DEV Community: Alfredo Romero

Best Voice AI Platforms for Agencies in 2026: Hermes vs Synthflow vs Retell vs Vapi

Alfredo Romero — Thu, 25 Jun 2026 20:30:37 +0000

The four platforms every AI voice agency evaluates right now are Hermes, Synthflow, Retell, and Vapi. They are not equivalent options. They serve different buyers, carry different risks, and have meaningfully different economics once you move past the advertised rate. This guide covers what the comparison actually looks like when you run real agency workloads across all four. We looked at data from 1,200 calls across client deployments, at published pricing, at incident histories, and at the structural question that matters most for agencies: which platform was actually built for you, and which one tolerated you until it found a bigger customer.

The voice AI market is growing fast. The global voice AI agents market is projected to hit $47.5 billion by 2034 , up from $2.4 billion in 2024, at a 34.8% CAGR. The segment is currently at $22.5 billion in 2026 , and venture capital investment in voice AI startups hit $1.2 billion in 2025 alone. The money coming into this space is real. What is also real is that most of it is going to enterprise infrastructure and API providers, not to the agencies doing the actual deployment work. The platform you choose determines how much of that margin you capture versus how much you send upstream.

By builders, for builders. Here is the actual comparison.

What separates these four platforms at the architecture level?

Before comparing pricing or features, it helps to understand what each platform actually is at the structural level. This matters because the architecture determines what breaks when volume grows, what you can and cannot white-label, and who controls your pricing.

Retell and Vapi are API infrastructure providers. They give you a programmable voice layer. That is genuinely useful if you have engineering resources and want to build custom implementations. It is not a complete agency platform. You still need to wire in a CRM, build a white-label client portal, manage client billing separately, and handle campaign orchestration yourself or with additional tools.

Synthflow is a no-code builder with a white-label wrapper. It was built for agencies who want a visual interface and do not want to touch code. That made it genuinely useful for agency operators in 2023 and 2024. In 2025, Synthflow raised a $20 million Series A and repositioned toward enterprise BPOs. The agency pricing tiers are gone for new users. New customers start at pay-as-you-go or custom enterprise at a minimum annual commitment around $30,000. The platform still works for agencies, but it is no longer designed for them.

Hermes is agency infrastructure built from the ground up. That means native multi-tenant workspaces per client, native white-label where your clients never see the word Hermes, a built-in CRM, campaign orchestration, and a single invoice at a flat rate. It was not retrofitted for agencies. It was designed around how agencies operate from day one. From $149/month.

What does each platform actually cost per minute?

The advertised rate and the real rate are not the same for three of the four platforms. Here is what the math looks like at 2,000 minutes per month, which is a reasonable mid-market agency volume.

Vapi advertises $0.05/min for the platform layer. But that number excludes the LLM (add $0.03 to $0.10/min for GPT-4o or Claude-class models), the text-to-speech voice (add $0.02 to $0.05/min for ElevenLabs or similar), and telephony via Twilio or Telnyx (add $0.01 to $0.02/min). The real all-in rate lands between $0.23 and $0.33 per minute , depending on model and voice choices. At 2,000 minutes, that is $460 to $660 in usage costs alone, arriving as 4 to 5 separate invoices from Vapi, OpenAI, ElevenLabs, and Twilio. See the 5-invoice problem breakdown for how this plays out at scale.

Retell advertises $0.07/min with no platform fee. That rate is also an undercount. According to the Retell AI 2026 pricing analysis , the real all-in rate lands between $0.13 and $0.31 per minute once you add LLM and telephony. The headline rate is the floor on the simplest possible configuration, not the ceiling on a production deployment. At 2,000 minutes, that is $260 to $620. Still no white-label on standard plans. Still no built-in CRM or campaign tools.

Synthflow removed its published agency tiers for new users in 2026. Pay-as-you-go users pay the component costs separately. Enterprise users negotiate a custom rate. The legacy agency plan was around $1,250/month at the entry tier, with the Retell AI review putting enterprise agency pricing at $3,400/month . If you are a new agency evaluating Synthflow in June 2026, the pricing page will not give you a number to plan against.

Hermes is the only platform on this list with a published flat rate that covers everything. Starter is $149/month with 300 included minutes. Business is $399/month with 1,000 included minutes. Agency is $699/month with 2,000 included minutes. Overage is $0.24/minute flat, including STT, LLM, TTS, and telephony. One invoice. One rate you can price your retainers against.

How does reliability compare across these platforms?

Platform reliability is not an abstract concern for agencies. It is a client retention variable. A 4-hour platform outage during an active campaign means missed calls, callback queues, and a conversation with your retainer client about what happened. Here is the documented incident record for the platforms where data is publicly available.

Vapi reliability: In the 90-day period ending June 2026, Vapi recorded 18 incidents according to IsDown monitoring , including 5 major outages. On May 21, 2026, the entire Vapi US platform went offline for 4 hours and 7 minutes due to database connection exhaustion, taking down calls, API access, the dashboard, and logs simultaneously. Independent monitoring detected 33 Vapi outages in March 2026 before Vapi acknowledged them, plus 87 incidents Vapi never reported at all. One Trustpilot reviewer documented $50,000 in damages from downtime caused by Vapi bugs, citing unreliable systems and non-existent support.

Retell reliability: Retell has a published status page and a better incident disclosure track record than Vapi. The 31-incident analysis over 11 months shows Retell is not immune to outages, but is more transparent about them. For agencies running critical campaigns, the recommendation is to build fallback routing regardless of which platform you use.

Synthflow reliability: Multiple G2 reviewers in 2026 cite API instability, calls that drop mid-conversation, and minutes that register as consumed without a matching call event. The Retell AI Synthflow review flags these as persistent issues at mid-volume deployments.

How does white-label capability compare?

