Retell AI
Retell AI is a voice agent platform for phone call centers. Teams build, test, deploy and monitor agents that handle inbound and outbound calls across voice, chat, SMS and API, with a claimed 600 ms latency.
What is Retell AI?
Retell AI is a platform for building AI voice agents that take over business phone calls. It bills itself as the #1 AI Voice Agent Platform for Automating PHONE Calls, and the product is organized in three movements: Build, Deploy and Monitor. Build covers the agent itself, through a drag-and-drop agentic framework with built-in guardrails, real-time function calling for appointments, payments, record updates and call transfers, a knowledge base with streaming RAG that automatically resyncs against your website content, and IVR navigation for keypad menus. Deploy covers getting that agent onto the phone: SIP trunking to any carrier, Twilio or Telnyx numbers, branded caller ID and verified numbers built to survive spam filters, and batch campaigns. Monitor covers what happens afterwards, with transcription, post-call analysis and AI quality assurance.
The differentiator the company puts forward is its own voice orchestration layer: a claimed 600 ms latency, voices selected from real performance data, and a proprietary turn-taking model. The stack stays open where it matters. You choose the LLM, from the GPT, Claude and Gemini families or a custom model of your own, and the text-to-speech provider among Retell, MiniMax, Fish, ElevenLabs, Cartesia and OpenAI. 55 languages are documented on the speech synthesis and recognition side. Agents are not confined to voice either: the same platform runs web and in-app chat, SMS and direct API access.
On the security side, Retell AI documents HIPAA, SOC 2 Type 1 and 2 and GDPR, alongside SSO, PII redaction, role-based access control and an on-premise deployment for enterprises with data residency constraints. A 99.5% uptime SLA is claimed, with planned maintenance capped at six hours a month.
The company behind it is Retell AI Inc., based in Redwood City, California: Y Combinator in November 2023, public launch in February 2024, a USD 4.6M seed in May 2024, then USD 40M in annualized revenue announced in November 2025 with a team of 25. It claims 3,000+ companies and 40M+ calls a month. Published customer stories include Pine Park Health (+38% scheduling NPS), SWTCH (support costs more than halved), Medical Data Systems (100% of inbound calls answered, 30% transferred) and Anker (80.4% resolution, NPS 63).
What it does
- Design a call flow by drag and drop, with built-in guardrails
- Trigger real-time functions mid-call: book an appointment, take a payment, update a record, transfer to a human
- Bring an existing number through SIP trunking, or buy a Retell number
- Run batch calling campaigns with no concurrency ceiling
- Display a branded caller ID and verified, spam-resistant numbers
- Analyze 100% of calls through transcription, post-call analysis and automated QA
- Keep a knowledge base in sync with your website content through streaming RAG
When to use Retell AI / When not to
A quick filter to help you decide if Retell AI is the right fit.
When to use Retell AI
- High-volume contact centers drowning in repetitive calls: appointment booking, order status, lead qualification
- Healthcare, financial services, insurance, logistics, retail, travel and hospitality, debt collection and home services teams
- Engineering teams that want to build on an API, SDKs, an MCP server and a CLI, and to SIP trunk onto the telephony they already run
- Regulated buyers who need HIPAA with a BAA, SOC 2 Type 1 and 2, a GDPR DPA with Standard Contractual Clauses, PII redaction, SSO and role-based access control
- Operations leaders running a hybrid model, where the agent absorbs routine volume and hands everything else to a human
When not to use Retell AI
- Organizations that require voice data to be processed inside the European Union: Retell AI states it does not currently operate services within the EU, and personal information is primarily stored and processed in the United States
- Anyone looking for a permanently free tool: beyond the USD 10 of sign-up credits, every minute is billed
- Buyers who need a fixed, predictable monthly price, since the real cost depends on the LLM, the voice and the telephony you combine
- Healthcare teams tempted to push protected health information through the platform before a BAA has been signed, which the terms of service forbid
- Outbound teams with no internal handle on TCPA compliance, consent capture and do-not-call lists, because that burden sits entirely with the customer
How to use Retell AI
A typical end-to-end flow, from setup to results.
