The main Retell AI alternatives fall into two groups. Vapi, Bland, Synthflow and ElevenLabs Agents are hosted, closed-source platforms that bill per minute or by contract. Dograh, LiveKit Agents and Pipecat are open source and run on your own servers. Dograh ships as a full platform with a visual builder; the other two are frameworks you build on.
Key Takeaways
- Retell's $0.055 voice infra fee comes before the model, voice and phone line.
- Retell sells on-prem through Enterprise sales; anyone can self-host Dograh for free.
- Move phone numbers one at a time, with Retell kept as the fallback.
This guide compares seven Retell AI alternatives on five questions: can you read the code, can you run it on your own servers, what a minute costs at list price, which AI models you can use, and how you build the call flow. Every price comes from the vendor's own pricing page.
Why teams look for a Retell AI alternative
Teams usually look for an alternative to Retell AI when the per-minute bill gets large at volume, or when the platform has to run somewhere Retell's standard cloud does not.
Retell is a hosted platform: you build an agent from a single prompt or on a drag-and-drop Conversation Flow canvas, and Retell runs the calls on its own cloud. Its compliance page lists SOC 2 Type 1 and Type 2, and Retell signs a business associate agreement (BAA) for HIPAA at no extra fee.
Retell's pricing page puts a call at $0.07 to $0.31 a minute, made of four parts. A fixed $0.055 voice infra fee comes first; Retell lists no separate speech-to-text charge, so it appears to cover that too. The AI model you pick adds $0.0016 to $0.32 a minute, the voice adds $0.015 for most voices (up to $0.10 for ElevenLabs v3), and a Retell phone number adds $0.015.
At 1,000 minutes a day, the voice infra fee alone comes to about $20,075 a year ($0.055 x 365,000 minutes), before the model, voice and phone line. Part of it would be a speech-to-text cost on another platform, and it is the one line that stays the same whichever models you pick.
Paid add-ons sit on top. Retell's AI quality assurance (QA) costs $0.10 a minute after the first 100 free minutes and scores a sample of your calls. Removing personal details costs $0.01 a minute, and noise removal and safety guardrails add $0.005 a minute each. Twenty concurrent calls are included, and each extra one costs $8 a month.
Retell's code is closed. For teams that need it inside their own network, Retell sells an Enterprise on-prem option through its sales team, with no published price, requirements or install package. Retell also states that it does not currently operate services within the European Union.
Put simply, the voice infra fee is the line you can remove. On an open-source platform you host yourself, such as Dograh, that fee is gone: you pay the model, voice and phone providers directly, plus your server. At a few hundred minutes a month the difference is small. At 1,000 minutes a day it is the roughly $20,000 a year worked out above, which is when a cheaper alternative to Retell AI starts to pay for the move. Where your call data has to live matters as much as the price, and the table below shows which platforms can run where. Our breakdown of Retell AI pricing works through the full bill for three example businesses.
How to choose a Retell AI alternative
Five things decide whether a Retell AI alternative will work for your team. The licence decides whether you can read the code, change it and keep running it if the vendor changes course. Where it runs decides whether that can be your own servers, or on-prem inside your network, without a sales contract. Billing uses each vendor's regular list price, never averaged into one "typical" per-minute rate, because the total depends on the models you pick. Model choice covers whether you can bring your own key (BYOK) or host open-weight models yourself. The builder matters because teams leaving Retell are used to a visual flow canvas.
Latency is left out of the table, since it depends on your models and phone path far more than on the platform, so test it on your own calls. The list covers platforms you can run a Retell-style phone agent on, which leaves out phone carriers such as Telnyx and Twilio and contact-centre suites such as PolyAI, Parloa and Cognigy. Each product below is available today.
Retell AI alternatives compared side by side
The table puts the best-known Retell AI alternatives next to Retell on the same questions.