White-label is not a checkbox. It is an architecture question. There is a meaningful difference between a platform that slaps a logo onto a shared subdomain and a platform that was built from the ground up to be invisible to your clients. Here is where each platform actually stands.

Vapi has no white-label offering on standard plans. The API is brandless, but there is no client portal, no white-label dashboard, and no sub-account structure you can surface to clients under your brand. You are building any client-facing interface yourself.

Retell does not offer white-label on standard plans either. According to the VoiceAIWrapper white-label analysis , white-label for agencies requires an enterprise negotiation with Retell. Agencies on standard plans who want a branded client experience typically use a third-party wrapper on top of Retell, which introduces the upstream dependency risks covered in the white-label audit checklist .

Synthflow does offer white-label as part of its agency offering, and that was the main reason agencies chose it over Retell and Vapi. With the removal of published agency tiers, the white-label feature is still present but accessed at enterprise pricing. G2 reviewers in 2026 also note cases where Synthflow branding surfaced to end clients in support escalations, which is a structural limitation rather than a configuration issue.

Hermes ships native white-label on every plan, including Starter at $149/month. Your clients log into your branded portal. Your support tickets stay inside your workspace. Every outbound communication is under your brand. Your clients never see the word Hermes. This is not a premium feature. It is how the platform was built because that is what agencies actually need.

Which platform is right for which agency size?

The honest answer is that Retell and Vapi are infrastructure layers, not agency platforms. They give you a raw capability and expect you to build the rest. That is appropriate if you have engineering resources and a specific technical vision. It is not appropriate if you are a 1-to-5-person agency that needs to be selling and delivering, not building infrastructure.

Here is the breakdown by agency profile.

Solo operator or new agency (0 to 3 clients): You need to get to revenue fast, not to become a backend engineer. Retell PAYG is technically cheaper at very low call volume if you are willing to wire up your own CRM and billing. In practice, the time cost of that setup is significant. Hermes Starter at $149/month gives you the full stack from day one. The day-one stack guide covers this in detail.

Growing agency (3 to 10 clients): This is where the hidden costs of Vapi and Retell start to compound. At $3,000 to $15,000/month in client revenue, the margin impact of a 5-invoice billing stack becomes real. The margin math analysis shows the typical pattern. Hermes Business at $399/month with 1,000 included minutes and 7 client workspaces is the right tier for this profile.

Established agency (10 to 20 clients): At this scale, the operational overhead of managing multiple billing relationships, a custom-built white-label layer, and separate campaign tooling is a real cost. Hermes Agency at $699/month with 2,000 included minutes and 20 workspaces is built for exactly this profile. If you are currently on Synthflow and evaluating alternatives because the pricing structure changed, the Synthflow migration guide covers the full process.

Developer-first team building custom voice products: Vapi and Retell are the right tools here. If you have engineers, need to swap LLM providers per call stage, or are building voice as a core product capability rather than an agency service, the API-first approach is appropriate. You will pay more per minute in practice and build more infrastructure, but you get the most flexibility.

What does latency look like across 1,200 agency calls?

Latency is the variable your clients feel most directly. A voice agent that pauses 1.5 seconds before responding does not feel like a person. It feels like bad customer service. Published benchmarks and production reality are different numbers.

Published benchmarks from 2026 show Vapi at 500 to 600ms under optimal provider pairings, Retell at 580 to 620ms in production, and Synthflow claiming sub-500ms in controlled tests. These are best-case numbers from the vendors themselves. Independent testing across platforms puts the realistic range at 500 to 800ms for Vapi, 580 to 620ms for Retell, and variable for Synthflow depending on model and region.

What the benchmarks do not capture is the latency impact of instability. A platform averaging 600ms that experiences a 4-hour outage every 2 weeks produces worse real-world call quality for your clients than a platform averaging 700ms that does not go down. The relevant question for agency deployments is median latency under normal load plus variance during incidents, not the best-case benchmark under ideal conditions.

What does Synthflow's enterprise pivot mean for existing agency customers?

If you are an existing Synthflow customer on a legacy agency plan, your plan is grandfathered for now. The platform is not shutting down. But the strategic direction is clear. The $20M Series A in 2025 funded a pivot toward enterprise BPOs and large contact centers. The removal of published agency tiers for new users is the first public signal of where the product roadmap is headed.

The pattern here is not new. Voicerr raised prices 7 to 10x overnight in early 2026. Synthflow's move is slower and softer, but the structural logic is the same: once a platform finds a higher-value buyer segment, the agency pricing gets deprioritized. Agencies building on Synthflow should be thinking about what the pricing structure looks like in 12 months when the next contract renewal comes due.

The data from the Synthflow enterprise pivot analysis shows the full arc of this repositioning and what it means for agencies currently on the platform.

What should you actually look for in an agency voice platform?

After running 1,200 calls across these platforms and speaking with agency owners managing everything from 2 clients to 30, the evaluation framework that actually matters comes down to six questions.

Can you price your retainers against a fixed number? If the platform bills 4 to 5 components separately and any one of them can change price without notice, you cannot set accurate retainer rates. You need a flat all-in ceiling.
Is white-label native or bolted on? A logo toggle on a shared subdomain is not white-label. If your clients ever see a vendor name in a support escalation, a system email, or an error message, the white-label is incomplete. Ask to see the full client-facing flow before you commit.
What is the platform's incident rate and response pattern? Ask for the status page URL and look at the last 90 days before you sign. The difference between 2 incidents and 18 incidents in 90 days is not small.
Does the platform have a native CRM and campaign layer? If you need separate tools for contact management and outbound campaigns, you are adding tool sprawl and additional failure points. The fewer the external dependencies, the more reliable the system.
Is the platform built for agencies or tolerating them? A platform that added agency features as an afterthought will route its roadmap toward its highest-value segment. Check whether agency-specific features (multi-tenant isolation, sub-account billing, per-client reporting) are in the core product or in an enterprise add-on tier.
What happens when you need to leave? Self-serve account management, number portability, and data export should be standard. If any of these require a support ticket, that is a yellow flag. If account deletion requires a support email, that is a red flag. The Synthflow cancellation experience is a useful reference point here.

