Retell AI
Last reviewed
May 6, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 4,065 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 6, 2026
Sources
15 citations
Review status
Source-backed
Revision
v1 ยท 4,065 words
Add missing citations, update stale details, or suggest a clearer explanation.
Retell AI is a voice AI agent platform that enables businesses to build, deploy, and manage AI-powered phone agents for inbound and outbound call automation. Founded in 2023 and launched publicly in February 2024, the company is headquartered in Redwood City, California, and completed Y Combinator's Winter 2024 batch. As of early 2026, the platform processes more than 50 million real-time AI phone calls monthly and generates over $50 million in annualized recurring revenue, making it one of the fastest-growing companies in the voice AI infrastructure market.
The platform is certified under HIPAA, SOC 2 Type II, and GDPR, and it serves customers in healthcare, insurance, logistics, real estate, and enterprise contact centers. Retell AI competes primarily with Vapi, Bland AI, and Hume AI in the market for developer-oriented voice agent infrastructure.
Retell AI was founded in 2023 by five co-founders who met through a combination of prior professional connections and the entrepreneurial community: Bing Wu (CEO), Zexia Zhang (CTO), Todd Li (President), Weijia Yu (COO), and Evie Wang (CMO).
The founding team brought experience from major technology companies. Bing Wu had previously worked at ByteDance/TikTok as a product manager shipping products to hundreds of millions of users, and had previously founded two investor-backed startups with six-figure revenues. Zexia Zhang, who serves as CTO, worked at Google on speech translation and natural language processing for call analysis. Todd Li is a repeat founder and former Google Ads engineer. Weijia Yu spent time at Meta as a product manager and tech lead, and previously shipped full-stack AI systems including text-to-speech and computer vision at ObEN, a Series B startup. Evie Wang, who leads marketing, came from ByteDance where she worked as a product designer and led onboarding initiatives that reduced user drop-off by 25 percent.
The team applied to Y Combinator and was accepted in November 2023. By their own account, they were initially rejected before refining their product demo sufficiently for acceptance into the Winter 2024 batch.
Retell AI launched publicly in February 2024 via a post on Hacker News titled "Retell AI (YC W24) - Conversational Speech API for Your LLM," which attracted significant developer attention. The Hacker News thread drew hundreds of comments and pointed the company toward early adopters in the developer community.
Within months of launch, the company had signed contracts with leading business process outsourcing (BPO) firms and reached $3 million in annualized recurring revenue. By the time the seed round closed in August 2024, Retell reported serving hundreds of customers including the telehealth company Ro.
Retell AI grew its monthly recurring revenue by 3x in the second half of 2024, and by mid-2025 had reached $7.2 million in annual revenue with a team of 41 employees. The company achieved $35 million ARR by the time it launched its automated QA product in December 2025, crossing $40 million ARR by January 2026, and $50 million ARR by April 2026 when it was named to the Wing VC Enterprise Tech 30 list. The company has remained profitable throughout its growth.
Retell AI has raised $5.1 million in total funding across two rounds.
The company's first round was a pre-seed investment from Y Combinator as part of the Winter 2024 batch, which closed in early 2024.
In August 2024, Retell AI closed a $4.6 million seed round led by Alt Capital, with participation from Y Combinator and Carya Venture. The round also included more than 20 individual founders and technology executives as angel investors, including Rajat Suri (co-founder of Lyft and Presto), Aaron Levie (co-founder and CEO of Box), Siqi Chen (founder and CEO of Runway Financial), Alex Levin (co-founder and CEO of Regal), Michael Seibel (co-founder of Twitch and YC group partner), and Tyler Bosmeny (co-founder of Clever and YC group partner).
The relatively modest funding relative to the company's revenue growth reflects the founders' stated preference for capital efficiency. By early 2026, Retell AI was generating over $50 million in ARR with a team of approximately 25 to 40 people, operating on a lean budget compared to other voice AI startups.
Retell AI's core product is a voice orchestration engine that sits between telephony infrastructure and large language models. The platform handles the coordination of several components that must work in tight sequence for a voice conversation to feel natural: speech-to-text transcription, LLM inference, text-to-speech synthesis, turn-taking detection, and interruption handling.