| Product | Open source | Self-host or on-prem | List price | What you get, and what's missing |
|---|---|---|---|---|
| Retell AI (baseline) | No | No self-host package. Runs on Retell's cloud, with no services operated in the EU. On-prem offered through Enterprise sales, with no published price or install details | $0.055/min voice infra, plus the AI model per minute ($0.0016 to $0.32), the voice, and $0.015/min for a Retell number. 20 concurrent calls included, then $8/mo each | Hosted platform with prompt agents, a drag-and-drop Conversation Flow builder, SOC 2 Type 1 and Type 2, and a BAA at no extra fee. But it is closed source and bills the AI model itself per minute. QA, personal-detail removal, noise removal and safety guardrails cost extra per minute |
| Dograh | Yes, BSD-2-Clause | Yes. Self-host the full platform for free on your own servers with Docker Compose. On-prem or private cloud on Enterprise. Or use managed Dograh Cloud | Self-hosted: no per-minute platform fee. Dograh Cloud: per-minute platform fee (see below). Models and phone extra | Full platform with a visual workflow builder, telephony on your own carrier, call traces, and QA that scores each step of every call. Use Dograh's models, your own keys, or open-weight models you host. But the free version has community support rather than a vendor service level agreement (SLA), and call transfer to a person works on Twilio only |
| Vapi | No | No. Runs only on Vapi's cloud, in the US or its EU region. Vapi does not support on-premise deployments | $0.05/min hosting, plus models at the provider's price and phone. Core $29/mo, Pro $999/mo minimum, HIPAA add-on $2,000/mo | Hosted developer platform with Assistants and Squads, set up in code or the dashboard. But it is closed source, and its visual flow builder, Workflows, was retired in August 2026 |
| Bland | No | Only on Enterprise: on-prem, air-gapped, or in your own virtual private cloud (VPC), by custom contract. Start and Build run on Bland's cloud | Start $0.14/min, no platform fee. Build $0.12/min plus $299/mo. Phone billed separately. Enterprise custom | Hosted platform with one per-minute rate that includes the AI models, and a visual Pathways builder. But it is closed source, it says it runs only its own models, and on-prem and the BAA need an Enterprise contract |
| Synthflow | No | No self-host or on-prem option on its pricing page. Runs on Synthflow's cloud | Enterprise only, from $30,000 a year | Hosted platform with a visual Flow Designer and launch support. But there is no self-serve plan, and contracts start at $30,000 a year |
| ElevenLabs Agents | No | No self-host package. VPC deployment of its models through sales; on-prem announced in early access. No published price for either | Free plan with 15 min/month. Paid plans $6 to $990/month with included minutes; extra minutes $0.08. AI model billed on top. No telephony fee; your carrier bills you | Hosted agent platform from a text-to-speech company, with a visual workflow builder and a free plan. But it is closed source, and the AI model is billed on top of the per-minute price |
| LiveKit Agents | Yes, Apache-2.0 (its turn-detector model has a separate LiveKit licence) | Yes. Self-host the framework and the LiveKit media server on your own servers or on-prem. You build and run the rest of the product. Managed option: LiveKit Cloud | Free to self-host. LiveKit Cloud: free plan with 1,000 min/month; $50/month for 5,000 min or $500/month for 50,000 min, then $0.01/min | Open-source framework with telephony and any model you wire in. But its Agent Builder is a single prompt with no workflows, and call insights with recordings exist only in LiveKit Cloud |
| Pipecat | Yes, BSD-2-Clause | Yes. Self-host the framework on your own servers or on-prem. You build and run the phone wiring, call history and live-call QA. Managed option: Pipecat Cloud | Free to self-host. Pipecat Cloud from $0.01/min per active agent | Open-source Python framework with a visual flow editor and any model you wire in. But its evals are for development, so there is no built-in QA for live calls, and call transfer does not work on the Twilio WebSocket route |
The table falls into two groups. Five hosted products, Retell included, bill per minute or by contract for as long as you use them, and none gives you code you can run. Where on-prem or VPC is offered, it sits behind a sales contract with no published price. Of the three you can run yourself, LiveKit Agents and Pipecat are frameworks, and Dograh is a full platform.
The prices use different units, so convert each one to your own monthly minutes before you compare.
Dograh: the open-source Retell AI alternative you can self-host
Dograh is an open-source, self-hostable voice agent platform under the BSD-2-Clause licence. Dograh on GitHub has more than 5,800 stars, and the whole platform sits in that repository: the visual workflow builder, telephony, call transfer, call history, call traces and automatic QA.
If your team built agents on Retell's Conversation Flow canvas, the builder will feel familiar: you draw the call as steps joined by pathways, each with a written condition the AI uses to pick its route. That makes Dograh a Retell AI alternative with a visual builder that you can also run on your own servers.
Dograh runs on the phone setup you already have: Twilio, Vonage, Plivo, Telnyx, Cloudonix, Asterisk ARI or your own Session Initiation Protocol (SIP) trunk. Dograh sells no numbers, so yours stay with your carrier. Call transfer to a person works on Twilio. After a call, automatic QA reviews the transcript and scores each step of the flow separately, on editor test calls and on live calls. It runs on the AI model you choose, billed as normal model usage with no per-minute QA fee, and you can analyze only a sample of calls to keep costs down.