- Create a free account on the Retell AI dashboard and collect the USD 10 of sign-up credits
- Build a first phone agent, which the documentation times at 15 minutes, from a prompt or a conversational flow, or let Conductor build and test one through chat
- Configure the agent: functions, guardrails, language, and the LLM and voice it should run on
- Connect your providers from the Integrations page, Available tab, using an API key or an OAuth connection, sometimes a tenant URL or an account identifier; an Admin or Developer role is required
- Let Retell test the credentials at connection time, then check that the provider has moved to the Connected tab
- Attach a number: SIP trunk the telephony you already use, or take a Retell number at USD 2 a month
- Set the per-agent retention, from 1 day to 2 years, and the PII controls before anything goes live
- Test the agent in simulation before pushing it to production
- Decide which tools run before the call, during it or after it, using pre-call and post-call functions, with pre-chat and post-chat equivalents for chat agents
- Publish, then follow performance through transcripts, post-call analysis and automated quality assurance
Pros & Cons
Pros
- Claimed 600 ms latency, presented as validated by independent benchmarks
- No annual contract: you start at USD 0 and pay by the minute
- Open model choice, with the GPT, Claude and Gemini families or a custom LLM, and six text-to-speech providers
- Compliance posture that is serious for a company of this size: HIPAA with a BAA, SOC 2 Type 1 and 2, GDPR with a DPA and SCCs that can be self-signed at no extra cost
- Retention configurable per agent from 1 day to 2 years, with a published subprocessor list
- SIP trunking means keeping your carrier and your existing numbers, with 20 concurrent calls and 10 knowledge bases included from the start
- 55 documented languages, 8 natively documented connectors, a 99.5% SLA, a public status page and a Vanta Trust Center
Cons
- Pricing is hard to forecast: the per-minute rate moves between USD 0.07 and USD 0.31 depending on the LLM, the voice, the telephony and the add-ons chosen
- Many capabilities are billed as add-ons on top: knowledge base +USD 0.005/min, advanced denoising +USD 0.005/min, PII removal +USD 0.01/min, AI quality assurance USD 0.10/min, branded calls +USD 0.10 per outbound call
- No permanent free plan, the USD 10 of credits being a one-off
- The strongest guarantees, from custom MSA, DPA and BAA to custom SSO, a dedicated server, RBAC and 24/7 support, are reserved for the Enterprise plan and quoted on request
- No services are operated within the European Union and data is processed primarily in the United States
- Without an explicit opt-out, communications feed model development and training after de-identification and aggregation, while TCPA compliance, consent and do-not-call lists remain entirely the customer's responsibility
- No mobile app, no features overview page and no dedicated legal or GDPR contact, everything going through support@retellai.com; the Trust Center is JavaScript-rendered, so certification detail is unreadable without a browser
Pricing & Plans
There is no permanent free plan. New accounts receive a one-off USD 10 of credits, after which billing is strictly pay-as-you-go, by the minute consumed, with no annual commitment. Voice agents run from USD 0.07 to USD 0.31 per minute depending on the combination selected, and the pricing simulator on the site shows USD 0.11 per minute as a worked example. Chat agents start at USD 0.002 per AI message. The per-minute rate breaks down into Retell voice infrastructure at USD 0.055, text-to-speech at USD 0.015 and USD 0.040 for ElevenLabs, the LLM from USD 0.003 to USD 0.32 depending on the model, and telephony at USD 0.015, which is free when you bring your own SIP trunk. Real-time speech-to-speech models are priced separately, at USD 0.345 per minute for GPT Realtime and USD 0.07 for GPT Realtime mini. Add-ons are billed on top, and monthly subscriptions apply to Retell phone numbers at USD 2.00, Retell SMS at USD 20.00, additional concurrency at USD 8.00 per concurrent call beyond the 20 included, additional knowledge bases at USD 8.00 beyond the 10 included, and verified numbers at USD 10.00. The Enterprise plan is quoted on request, with volume discounts.