The honest summary

Vapi and Retell are excellent infrastructure layers. They are not agency platforms. The gap between their advertised rate and their real rate is not a marketing trick, it is an architecture choice: they provide one layer and bill for it honestly, and everything else is your problem. If you have engineering resources and need maximum flexibility, they are the right starting point.

Synthflow was the right choice for no-code agency operators in 2023 and 2024. The enterprise pivot does not make it bad software. It makes it software whose roadmap is now aligned with a different customer. That matters when you are planning 12 months out.

Hermes is the one platform on this list that was built specifically for what AI voice agencies actually do: manage multiple clients under one roof, maintain white-label isolation, control margins, and scale without depending on 5 separate vendor relationships. One invoice. Your brand. From $149/month. First agent live in 72 hours.

The comparison page below has the full feature breakdown in a structured format: Hermes vs Synthflow . If you want a cost breakdown against your current Vapi or Retell bill, the voice bill audit tool runs the comparison automatically.

Frequently asked questions

{f.q}

{f.a}

See how Hermes compares against your current bill

Upload your last Vapi, Retell, or Synthflow invoice and get a side-by-side cost comparison with Hermes. Or apply for the Founders' Beta and get your first agent live in 72 hours.

Alfredo Romero is CEO of Hermes, the voice infrastructure platform for AI agencies. Connect on LinkedIn .

Frequently asked questions

Which voice AI platform is best for a small agency with 1 to 5 clients?

For a solo operator or small agency just getting started, Hermes Starter at $149/month gives you 300 included minutes, 3 client workspaces, and a built-in CRM without needing additional tools. Retell PAYG is cheaper at low volume but requires you to build or buy CRM, white-label, and billing tooling separately. Vapi has the same problem at a higher per-minute rate once you add all the upstream costs. Synthflow removed its entry-level agency tiers in 2026, so new agency users on Synthflow now start at pay-as-you-go or custom enterprise pricing.

What is the real all-in cost per minute for Vapi?

Vapi's advertised rate is $0.05/minute for the platform layer. But that number does not include the LLM (add $0.03 to $0.10/min for GPT-4o or Claude-class models), the text-to-speech voice (add $0.02 to $0.05/min for ElevenLabs or similar), and telephony via Twilio or Telnyx (add $0.01 to $0.02/min). The real all-in rate lands between $0.23 and $0.33 per minute depending on configuration choices. At 2,000 minutes per month, that is a range of $460 to $660 in usage costs alone, before any platform fee.

Does Retell AI offer white-label for agencies?

Retell does not offer native white-label on standard plans. White-label access requires an enterprise negotiation. This means agencies on standard Retell plans cannot rebrand the platform under their own logo for clients. Third-party wrapper tools exist to add a white-label layer on top of Retell, but those wrappers inherit Retell's upstream risk, pricing changes, and outage exposure. Hermes, by contrast, ships native white-label as a default across all plans, including Starter.

How reliable is Vapi for a production agency deployment?

Vapi had 18 incidents in the 90-day period ending June 2026, including 5 major outages. The May 21, 2026 outage took the entire US platform offline for over 4 hours due to database connection exhaustion, affecting calls, the API, the dashboard, and logs. Independent monitoring by IsDown detected 33 Vapi outages in March 2026 before Vapi acknowledged them, plus 87 incidents that Vapi never publicly reported. For agencies with active client campaigns, a 4-hour platform outage is a client retention risk, not just a technical inconvenience.

Why did Synthflow remove its Agency pricing plan?

Synthflow raised a $20 million Series A in 2025 and subsequently repositioned toward enterprise BPOs and contact centers. The removal of tiered agency plans in favor of a pay-as-you-go and custom enterprise model reflects that strategic shift. Agencies that were on Synthflow's legacy Agency plan (previously around $1,250/month) now face either pay-as-you-go rates or enterprise negotiations starting around $30,000 annually. The Retell AI 2026 Synthflow review puts the agency plan starting price at $3,400/month at the enterprise tier.

What is the fastest-responding voice AI platform for agency use in 2026?

Latency benchmarks across the tested platforms in 2026 show Synthflow and Vapi both claim sub-500ms response times under optimal conditions, while Retell AI consistently achieves 580 to 620ms in production. Hermes is built on top of optimized inference routing and targets sub-700ms end-to-end in standard configurations. The practical question for agencies is not which platform has the best benchmark in isolation, but which platform delivers consistent latency across all your client use cases under production load. Outage-prone platforms with lower benchmark latency produce worse real-world call quality than stable platforms with slightly higher targets.

Can I use Hermes if I am already running Retell or Vapi?

Yes. Hermes is designed as a migration destination for agencies moving off Retell, Vapi, or Synthflow. The platform ships with a guided migration workflow. If you are using Twilio numbers with Retell or Vapi, those numbers can be routed through Hermes without porting. Existing agent prompts and flows are rebuilt inside Hermes, which typically takes 2 to 4 hours per agent. Hermes offers a parallel-run period so you can test agents on Hermes before cutting over from your current platform.

This article was originally published on the Hermes blog. Hermes is the white-label operating platform for AI voice agencies. One platform, your brand, from $149/month.

OpenAI Shipped Your Voice Stack at $0.25/Min. Vapi Went Enterprise. The Infra Layer Abandoned Agencies in Eleven Days.