The platform accepts audio from a phone call, transcribes it in real time, routes the transcription to an LLM, generates a response, synthesizes that response to audio, and plays it back to the caller. The full round-trip latency averages approximately 600 milliseconds, which Retell claims is the lowest among competing platforms based on independent benchmarks. This figure covers the time from when a caller finishes speaking to when the agent begins its response.
A custom turn-taking model detects when a caller has finished speaking versus when they are merely pausing mid-sentence. The model uses voice activity detection (VAD) and backchanneling cues to avoid cutting off callers prematurely. Retell updated this turn-taking model in late 2025, reducing additional latency by 150 milliseconds.
Retell AI supports four categories of agent architecture:
| Agent type | Description | Best for |
|---|---|---|
| Single prompt agent | One system prompt drives the entire conversation | Short, simple interactions like FAQ answering |
| Multi-prompt agent | Multiple prompts chained together based on conversation stage | Structured flows like intake forms |
| Conversation flow agent | Visual node-based workflow builder | Complex branching logic with precise control |
| Custom LLM agent | Developer-supplied LLM endpoint | Teams with proprietary fine-tuned models |
The conversation flow agent type uses a no-code visual builder similar to a flowchart. Businesses can drag nodes representing conversation states (greeting, data collection, transfer decision) and connect them with conditional logic. For engineering teams that want full control, the custom LLM agent type accepts any LLM via a standardized WebSocket API, meaning teams can plug in a fine-tuned model trained on their own call transcripts.
The platform natively supports several major telephony providers for both inbound and outbound calling:
| Provider | Type |
|---|---|
| Twilio | CPaaS (cloud communications) |
| Vonage | CPaaS |
| Telnyx | CPaaS |
| Amazon Connect | Contact center platform |
| Avaya | Enterprise contact center |
| Genesys | Enterprise contact center |
| Five9 | Cloud contact center |
Retell also supports native SIP trunking, which lets enterprises route calls through their own telephony infrastructure without going through a third-party CPaaS provider. This is particularly relevant for large enterprises with existing carrier contracts or contact center deployments.
The platform integrates with CRM and automation tools through webhooks and native connectors, including HubSpot, Salesforce, Go High Level, Cal.com (for appointment booking), Zapier, Make, and n8n. These integrations allow agents to read and write CRM records mid-call, book appointments directly into calendars, and trigger downstream workflows when a call ends.
Retell supports a streaming retrieval-augmented generation (RAG) knowledge base that the agent can query in real time during a conversation. The knowledge base auto-syncs from specified URLs or document uploads. Adding the knowledge base costs an additional $0.005 per minute on top of the base platform fee.
Retell AI holds HIPAA, SOC 2 Type I, SOC 2 Type II, and GDPR certifications. The compliance infrastructure is hosted on AWS, which provides a GDPR-compliant Data Processing Addendum (DPA) in its service terms.
For HIPAA compliance, the platform supports Business Associate Agreements (BAAs) through a self-service portal at click-agreements.retellai.com. This means healthcare customers do not need to go through a sales process to execute a BAA; they can sign one directly from the portal. All voice recordings and call transcripts are encrypted at rest and in transit. The platform also supports customer-managed encryption keys for customers who require additional data control.
Access controls include role-based access control (RBAC) for enterprise customers, granular permission settings, audit logs, and real-time monitoring. PII (personally identifiable information) redaction is available as an add-on for $0.01 per minute, which strips sensitive data from transcripts before storage.
The company's compliance documentation and certificates are publicly accessible through a Trust Center hosted on the Vanta platform.
One notable gap: Retell AI does not currently hold ISO 27001 certification, which some enterprise procurement teams require. On-premises deployment is also not available, which can be a blocker for customers with strict data residency requirements. The company's GDPR compliance relies on AWS infrastructure rather than EU-resident servers, which has raised questions for some European customers.