- USD 0.07 to USD 0.31 per voice minute and from USD 0.002 per AI chat message
- covering full platform access
- prebuilt agent templates
- analytics and transcripts
- simulation testing
- webhooks and API
- 20 free concurrent calls
- a shared server
- custom voice and chat pricing on request
- including everything in pay as you go
- plus deployment support as an add-on
- a custom MSA and DPA
- HIPAA BAA
- custom SSO
- role-based access control
- a dedicated stable server
- knowledge base +USD 0.005/min
- batch calling +USD 0.005 per dial
- branded call +USD 0.10 per outbound call
- advanced denoising +USD 0.005/min
- guardrails +USD 0.005/min
- PII removal +USD 0.01/min
- AI quality assurance USD 0.10/min with the first 100 minutes free
- SMS USD 0.01 per message
- Retell phone number USD 2.00
- Retell SMS USD 20.00
- additional concurrency USD 8.00 per concurrent call
- additional knowledge base USD 8.00
- verified number USD 10.00
Data, GDPR & hosting
A consolidated view of how Retell AI handles your data.
GDPR overview
GDPR compliance is claimed explicitly: the documentation states that Retell complies with the GDPR by using Amazon Web Services, which includes a GDPR-compliant Data Processing Addendum. That DPA, carrying the EU Standard Contractual Clauses, is self-signable at no cost on click-agreements.retellai.com. The privacy policy names the Article 6 bases relied on, namely consent, performance of the contract and legitimate interests, and lists the rights available: access, rectification, erasure, objection, restriction, portability and complaint to a supervisory authority. Transfers outside the EEA rely on SCCs, the UK IDTA or another approved mechanism, with a copy on request. Requests go to support@retellai.com or by post to Redwood City; erasure means deleting the account, then emailing. Two blind spots remain: no Article 27 EU representative and no named DPO, and the same page states that no services are operated within the European Union.
Who owns the data?
Section 10 of the terms of service is explicit: you retain ownership of any content you submit, upload or provide to the service, including software, text, audio, video, images and user-generated content. Retell AI and its licensors retain ownership of the underlying AI models and algorithms used to generate the AI output, so the engine never travels with the result. Feedback sent to the company is treated differently and is covered by a perpetual, irrevocable, worldwide, royalty-free licence. And when Retell AI acts as a service provider on a customer's behalf, it is that customer, not Retell AI, who sets how long the data is kept.
Reuse rights
Because you keep ownership of your User Content, you are free to reuse your own transcripts, recordings and call data without asking Retell AI for permission. The reverse direction is where attention is required. Under section 11 of the terms, if you do not opt out of recording, you give Retell AI permission to record calls made through the service and to process communications and User Content for AI-powered analytics and for the development, training and improvement of the machine learning models included in the service; the company states that the data is de-identified and aggregated using commercially reasonable industry-standard technologies before being used for those purposes. The privacy policy lists the same intent in plain words: to train the artificial intelligence models that support the services, and to improve, train and enhance those models and related technologies. Training is therefore the default state rather than an option you switch on, and the recording and transcription opt-out, listed in the pricing comparison table, is what disables it. Two further controls sit alongside it: per-agent PII handling, where you choose to store everything, exclude PII, or keep basic attributes only, and the exclusion of personal identifiers from internal logs and product analytics by default. Be aware that third-party LLMs are called for certain features, such as generating and refining an agent prompt.
Data retention & training
Hosting summary
Retell AI states in its privacy policy that the company is based in the United States and that the personal information it collects is primarily stored and processed in the United States. The compliance documentation names Amazon Web Services as the infrastructure layer and grounds the GDPR claim on the AWS Data Processing Addendum, while stating on the same page that no services are currently operated within the European Union. Transfers from the EEA, the United Kingdom and Switzerland rely on Standard Contractual Clauses, the UK IDTA or another approved mechanism, and a copy can be requested. The telecom carriers used for the calls, Twilio among them, also transfer data internationally under their own mechanisms. Enterprises with strict data residency requirements are offered an on-premise deployment, and the subprocessor list is published at trust.retellai.com/subprocessors. One technical caveat: the domain resolves to 185.158.133.1, a Cloudflare anycast node geolocated in Poland, which is a network indicator rather than a hosting location.