Alfredo Romero — Thu, 25 Jun 2026 14:36:27 +0000

On May 7, OpenAI made gpt-realtime-2 generally available in the Realtime API at roughly $0.25 to $0.35 per minute of conversation, with GPT-5-class reasoning, 128K context, native translation, and 70+ languages. Coverage: TechCrunch, OpenAI's announcement, and a clean teardown at DataCamp. Five days later, on May 12, Vapi closed a $50M Series B at a $500M valuation and publicly positioned itself as the enterprise infrastructure layer, with Amazon Ring running 100% of inbound through them. Two days after that, Synthflow's enterprise page started leading with two deal proof points: a $230M multinational BPO operator running 40+ branded agents at 600K calls per month, and a top U.S.-based CRM platform white-labeling Synthflow at 500K calls per month, both inside 60 days. The full receipts are on Synthflow's enterprise comparison page.

Eleven days. Three events. One outcome. The middle of the AI voice stack collapsed, and an agency owner with five clients is the only customer left in the room nobody is engineering for.

Why this matters for AI voice agencies

For the last 18 months, the canonical AI voice agency stack was a wrapper or infra platform (Retell, Vapi, Voicerr, Synthflow) plus GoHighLevel plus Zapier plus Stripe plus Twilio plus a custom dashboard. The wrapper layer existed because the orchestration was hard. Stitching STT, an LLM, TTS, telephony, turn-taking, barge-in, latency tuning, and call quality together was a four-vendor problem with 200ms of jitter to manage. You paid the wrapper for the integration, not the model.

On May 7 the orchestration disappeared. OpenAI now handles speech-to-speech reasoning inside one API call, in one model, with caching that drops the dominant cost line by 80x. The wrapper's reason to exist was the stitching. The stitching just got eaten by the model layer. eWeek summarized the strategic shift cleanly:

"OpenAI's bet is that doing the audio reasoning inside one model is more defensible than stitching three vendors together. Whether ElevenLabs, Deepgram and the rest can hold their wedge depends on how quickly they push their own integrated stacks."

At the same time, the wrappers and infra companies that could not be commoditized from below decided to escape upward. Vapi's TechCrunch story is the public version of this. The internal version is the customer list at the launch of gpt-realtime-2 on the same week: Zillow, Glean, Genspark, Bluejay, Intercom, Priceline, Foundation Health, Deutsche Telekom. Read it aloud. Not one agency. Synthflow's customer list is the same shape: a $230M BPO, a national CRM platform, 600K calls per month, 500K calls per month, dedicated success engineers, procurement-driven roadmaps.

The implication for the agency layer is concrete, not vibes. Your platform's product roadmap is now being written for a customer who pays a six-figure annual contract and signs a master service agreement. Your support ticket about a missing white-label string in a webhook payload sits behind their compliance review. Your feature request for a campaign retry rule sits behind their procurement cycle. That is what "enterprise focus" actually means in the day-to-day. It is not a marketing line. It is the prioritization queue.

Meanwhile the wrapper era's pricing power evaporated in the other direction. Voicerr already moved from $28 per month to $199 to $299 per month, a 7x to 10x hike documented at Trillet, because the upstream cost moved underneath them and they had no other lever. They lost both directions in the same quarter. Pricing power from above, model commoditization from below. A wrapper sitting on top of an infra layer that just became one OpenAI call has nowhere to go.

Three paths are now visible for the agency operator with three to fifteen clients:

DIY on Pipecat plus OpenAI plus Twilio plus your own glue. Pipecat hit v1.0.0 on April 14, 2026, so the orchestration scaffolding is production-grade open source. Honest cost: $40K to $60K in dev time before you close your sixth retainer, plus a permanent on-call rotation. Real path for a builder with a co-founder engineer and four months. Not the path for an agency that wants to ship a new agent this week.
Stay on a wrapper that is losing its reason to exist. Pay 7x to 10x the old sticker price. Inherit every upstream pricing shock. Wait for the next "subscription update" email.
Move to an application-layer platform built for agencies. Get the CRM, the campaign engine, the white-label portal, the workspace structure, the billing surface, the A2P submission flow, the recording and transcript pipeline, and the prompt versioning in one place. Let the platform handle the cache strategy, the model routing, and the upstream commercial relationship. Keep the spread.

What we're doing at Hermes about it

Hermes is the operating platform for AI voice agencies. It is the third path above. We are not a wrapper on Vapi. We are not a thin UI on Retell. We are the application layer the infra companies just walked away from.

The pricing stays where it has been since we wrote it down. Starter is $149 per month with 300 included minutes and 3 workspaces. Business is $399 per month with 1,000 included minutes and 7 workspaces. Agency is $699 per month with 2,000 included minutes and 20 workspaces. Overage is $0.24 per minute against a $0.18 landed cost on the new gpt-realtime-2 economics, and we run a multi-model routing strategy on top so the cache hit rate is actually load-bearing instead of theoretical. The 25% spread is locked because we run the upstream relationship. Wrappers cannot promise this. They do not control the cost structure and they have already proved it by raising prices 7 to 10 times.

The application layer is the work product. White-label is native, not bolted on. Every workspace has its own subdomain, custom branding, end-client portal, and per-workspace billing. The CRM is built into the same database as the campaign engine, so a call outcome writes a lead status without a Zapier hop. The campaign engine knows how to retry, schedule callbacks, hand off to a human, and reconcile A2P 10DLC submissions ($30 pass-through, $0 margin). The billing surface knows the difference between an included minute and an overage minute. None of this is a feature roadmap. It is the live platform, and it is the layer OpenAI's API does not ship.