Retell AI supports multiple language models from major providers. The following table shows the LLM options available and their per-minute pricing as listed on the platform's pricing page (all costs are in addition to the base infrastructure fee of $0.055 per minute):
| LLM | Provider | Cost/minute |
|---|---|---|
| GPT 4.1 | OpenAI | $0.045 |
| GPT-4o | OpenAI | varies |
| Claude 4.5 Sonnet | Anthropic | $0.080 |
| Gemini 2.5 Flash | $0.035 | |
| Custom LLM | Any (via API) | Variable |
For teams that want to bring their own fine-tuned or proprietary model, Retell provides a custom LLM integration path via WebSocket. This is particularly useful for healthcare or legal companies that have trained models on domain-specific data.
Retell supports six or more text-to-speech providers, giving developers the ability to select voices based on quality, latency, and cost:
| Provider | Cost/minute | Notes |
|---|---|---|
| Retell Platform (native) | $0.015 | Default; optimized for low latency |
| Cartesia | $0.015 | Ultra-smooth synthesis; can be set as default or fallback |
| ElevenLabs | $0.040 | High expressiveness; ElevenLabs v3 supported |
| OpenAI TTS | $0.015 | OpenAI-native voices |
| MiniMax | $0.015 | Multilingual capability |
| Fish Audio | $0.015 | Community voice library |
Retell has a formal partnership with Cartesia, which allows agents to use Cartesia voices as a primary voice or as an automatic failover if the primary provider goes down. ElevenLabs integration supports the latest v3 model, which delivers stronger emotional expression and more varied cadence compared to earlier versions.
Retell handles transcription internally as part of its voice orchestration layer. The platform's documentation does not publicly specify which underlying STT provider is used, as this is abstracted away from the developer.
The platform supports 50 or more languages with native speech patterns. The system can detect a caller's preferred language and adapt automatically, which is particularly useful for contact centers serving multilingual customer bases.
Retell AI uses a consumption-based pricing model with no platform fees or feature gating. Every plan gets full platform access.
The pay-as-you-go tier starts with $10 in free credits. Pricing is modular, with costs depending on which LLM, voice provider, and add-ons are selected. The base infrastructure fee is $0.055 per minute, and the total cost per minute is the sum of infrastructure plus the chosen LLM and TTS provider fees.
Typical all-in costs for a standard configuration (Retell infrastructure + GPT 4.1 + Retell native voice) come to approximately $0.115 per minute. Using ElevenLabs TTS instead adds $0.025 per minute, bringing the total to around $0.14 per minute.
Concurrency on the pay-as-you-go plan includes 20 concurrent calls at no additional cost. Additional concurrency is $8 per concurrent call per month.
Enterprise pricing is custom and requires contacting sales. Published details include:
The enterprise tier is aimed at organizations spending over $3,000 per month on the platform. Reported entry-level enterprise contract sizes start around $8,000 per month.
| Add-on | Cost |
|---|---|
| Knowledge base (RAG) | +$0.005/min |
| PII removal | +$0.01/min |
| Safety guardrails | +$0.005/min |
| Phone number (standard) | $2.00/month |
| Phone number (toll-free) | $5.00/month |
| Additional concurrency | $8.00/concurrent call/month |
In December 2025, Retell AI launched Retell Assure, described as the first automated quality assurance (QA) solution purpose-built for voice AI. The product became generally available on January 1, 2026.
Traditional call center QA relies on human reviewers sampling 1 to 2 percent of calls. Retell Assure monitors 100 percent of calls automatically using a multi-model pipeline that analyzes each call against customizable criteria including latency, interruptions, hallucinations, sentiment, and adherence to the agent's intended script. When the system detects a failure or a drift from expected behavior, it assigns a score, attaches a reason, flags the issue, and in some cases automatically adjusts the voice AI model's behavior without requiring human intervention.
The product was developed in direct response to customer requests. Bing Wu stated at launch: "Automated quality assurance was the #1 request of our enterprise customers."