Self-hosting uses Docker Compose on a server with at least 8 GB of RAM and 4 vCPUs, and needs no GPU when your models run through API keys. Self-hosted Dograh has no per-minute platform fee; you still pay for the server, any paid model keys and the phone line. For an on-prem Retell AI alternative with a contract behind it, Dograh Enterprise offers on-prem or private-cloud deployment, and our post on why on-prem will win enterprise voice AI makes that case. If you would rather not run servers at all, Dograh Cloud is the managed version, at 1¢ a minute.
Dograh is the orchestration layer that runs the call flow, tools and telephony. The speech and language models are a separate layer: Dograh's own, any supported provider on your own API key, or open-weight models you host. Call records stay on your servers once you self-host. The audio stays there too only when you also host the AI models; with API keys it reaches the model provider, and the phone carrier carries the call either way. Our guide to running voice AI on local, self-hosted models covers that stack.
Self-hosted, you control the server and choose when to upgrade. Dograh Enterprise adds support with SLAs if you need it. Our Dograh vs Retell AI comparison goes through every row, from licence to call transfer.
Open Source Alternative to Vapi / Retell
Self-hosted voice agent platform — no per-minute fees
dograh-hq/dograh
Star on GitHub
Hosted alternatives to Retell AI
Four hosted products come up most often as Retell AI competitors, each running on the vendor's cloud like Retell. If your shortlist is Retell AI vs Vapi vs Bland, the first two sections below cover it.
Vapi
Vapi charges a prepaid $0.05 a minute for hosting and bills speech-to-text, the AI model and the voice at each provider's own price on top. Its packages set how many calls run at once: 4 with no package, 10 on Core at $29 a month, and 30 on Pro, which costs 10% of your hosting fee with a $999 monthly minimum. You build agents as Assistants and Squads, in the dashboard or in code. Vapi is closed source and does not support on-premise deployments. Its visual flow builder, Workflows, was retired in August 2026, and HIPAA handling is a $2,000 a month add-on.
Our Vapi vs Retell AI comparison puts the two side by side, and Vapi alternatives covers the same field from the Vapi side.
Bland
Bland sells one per-minute rate that includes the AI models. Its Start plan is $0.14 a minute with no platform fee and 10 concurrent calls, and Build is $0.12 a minute plus $299 a month, with 50. The phone line is billed separately. You build calls on a visual Pathways canvas. Bland is closed source, and it says it runs only its own speech, language and voice models, so you cannot choose a model provider. Its on-prem and VPC options need the custom Enterprise plan, and so does its BAA for HIPAA.
Our Retell AI vs Bland AI comparison goes through the two row by row, including what each one charges for a failed call.
Synthflow and ElevenLabs Agents
Synthflow sells only Enterprise contracts, and they start at $30,000 a year, scoped around call volume, concurrency, telephony, integrations, security needs and launch support. You build calls in its visual Flow Designer. Synthflow is closed source, and its pricing page lists no self-hosted or on-prem option.
ElevenLabs Agents has a free plan with 15 call minutes a month, then paid plans from $6 a month for 75 minutes up to $990 a month for 12,375 minutes. Extra minutes cost $0.08, with no telephony fee, since your own carrier bills you. It has a visual workflow builder and bills the AI model on top. The platform is closed source. ElevenLabs sells VPC deployment of its models and announced an on-prem option in early access in April 2026, both through sales with no published price.
Open-source frameworks: LiveKit Agents and Pipecat
LiveKit Agents and Pipecat are open-source frameworks that give developers solid building blocks for a voice agent. The product around those blocks is yours to build.
Pipecat is a Python framework from Daily under the BSD-2-Clause licence, with more than 16,000 GitHub stars, and Dograh itself is built on a fork of it. In our experience Pipecat is an excellent base, but a production deployment needs a lot of extra plumbing: phone integrations, call history, QA on live calls and the call lifecycle. Pipecat has a visual flow editor, and its evals are built for development. Daily also runs a managed version, Pipecat Cloud, from $0.01 a minute per active agent on its smallest agent size.
LiveKit Agents runs on the LiveKit media server under Apache-2.0, with more than 14,000 stars. If open source is your reason for leaving, note that LiveKit's turn-detector model ships under a separate LiveKit Model License. Its Agent Builder is a single-prompt form with no workflows, and call insights with recordings are only available in LiveKit Cloud. LiveKit Cloud has a free plan with 1,000 agent minutes a month, and paid plans start at $50 a month for 5,000 minutes and $500 for 50,000, then $0.01 a minute.