Action steps for agencies affected (this week, not next quarter)

Recompute your real per-minute cost on the new gpt-realtime-2 economics. Pull your last 30 days of upstream invoices (Vapi, Retell, Voicerr, whoever). Divide by minutes used. If your landed cost per minute is north of $0.20 and you are charging clients $0.30 to $0.45 per minute, your spread is closing while the model layer's spread is widening. That is the leak.
Ask your current platform, in writing, where the agency tier sits in the product roadmap. Phrase it specifically: "What new features shipped for sub-100-call-per-day operators in Q1 2026?" If the answer is enterprise SSO, audit logs, dedicated success engineers, or a partner program for BPOs, you have your answer.
Stress-test your client-facing surface, not the model. Ten simultaneous calls. Thirty. Watch the latency curve. Watch which dashboard your client logs into. Watch whether the email confirmations carry your domain or a "Powered by" footer. The model is not your problem anymore. The application layer is.
Tell your clients before they hear it from a YouTube tutorial. Liam Ottley, Caleb Casas, Brendan Jowett, and Daniel Walter have not yet published "what gpt-realtime-2 means for your AI voice agency stack" videos. They will this week. Send the operator-level note to your clients first.
Move the parts of the stack you do not control. You cannot control OpenAI's pricing or Vapi's customer focus or Synthflow's enterprise pivot. You can choose whether you build the application layer yourself, rent it from a wrapper that just lost its wedge, or run on a platform built for agency operators who need new agents live in 72 hours.

Frequently asked questions

If gpt-realtime-2 is $0.25 to $0.35 per minute and Hermes overage is $0.24 per minute, how does the math even work?

Hermes runs a multi-model routing strategy with cache configuration, system prompt reuse, and per-call telemetry tuned at the platform level. Cached audio input on gpt-realtime-2 drops from $32 per 1M tokens to $0.40 per 1M tokens, which is an 80x reduction on the dominant cost line in a real outbound conversation. Combine that with selective model routing for non-reasoning segments (greeting, clarifying questions, scheduled callbacks, voicemail) and the landed cost lands around $0.18 per minute. We charge $0.24. That is a 25% spread, locked, on top of three included-minute tiers ($149 / 300 min, $399 / 1,000 min, $699 / 2,000 min). An agency owner does not need to engineer cache hits or pick which model handles which conversation segment. The platform does it.

If OpenAI now ships everything in one API call, why does an agency need a platform at all?

Because the platform is no longer the voice. The platform is the CRM, the campaign engine, the workspace structure, the white-label portal, the billing reconciliation, the A2P 10DLC submission, the lead status writebacks, the call attribution, the recording storage, the transcript pipeline, the knowledge base management, the prompt versioning, the retry logic, and the client-facing dashboard. None of that ships in the gpt-realtime-2 endpoint. The voice infrastructure collapsed into one API call. The application layer did not. That is the wedge an agency-built platform fills.

Should I just build this myself on Pipecat plus OpenAI now that the infra is cheap?

If you have a dedicated engineer and four months to burn, yes, the DIY path got cleaner with Pipecat hitting v1.0.0 on April 14. If you have five clients and need new agents live this week, no. The honest math is that a self-built stack on Pipecat plus OpenAI plus Twilio plus a CRM plus a billing surface plus a white-label portal costs $40K to $60K in developer time before you close your sixth retainer, and the maintenance does not stop. The reason platforms still exist is that the application layer is wide, not deep. We built it once. You charge clients $1,500 to $3,000 a month on top.

Where this leaves you

The infrastructure layer did not pivot to enterprise out of malice. It pivoted because the model layer commoditized it from below and the only revenue ramp left is the six-figure logo. Vapi did the right thing for its cap table. Synthflow did the right thing for its cap table. OpenAI did the right thing for its cap table. None of them owe the AI voice agency owner a different roadmap.

The agency layer does the right thing for itself by noticing this eleven-day window early. The wedge that opens when an infra layer graduates is the wedge an application-layer platform built specifically for agencies should occupy. By builders, for builders. One platform. Your brand. Your margins. From $149 per month. First agent live in 72 hours.

Originally published at buildwithhermes.com/blog/infra-layer-abandoned-agencies-eleven-days-2026-05-18.

The Voicerr 10x Price Hike: A Migration Playbook for AI Calling Agencies

Alfredo Romero — Wed, 24 Jun 2026 16:39:20 +0000

If you run a voice AI agency on Voicerr, you woke up to a brutal reality in late 2025: Voicerr raised prices 7-10x overnight. What cost $28/month six months ago now costs $199-$299/month. For agencies running 5-20 clients, that's a swing from ~$140/month to $1,500+/month in fixed subscription costs alone, before you add per-minute voice charges on top.

This isn't a bug in your invoice. This is a deliberate platform pivot. Voicerr, built as a white-label wrapper on top of VAPI and Retell, decided to charge enterprise prices while keeping enterprise features out. And they're not alone. PlayAI shut down entirely after Meta acquired them. Synthflow users are reporting cancellation billing traps. Assistable just migrated everyone to V3 with zero backwards compatibility. The wrapper economy is imploding.

The good news: migrating off Voicerr is faster than you think, and there are platforms built specifically for agencies that won't pull this again. This playbook walks you through the migration decision, the technical steps, and how to communicate this to your clients without losing deals.

Why Voicerr (and every wrapper) raised prices

Voicerr is not a voice AI platform. It's a white-label UI on top of VAPI and Retell's infrastructure. Voicerr doesn't control pricing, uptime, or features. It just wraps them and sells access.

When VAPI and Retell prices rose, Voicerr's business model broke. They had two choices: raise their flat fee, or watch margins collapse. They chose option one.

The fatal flaw: Voicerr's per-minute pricing didn't change. Voice minutes still cost $0.15-$0.25/min on VAPI or $0.12-$0.15/min on Retell, exactly the same as if you connected directly. So agencies now pay $199-$299/month for a white-label dashboard that doesn't add real value over connecting directly to Retell and building their own ops.

This is the structural flaw in wrapper economics: wrappers don't control their unit economics. When upstream providers raise prices, wrappers have no leverage and must choose between raising flat fees or going bankrupt.