Retell AI operates in a crowded market for voice agent infrastructure. The main competitors are Vapi, Bland AI, and Hume AI, each with a distinct approach.
| Feature | Retell AI | Vapi | Bland AI | Hume EVI |
|---|---|---|---|---|
| Primary audience | Developers and enterprises | Developers (API-first) | High-volume outbound | Emotionally intelligent apps |
| Latency | ~600ms | ~500-600ms | ~800ms | ~400-600ms |
| HIPAA compliance | Yes (self-service BAA) | Yes | Yes | Yes |
| SOC 2 Type II | Yes | Yes | Yes | Yes |
| Pricing (base) | $0.055/min infra + LLM + TTS | $0.05/min + $0.066 platform fee | $0.11/min (scale) | API-based, varies |
| Custom LLM support | Yes | Yes | Limited | No (proprietary model) |
| Warm call transfer | Yes | No | Limited | No |
| Automated QA | Yes (Retell Assure) | No | No | No |
| Concurrent calls (free) | 20 | Varies | Varies | Varies |
| No-code builder | Yes (Conversation Flow) | Limited | Limited | No |
| Emotional prosody detection | No | No | No | Yes (core feature) |
| ISO 27001 | No | No | No | No |
| On-premises deployment | No | No | No | No |
Vapi is the closest architectural analog to Retell. Both are middleware platforms that sit between telephony and LLMs. Vapi has historically been the preferred choice for hardcore engineering teams because of its granular pipeline control and extensive webhooks, but it has a steeper learning curve and a less transparent pricing structure. Retell's pricing is simpler: there is no fixed platform fee, and most features are included at no extra cost. Vapi charges a $0.066 per minute platform fee on top of provider costs, which can push real production costs substantially higher than Retell at scale.
Retell also offers warm call transfers and branded caller ID, which Vapi does not provide at the same level. Independent benchmarks have found Vapi's uptime at 99.94 percent versus Retell's claimed 99.99 percent.
Bland AI positions itself as a high-throughput infrastructure platform optimized for volume. It claims to build at the infrastructure level, managing transcription, inference, and synthesis natively rather than routing through external APIs. This gives Bland more control over end-to-end behavior but at the cost of flexibility: developers cannot plug in arbitrary LLMs or voice providers.
Bland's pricing is higher at scale. For 10,000 calls averaging 4 minutes each, Retell costs approximately $2,800 per month while Bland's scale plan runs around $4,900 per month. Bland's voice quality is generally considered lower than ElevenLabs or Cartesia integrations available on Retell, with less emotional resonance.
Hume AI's Empathic Voice Interface (EVI) takes a fundamentally different approach. Where Retell and Vapi are horizontal infrastructure platforms, Hume EVI is a vertical solution that tightly couples its speech model, emotion detection, and synthesis pipeline. The core differentiator is emotional prosody measurement: EVI detects and generates emotional cues in voice, allowing the agent to modulate its tone based on how the caller sounds.
This makes Hume EVI well-suited for mental health applications, coaching tools, and any scenario where the agent needs to respond to a caller's emotional state. Retell, by contrast, does not natively measure or generate emotional prosody, though it can integrate Hume voice models as a TTS provider. For contact center automation where throughput and compliance matter more than emotional intelligence, Retell is the more common choice.
Healthcare has been one of Retell AI's most active verticals. The platform's HIPAA compliance and self-service BAA portal lower the barrier to deployment for clinics, hospitals, and health tech companies. Common applications include appointment scheduling and reminders, insurance verification, post-discharge follow-up calls, and after-hours patient intake.
Pine Park Health, a primary care provider, deployed Retell to automate scheduling and reported a 38 percent increase in its scheduling net promoter score (NPS). The platform enabled the company to handle higher call volumes without proportional increases in staffing.
Medical Data Systems (MDS), a healthcare accounts receivable company, uses Retell to handle 100 percent of its inbound calls. Under the previous system, the company transferred 30 percent of calls to human agents; that rate held steady after the AI deployment, but MDS eliminated the inbound staffing overhead. The company reports collecting approximately $280,000 per month using the AI agents.
A major U.S. insurer deployed Retell to automate 75 to 80 percent of its support calls, added 24/7 multilingual inbound support, and reduced call abandonment rates from 20 percent to 5 percent.
Contact center operators were among Retell's earliest and largest customers. The platform's ability to run unlimited concurrent calls without incremental staffing costs makes it attractive to business process outsourcing firms that handle high call volumes. Retell's batch calling feature enables outbound campaigns at scale, dialing thousands of numbers per day for lead qualification, appointment confirmations, or collection outreach.