Where Retell AI works
Country-level availability.
Not available in
Things to keep in mind
Risks and trade-offs to weigh before adopting Retell AI.
- The final bill depends on six stacking variables, namely the LLM, the text-to-speech voice, the telephony, the add-ons, the concurrency and the phone numbers: model your estimate before committing
- The recording opt-out is not active by default, and leaving it off means your communications feed the development and training of the models
- A BAA must be signed before any health data transits the platform, since the terms forbid protected health information without one
- TCPA compliance, documented consent and do-not-call lists sit entirely with you, including keeping the evidence for at least five years
- No services are operated within the European Union despite the GDPR compliance claim: check that this is acceptable for sensitive European processing
- Advanced compliance controls such as RBAC, SSO, a custom MSA and DPA and a dedicated server are locked behind the Enterprise plan
- The domain's IP geolocation points to a Cloudflare CDN node in Poland, which is a network artifact and not the country where your data is hosted
Setup & Integrations
Technical difficulty
Low to moderate. The documentation targets a first phone agent in 15 minutes, Conductor builds and tests an agent through chat, and the no-code drag-and-drop framework with prebuilt templates covers the basic path. Effort rises with the connections: SIP trunking, a custom LLM, webhooks, the API, the MCP server and the CLI all assume a technical owner. Connecting an integration needs an API key or OAuth, sometimes a tenant URL or account identifier, and an Admin or Developer role. Enterprise buyers can add engineer-led deployment support. The stated pace is weeks, not months.
Deployment
Integrations
Supported languages
Behind Retell AI
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Retell AI.
Frequently asked questions
What is Retell AI?
How much does Retell AI cost?
Is there a permanent free plan?
Can I keep my existing phone number?
Which languages does Retell AI support?
Does Retell AI provide an API and native integrations?
Is Retell AI HIPAA and GDPR compliant?
Are my calls used to train the AI models?
How long is my data kept, and where is it hosted?
What uptime does Retell AI commit to, and how do I get support?
Should you pick Retell AI?
Retell AI positions itself as production voice infrastructure rather than a demo generator, production first, demos second, and the product largely backs that up. The claimed 600 ms latency, the freedom to pick your own LLM and voice provider, SIP trunking onto whatever telephony you already run, and a compliance file that covers HIPAA with a BAA, SOC 2 Type 1 and 2 and a self-signable GDPR DPA with Standard Contractual Clauses all point at buyers who intend to put agents on real customer calls rather than experiment with them. The traction is consistent with that ambition: 3,000+ companies claimed, 40M+ calls a month, and stated profitability on USD 40M of annualized revenue with a team of 25.
Two reservations deserve weight before signing. The first is cost. Pay-as-you-go sounds simple, yet the per-minute price moves between USD 0.07 and USD 0.31 according to the LLM, voice, telephony and add-ons you combine, and several capabilities most buyers would consider standard, from the knowledge base to denoising, PII removal and quality assurance, are billed separately. The strongest guarantees, including a custom MSA and DPA, RBAC, SSO and a dedicated server, sit behind an Enterprise quote.
The second is jurisdiction. GDPR compliance is claimed on the strength of AWS infrastructure and a signed DPA, yet the same documentation states that no services are currently operated within the European Union, and the privacy policy places storage and processing primarily in the United States, with no Article 27 representative and no named DPO. Transfers are covered by SCCs and an on-premise deployment exists for strict residency needs, but a European buyer handling sensitive data should test that arrangement against its own requirements rather than take the compliance badge at face value.
- Choosing a selection results in a full page refresh.
- Opens in a new window.