The broader wrapper collapse of 2025-26

Voicerr's move is part of a larger pattern. By early 2026, the white-label voice AI wrapper ecosystem had fractured:

PlayAI: Acquired by Meta in July 2025, entire platform shut down by December 31, 2025. 40,000+ users lost everything.
Voicerr: 7-10x price hike from $28 to $199-$299/month in late 2025.
Synthflow: Reported cancellation billing traps and unresponsive customer service post-cancellation.
Assistable: V3 migration broke backwards compatibility, forcing reconfigurations.

The common thread: wrappers are dependent on upstream providers, have no negotiating power, and fail fast when their cost model breaks or they're acquired.

Should you migrate? The decision framework

Before migrating, ask these three questions:

1. What is your true all-in cost on Voicerr vs. direct?
Voicerr's new pricing: $199-$299/month + $0.15-$0.25/min on VAPI or $0.12-$0.15/min on Retell. Direct VAPI connection: $0 subscription (just API calls) + $0.15-$0.25/min. Direct Retell connection: $0 subscription + $0.12-$0.15/min.

Math: If you run 10,000 minutes/month on Voicerr at $0.15/min, that's $1,500 in minutes + $250/month subscription = $1,750. The same 10,000 minutes direct to Retell at $0.12/min = $1,200. You're paying Voicerr $550/month for a white-label UI you could build yourself in 2 weeks with Hermes' templates.

2. Do you need white-label? Really?
If your clients never see "Voicerr" and you've already white-labeled the agent dashboard, migrating to a full platform won't change your client's experience. You'll just have a cleaner ops setup and better margins.

3. How much runway do you have?
If Voicerr's new bill just ate 3 months of profit, migrate now. If you can absorb it for 60 days, take time to pick the right platform.

Migration target comparison (as of June 2026)

Platform	Subscription	Per-Min Cost	White-Label	Migration Time
Hermes	$149-$699/mo	$0.24/min	Native	2-3 days
Synthflow	$99-$399	$0.10-$0.20	Yes	3-5 days
Direct Retell	None	$0.12-$0.15	DIY build	2-4 weeks
VoiceAIWrapper	$29/mo	Pass-through	Yes	1-2 days

Step-by-step migration playbook (3-5 days)

Step 1: Export your agents and knowledge bases. In Voicerr, download all agent configurations as JSON. Export knowledge base documents. If Voicerr doesn't have a bulk export, use your browser's DevTools to inspect API calls and replicate the export manually. Expect 30-60 minutes per agent.

Step 2: Choose your target platform. For agencies with 3+ clients: Hermes (purpose-built for agencies, native white-label, CRM included). For quick wins on a budget: VoiceAIWrapper ($29/mo). For compliance-heavy industries: native platforms like Synthflow. Avoid other wrappers unless you like this situation in 12 months.

Step 3: Set up phone number forwarding (not porting). Don't port numbers immediately. Instead, configure call forwarding from Voicerr to your new platform for 7 days. This gives you a rollback path. Once both systems are live and agents match, then port the numbers. Porting typically takes 1-3 business days. Use those days to clone and test agents in parallel.

Step 4: Recreate agents in new platform. Most platforms (Hermes, Synthflow, VoiceAIWrapper) let you import agent configs from JSON or scrape knowledge base URLs directly. Import your exported agents. Re-test each one against a test number. Most agencies complete this in 4-8 hours for 5-10 agents.

Step 5: Shadow-run both systems for 24-48 hours. Send test calls to both Voicerr and the new platform simultaneously. Compare output quality, response time, and error handling. If something is wrong, you catch it before clients notice.

Step 6: Cut over and monitor closely. On cutover day, port numbers, flip your IVR routing, and watch the dashboard for the first 2 hours. Have your backup number (Voicerr still live on forwarding) ready to flip to if needed. After 24 hours with zero errors, cancel Voicerr.

How to tell your clients without losing deals

Your clients should not care what platform you use. But if they hear you're leaving Voicerr, the first thing they think is "instability," not "better margins."

Here's the framing:

"We're migrating to [Platform] to give you better service: faster support, better integrations, and cleaner reporting. Your agent stays the same, your numbers stay the same, but you'll see improvements in response time and reliability. We're handling the whole migration, zero downtime. Here's your new dashboard link. Let us know if you see anything different."

That's it. Don't explain wrappers or Voicerr's financial problems. Just frame it as an upgrade. Your clients care about results, not infrastructure.

The real lesson: own your infrastructure or plan to migrate every year

Voicerr's 7-10x hike isn't malice. It's math. Wrappers have zero negotiating power with their upstream providers. When costs rise, they have three options: raise prices, die, or get acquired and shut down (see PlayAI).

The agencies that avoid this trap build on platforms that own their infrastructure. Hermes owns our voice routing, CRM layer, and billing. We don't pass upstream pricing risks to you. When per-minute costs fluctuate, it's our problem, not yours.

If you're going to migrate anyway, migrate to a platform that won't force you to do this again in 12 months. That's the only real escape from the wrapper trap.

Frequently asked questions

Will my phone numbers work on the new platform? Yes. Most platforms (Hermes, Synthflow, VoiceAIWrapper) support number porting. You own the numbers through your carrier. You just point them at a new IVR. Takes 1-3 business days to port, 24 hours to configure.

Will my existing agent data transfer? Mostly. Agent config (prompt, knowledge base, tools) transfers via JSON export/import. Call history and analytics don't transfer. You'll have a clean slate on reporting, but your live agents pick up seamlessly.

How long will the migration take? 3-5 days for most agencies with 5-15 clients. 1-2 hours to export, 4-8 hours to recreate agents, 24 hours shadow-run, then cutover. If you have 30+ agents or complex custom code, add 1-2 weeks.

Originally published on the Hermes blog. Hermes is the operating platform for AI voice agencies. One platform, your brand, from $149/month. Start a free trial.