Asbury Auto Group, one of the largest U.S. auto dealership chains, uses Retell for customer-facing call automation. StorageVault, a self-storage company, uses the platform for customer support. Anker, the consumer electronics accessories brand, has deployed Retell for support call handling.
Medical debt and consumer finance companies use Retell for outbound collection campaigns and inbound payment processing. Medical Data Systems' $280,000 monthly collection figure illustrates the revenue impact these deployments can achieve. The platform supports DTMF tone recognition (touch-tone input), which is important for self-service payment flows that require callers to enter account numbers or confirmation codes.
Home services businesses (HVAC, plumbing, electrical) use Retell for after-hours call handling, dispatch, and appointment booking. Real estate companies deploy agents for inbound lead qualification, where the agent captures buyer intent and schedules follow-ups with human agents.
Insurance companies use the platform for first notice of loss (FNOL) automation, policy inquiries, and renewal outreach. The multilingual capability is particularly relevant for insurers serving diverse geographic markets.
As of early 2026, Retell AI serves thousands of businesses. Named customers include Anker, Asbury Auto Group, StorageVault, Pine Park Health, Medical Data Systems, and Ro (a telehealth company). The company reports that nearly all pilots convert to production deployments, a claim supported by Bing Wu's statement that customers "quickly see the quantifiable impact... lower costs, faster resolutions, and higher customer satisfaction ratings."
In April 2026, Retell AI was named to the Wing VC Enterprise Tech 30 (ET30) 2026 list, which recognizes 30 enterprise technology companies. Retell was one of three voice AI companies included, and was placed in the early-stage category. The recognition was notable because Wing VC's ET30 list has historically preceded significant enterprise adoption waves for the companies it selects.
The Hacker News launch post for Retell AI in February 2024 was received positively by the developer community, with discussions centering on the latency claims and the flexibility of the custom LLM integration. The OpenAI customer spotlight published in 2024 highlighted Retell's use of GPT-4o for voice agent automation.
In third-party reviews and comparisons, Retell is frequently cited as the best option for teams that need enterprise compliance combined with custom LLM flexibility. Synthflow, a competitor, published a review giving the platform 6.7 out of 10, noting its strength as developer infrastructure while flagging limited GUI tooling for non-technical users as a weakness.
Several limitations have been noted by reviewers and customers:
Customer support: Non-enterprise users rely primarily on a community Discord for support, rather than dedicated ticketing. This is a common complaint in third-party reviews, with some users describing the support as "non-existent" for urgent issues. Enterprise customers get a dedicated Slack channel and 24/7 support.
Billing complexity: The modular pricing model, while transparent in per-component rates, can produce confusing invoices for teams running multiple configurations. Users have reported difficulty predicting monthly costs before signing enterprise contracts. Unexpected charges related to concurrency overages and per-feature add-ons have been flagged on review platforms.
Missing certifications: Retell AI does not hold ISO 27001 certification. This is a requirement for some European enterprise procurement processes. On-premises deployment is also unavailable, which blocks some regulated industries from deploying the platform entirely.
GDPR gaps: The platform's GDPR compliance depends on AWS infrastructure and does not currently include EU-resident servers. Some European customers and legal teams have raised unresolved questions about cross-border data transfer.
No voice cloning: The platform does not offer native voice cloning, which some competitors provide. Teams that want to create a custom branded voice must use an external provider like ElevenLabs or Cartesia's voice cloning tools separately.
Hallucinations: Like all LLM-based systems, Retell agents can produce incorrect or fabricated responses. The Retell Assure QA product monitors for hallucinations across all calls and can flag issues automatically, but cannot prevent them at inference time. Customers in regulated industries (healthcare, insurance, legal) need to implement additional safeguards in their agent prompts and knowledge bases.
Latency variability: Retell's average ~600ms latency claim is based on independent benchmarks, but real-world performance can vary based on LLM provider response time, network conditions, and call complexity. Some reviewers have observed latency spikes during peak usage that disrupt conversation flow.