A self-hosted install runs the Agent Builder with a Dograh Service Key, and the canvas works without one. An MCP server lets an AI assistant create and edit agents, and a QA step scores each step of test and live calls.
Because the code is open, your team can run the platform and change it too. You can self-host with Docker Compose on one server with 8 GB of RAM and 4 vCPUs, use the Helm chart, run on Dograh Cloud, or take on-prem and private-cloud deployment on the Enterprise plan.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
Voice and chat: how far each one reaches
Cognigy is built to meet a customer on whichever channel they pick. One agent can answer on voice, webchat, WhatsApp and SMS, in more than 100 languages, and Live Chat and Agent Copilot help the human agents who take over. Over 1,000 brands used Cognigy when NiCE agreed to buy it, Lufthansa and Mercedes-Benz among them.
Dograh is voice-first. Agents answer phone calls and browser calls, and the website widget has a Chat type where visitors type and the agent replies as text. WhatsApp support covers inbound voice calls through the WhatsApp Business Calling API. When an agent needs to send a text message, an HTTP API tool calls your own SMS provider.
Models and speech-to-speech on each side
Cognigy works with LLMs from OpenAI, Azure, Anthropic, Amazon Bedrock, Google and Mistral AI on your own credentials. Since 12 November 2026, NiCE says default models are included in the conversation price. Cognigy limits those included models to installs connected to NiCE CXone that carry a new SKU. Speech comes from providers you add with your own keys, such as Microsoft Azure, Google, Deepgram, ElevenLabs, Speechmatics and Nuance.
Speech-to-speech is where the two pages part. NiCE's launch announced native speech-to-speech models, while Cognigy's speech-to-speech docs list one model, OpenAI GPT Live, and mark the feature experimental and not for production. Ask which applies to your contract.
Dograh is the orchestration layer that runs the call flow, tools and telephony, and you choose the models in each agent. That can be Dograh's own models, your own provider keys, or open-weight models behind an OpenAI-compatible server on your own machines. Seven speech-to-speech options are built in, among them OpenAI Realtime, AWS Nova Sonic, Ultravox and Gemini Live.
Your own keys move the model contract to you, and the call audio still reaches each provider. The audio stays inside your network only when the speech and language models run on your own servers as well.
Phone lines, call data and leaving
Cognigy connects to SIP trunks from your own carrier, and the carrier has to allow Cognigy's voice IP addresses. G.711 is the preferred codec, and transfers to people go out by SIP REFER or SIP INVITE. Voice Gateway capacity is sold in packages of concurrent lines, and use above the package on a given day counts as overage.
Dograh connects to your own carrier, from Twilio, Telnyx, Plivo and Vonage to Exotel, Vobiz, Cloudonix and Asterisk ARI for an existing phone system. On Twilio, Telnyx, Plivo, Cloudonix and Asterisk ARI, Dograh calls the destination first and connects the caller only when someone answers.
Call data sits where each product runs. Cognigy runs SaaS clusters in Europe, the US, the UK and Australia, and NiCE CXone clusters add places such as Tokyo, Mumbai and Cape Town. A self-hosted Dograh install keeps recordings and transcripts in your own database and storage.
The exit differs too. Cognigy snapshots are encrypted and work only in Cognigy, and they leave out logs and analytics, so conversation data comes out through OData, read only. Dograh exports a CSV report of completed calls and each agent as JSON. Our post on the open-source alternative to Cognigy walks through a rebuild. Our page on migrating from Cognigy to Dograh sets out the move in order, from pulling data through OData to moving one phone line at a time.

How each one prices a month
Cognigy publishes no rates, so every deal is quoted. The unit is the billable conversation, up to 50 user inputs within 24 hours. On NiCE CXone, a voice call counts as one conversation for each 10 minutes, so a 12-minute call counts as two. Our guide to Cognigy pricing covers the units and the questions for a quote.
On the open-source side, the hosting choice sets the bill. Self-hosted Dograh has no per-minute platform fee, but servers, model keys and phone minutes still cost money. Dograh Cloud charges a 1¢-a-minute platform fee and includes 10 concurrent calls, with more on request. Self-hosting also means your team runs the server, the database and the upgrades, with community support on open source and custom SLAs on Enterprise.
Where Cognigy wins
Cognigy is the stronger pick for a large contact center that wants one AI layer across every channel. It covers WhatsApp, SMS and webchat out of the box, and Live Chat and Agent Copilot help the human agents who take over. It runs natively inside NiCE CXone and is built for Amazon Connect, Avaya and Genesys, so a team on one of those keeps its routing and reporting.
It also takes the infrastructure off your hands. NiCE runs it in many regions without your team running a server, and Cognigy says it handles more than 25,000 concurrent interactions with AI agents. Large enterprises with complex compliance needs get one vendor and one contract, plus a Simulator for testing at scale. If that is your shape, Cognigy fits better.
When Dograh is the stronger fit
Dograh fits when voice is the main channel and you want to own the platform. That means reading and changing the code, running it on your own servers as a new customer, or keeping call audio inside your network with local models.
It also fits a team that wants to choose and swap each model, pick its own carrier and take call data out in open formats. A team already on a contact center can put Dograh beside it through SIP or Asterisk ARI and transfer calls to a queue number.
To test the fit, self-host one agent on one server or start on Dograh Cloud, and run one real call flow next to your current setup. If you want more options first, our roundup of Cognigy alternatives covers the field.
Glossary
- Billable conversation
- How Cognigy counts usage for billing, with up to 50 user inputs inside 24 hours. On NiCE CXone, each 10 minutes of a voice call counts as one conversation.
- Voice Gateway
- Cognigy's telephony layer, which connects SIP trunks from your carrier to its agents. Capacity is sold in packages of concurrent lines.
- Speech-to-speech model
- A model that takes the caller's audio and answers in audio in one step, without separate speech-to-text and text-to-speech stages. It can cut response time on a call.
- Open-weight model
- An AI model whose weights you can download and run on your own servers. Running one is the only way to keep call audio from reaching an outside model provider.