The Voicerr 10x Price Hike: A Migration Playbook for AI Calling Agencies

Alfredo Romero — Wed, 24 Jun 2026 15:50:52 +0000

The Voicerr 10x Price Hike: A Migration Playbook for AI Calling Agencies (Without Losing Numbers or Clients)

Why Voicerr (and every wrapper) raised prices

Voicerr is not a voice AI platform. It's a white-label UI on top of VAPI and Retell's infrastructure. Voicerr doesn't control pricing, uptime, or features. It just wraps them and sells access.

When VAPI and Retell prices rose, Voicerr's business model broke. They had two choices: raise their flat fee, or watch margins collapse. They chose option one.

This is the structural flaw in wrapper economics: wrappers don't control their unit economics. When upstream providers raise prices, wrappers have no leverage and must choose between raising flat fees or going bankrupt.

The broader wrapper collapse of 2025-26

Voicerr's move is part of a larger pattern. By early 2026, the white-label voice AI wrapper ecosystem had fractured:

PlayAI: Acquired by Meta in July 2025, entire platform shut down by December 31, 2025. 40,000+ users lost everything.
Voicerr: 7-10x price hike from $28 to $199-$299/month in late 2025.
Synthflow: Reported cancellation billing traps and unresponsive customer service post-cancellation.
Assistable: V3 migration broke backwards compatibility, forcing reconfigurations.

The common thread: wrappers are dependent on upstream providers, have no negotiating power, and fail fast when their cost model breaks or they're acquired.

Should you migrate? The decision framework

Before migrating, ask these three questions:

1. What is your true all-in cost on Voicerr vs. direct?

Voicerr's new pricing: $199-$299/month + $0.15-$0.25/min on VAPI or $0.12-$0.15/min on Retell.

Direct VAPI connection: $0 subscription (just API calls) + $0.15-$0.25/min.

Direct Retell connection: $0 subscription + $0.12-$0.15/min.

Math: If you run 10,000 minutes/month on Voicerr at $0.15/min, that's $1,500 in minutes + $250/month subscription = $1,750. The same 10,000 minutes direct to Retell at $0.12/min = $1,200. You're paying Voicerr $550/month for a white-label UI you could build yourself in 2 weeks with Hermes' templates.

2. Do you need white-label? Really?

If your clients never see "Voicerr" and you've already white-labeled the agent dashboard, migrating to a full platform won't change your client's experience. You'll just have a cleaner ops setup and better margins.

3. How much runway do you have?

If Voicerr's new bill just ate 3 months of profit, migrate now. If you can absorb it for 60 days, take time to pick the right platform.

Migration target comparison (as of June 2026)

Platform	Subscription	Per-Min Cost	White-Label	Migration Time
Hermes	$149-$699/mo	$0.24/min	Native	2-3 days
Synthflow	$99-$399	$0.10-$0.20	Yes	3-5 days
Direct Retell	None	$0.12-$0.15	DIY build	2-4 weeks
VoiceAIWrapper	$29/mo	Pass-through	Yes	1-2 days

Step-by-step migration playbook (3-5 days)

Step 1: Export your agents and knowledge bases

In Voicerr, download all agent configurations as JSON. Export knowledge base documents. If Voicerr doesn't have a bulk export, use your browser's DevTools to inspect API calls and replicate the export manually. Expect 30-60 minutes per agent.

Step 2: Choose your target platform

For agencies with 3+ clients: Hermes (purpose-built for agencies, native white-label, CRM included). For quick wins on a budget: VoiceAIWrapper ($29/mo). For compliance-heavy industries: native platforms like Synthflow.

Avoid other wrappers unless you like this situation in 12 months.

Step 3: Set up phone number forwarding (not porting)

Don't port numbers immediately. Instead, configure call forwarding from Voicerr to your new platform for 7 days. This gives you a rollback path. Once both systems are live and agents match, then port the numbers. This is the single safest way to zero your downtime risk.

Porting typically takes 1-3 business days. Use those days to clone and test agents in parallel.

Step 4: Recreate agents in new platform

Most platforms (Hermes, Synthflow, VoiceAIWrapper) let you import agent configs from JSON or scrape knowledge base URLs directly. Import your exported agents. Re-test each one against a test number. Most agencies complete this in 4-8 hours for 5-10 agents.

Step 5: Shadow-run both systems for 24-48 hours

Send test calls to both Voicerr and the new platform simultaneously. Compare output quality, response time, and error handling. If something is wrong, you catch it before clients notice.

Step 6: Cut over and monitor closely

On cutover day, port numbers, flip your IVR routing, and watch the dashboard for the first 2 hours. Have your backup number (Voicerr still live on forwarding) ready to flip to if needed. After 24 hours with zero errors, cancel Voicerr.

How to tell your clients without losing deals

Your clients should not care what platform you use. But if they hear you're leaving Voicerr, the first thing they think is "instability," not "better margins."

Here's the framing:

"We're migrating to [Platform] to give you better service: faster support, better integrations, and cleaner reporting. Your agent stays the same, your numbers stay the same, but you'll see improvements in response time and reliability. We're handling the whole migration, zero downtime. Here's your new dashboard link. Let us know if you see anything different."

That's it. Don't explain wrappers or Voicerr's financial problems. Just frame it as an upgrade. Your clients care about results, not infrastructure.

What you'll actually save

Let's math this for a realistic agency with 8 clients, 5,000 minutes/month per client (40,000 total):

Voicerr (new pricing):
  Subscription: $299 (Agency plan) x 8 sub-workspaces = $2,392
  Voice: 40,000 min x $0.15/min = $6,000
  Total: $8,392/month

Hermes:
  Subscription: $699 (Agency plan, covers all 8)
  Included minutes: 2,000/month (rollover unused)
  Overage: (40,000 - 2,000) x $0.24 = $9,120
  Total: $9,819/month

Hermes: $1,427 more per month, but includes CRM, campaigns,
and analytics. Voicerr: pure voice only.