If open source is your main reason for leaving, our post on the open-source alternative to Retell AI covers what changes when you own the platform, and where the call data really goes.
When staying on Retell makes sense
Retell still fits some teams, and a switch only pays off if it fixes a real problem for you.
If you need SOC 2 and a signed BAA without running any servers, Retell gives you both at no extra fee. Self-hosted Dograh takes the vendor out of the data path, but the security and compliance work then sits with your own team.
If you run fewer than 20 calls at once and want one bill for the model, the voice and the phone line, Retell's pay-as-you-go plan covers that with no monthly fee. The per-minute total only becomes a real cost question at production volume.
If your team works in one vendor-run console, Retell's covers testing with simulated callers and live monitoring, where a person can listen in on a call and take it over. Moving means setting up those habits again on another platform.
How to switch from Retell AI to Dograh
Moving to Dograh can happen one piece at a time, so Retell keeps answering your calls until each piece is ready. Dograh is open source, so your team can read the code and test the self-hosted version on your own server before committing to anything. Each agent you rebuild can be tested over a real phone call first, and each number moves only when its agent passes.
Start with the agent. Retell documents no export button, so read each agent through Retell's API with a read-only key: list the agents, pull each one's prompt and tools from its Retell LLM or Conversation Flow, and keep that file as the spec. Dograh has no Retell importer, so you rebuild the agent in the visual builder, or ask an assistant such as Claude Code to build it through Dograh's Model Context Protocol (MCP) server. A Conversation Flow maps across almost step for step; a single-prompt agent becomes one or two steps.
Some Retell features need a different design. Dograh has no rule-based branches, since the AI picks each pathway. It has no keypad-press action and no built-in SMS. Values the agent pulls out of a call go to webhooks, transfer routing and HTTP tools, and cannot be dropped into a later prompt.
Tool servers mostly carry over. Dograh's HTTP tool body template can wrap the values the way Retell's custom functions send them, so only a server that reads Retell's call details or function name needs a small change.
Phone numbers need the most care. Retell has no documented way to move a number it sold you to another carrier, so plan to replace those numbers, or ask Retell support before counting on keeping one. A Twilio number you imported stays in your Twilio account, but Retell connects it through a SIP trunk, and Twilio ignores a number's own call settings while it sits on a trunk. So the switch happens in Twilio, when the number comes off the trunk, and putting it back on the trunk is your rollback.
Move one number at a time and keep the Retell setup as the fallback until each number works on the new platform. Test every call transfer over a real phone call, because a browser test call cannot transfer. Our guide to testing voice agents before you go live lists the checks in order, and the full step-by-step is in how to migrate from Retell AI to open-source voice AI.
Join the Dograh Community
Dograh is an OSS alternative to Vapi. Join our Slack community for queries, releases, best practices & community interactions.
Picking your Retell AI alternative
The right pick depends on where your calls must run and who will run the platform.
If call data has to stay on your own servers or inside your network, Dograh is an open-source platform you can install today, with on-prem on Enterprise when you need a contract behind it. If you want open source and plan to build most of the product yourself, LiveKit Agents and Pipecat are frameworks to start from. If you want a hosted product and accept per-minute billing, compare Vapi, Bland and ElevenLabs Agents at your own call volume, and expect an annual contract with Synthflow.
For a Retell AI alternative for enterprise, ask every closed vendor for its on-prem price and server requirements in writing before you compare.
Each part of this comparison has its own deep dive:
- Dograh vs Retell AI
- Open-Source Alternative to Retell AI
- Retell AI Pricing Explained
- How to Migrate From Retell AI to Open-Source Voice AI
- Retell AI vs Bland AI
Glossary
- Voice infra fee
- Retell's $0.055 per-minute charge for running the call. Retell lists no separate speech-to-text charge, so it appears to cover that too. The AI model, the voice and the phone line are billed on top of it.
- Concurrent calls
- The number of calls running at the same moment. Hosted plans include a set number and charge a monthly fee for each extra one.
- SIP trunk
- A connection that carries phone calls over the internet between a carrier and a voice platform. While a Twilio number sits on a trunk, the trunk decides where its calls go.
- Bring your own key (BYOK)
- Using your own account with a model provider instead of the platform's. You hold the contract and the bill, but the call audio still goes to that provider.