Voicerr is cheaper on voice minutes alone, but Hermes includes everything else. If you're currently building your own CRM layer and campaign orchestration on top of Voicerr, Hermes cuts that dev time by 80%.

More important: Hermes doesn't own an upstream provider. We control our own infrastructure. Prices won't jump 7x in six months.

The real lesson: own your infrastructure or plan to migrate every year

If you're going to migrate anyway, migrate to a platform that won't force you to do this again in 12 months. That's the only real escape from the wrapper trap.

Frequently asked questions

Will my phone numbers work on the new platform?

Yes. Most platforms (Hermes, Synthflow, VoiceAIWrapper) support number porting. You own the numbers through your carrier. You just point them at a new IVR. Takes 1-3 business days to port, 24 hours to configure.

Will my existing agent data transfer?

Mostly. Agent config (prompt, knowledge base, tools) transfers via JSON export/import. Call history and analytics don't transfer. You'll have a clean slate on reporting, but your live agents pick up seamlessly.

What if Hermes raises prices too?

Hermes controls its own infrastructure, so pricing is stable. But if you want insurance, use VoiceAIWrapper ($29/mo) as a hedge: it's cheaper and lets you wrap any provider. You can build Hermes' CRM layer on top if you want, and if prices spike again, you're already portable. No lock-in.

How long will the migration take?

3-5 days for most agencies with 5-15 clients. 1-2 hours to export, 4-8 hours to recreate agents, 24 hours shadow-run, then cutover. If you have 30+ agents or complex custom code, add 1-2 weeks.

Should I build my own on Retell/VAPI instead?

Only if you have 4+ weeks of dev time. You'll build the same features Hermes has (white-label, CRM, campaign routing, billing) from scratch. Budget $15K-$30K in dev. Hermes is $699/mo, so you break even in 6-12 months and then save $8K/year.

Next steps

If Voicerr's new pricing is hitting your margins, you have a window to move before 30 days. Here's what to do:

Calculate your true all-in cost (sub + voice minutes)
Compare against Hermes, Synthflow, or direct Retell/VAPI costs
Set up a test environment on the new platform
Export one agent and run it shadow-live for 24 hours
If zero issues, commit to the migration and schedule cutover

If you want to explore Hermes specifically, we've built migration templates and tools to make this frictionless. You can import your Voicerr agents in minutes, start a free trial, and we'll walk you through the whole process.

You shouldn't have to rebuild your infrastructure every time your platform has a financial crisis. Choose a platform that owns its infrastructure. Then move on.

Originally published at buildwithhermes.com/blog/voicerr-10x-price-hike-migration.

TCPA 2026: The New Compliance Rules Every AI Voice Agency Must Follow

Alfredo Romero — Wed, 24 Jun 2026 15:50:47 +0000

The FCC just confirmed that AI-generated voices trigger TCPA restrictions. Non-compliance costs $500 to $1,500 per call with no caps. Here is exactly what you need to know and do this week.

By Alfredo Romero, CEO Hermes. June 8, 2026.

What the FCC Just Ruled

On June 5, 2026, the FCC issued a formal declaratory ruling: AI-generated voices on voice calls are subject to the Telephone Consumer Protection Act (TCPA). This wasn't speculation or guidance. It was confirmation backed by law.

The ruling reaffirmed that AI-generated voices fall under "artificial or pre-recorded voices" under 47 USC 227(b)(1)(B). What that means in English: every AI voice call an agency makes requires prior express written consent from the recipient before the call is placed.

If you're running an AI calling agency and haven't been documenting consent, you're exposed. Colorado's new AI Act (effective June 30, 2026) adds another layer: high-risk AI systems must pass bias testing and documentation before deployment.

Impact: Non-compliance now exposes you to $500 to $1,500 per call penalties with no aggregate cap. Class actions are already rolling: average settlement 2025-2026 is $5M to $20M.

Why This Matters for AI Voice Agencies Right Now

If you're running an AI voice agency on Retell, VAPI, Synthflow, or Hermes, this ruling affects your business model immediately. Most agencies built their campaigns assuming TCPA didn't apply to AI. It does now, and retroactively.

The implications:

Consent is non-negotiable. Every call your agent places needs proof of consent on file.
EBR doesn't help you. You can't rely on Established Business Relationship to skip consent for AI calls.
Class action risk. Competitors and users are already filing. Get ahead of it.
State laws are tightening. Colorado AI Act, California SB 1047 (pending), and others add data handling and bias testing requirements.

The Penalty Structure (It's Brutal)

Negligent Violation: $500 per call. You didn't know about TCPA or made a good-faith mistake. Still $500 per call, no aggregate cap.

Willful Violation: $1,500 per call. You knew about TCPA and ignored it (or should have known and were reckless). Treble damages apply.

Real math, a 5,000-call campaign with non-compliance:

Negligent: 5,000 calls x $500 = $2.5 million
Willful: 5,000 calls x $1,500 = $7.5 million

This is why class actions averaged $5M to $20M in 2025-2026.

What You Need on File RIGHT NOW

Prior Express Written Consent (PEWC). Email, SMS, form submission, or signed agreement from the recipient explicitly agreeing to receive calls from your AI agent. Must be in writing. "Yes" in a chat doesn't count.
Call Timestamp and ID. Proof that the call was placed, who it went to, when it was placed, and which agent made it. Your phone provider (Twilio) handles this.
Consent Proof and Linked Record. The consent document tied to the phone number or contact. If you can't match consent to the call recipient, you lose in court.
AI Disclosure. Colorado AI Act (effective June 30) and proposed federal rules require disclosing that an AI system made the call within the first 2 seconds of the conversation.
Do-Not-Call Check. Screen all numbers against the National DNC Registry before dialing. Hermes, Retell, and Synthflow handle this, but verify it's running.

What Hermes Does About This

We built Hermes from day one with TCPA and state compliance in mind. Here's what's built in:

Consent Manager: Track and link PEWC to every contact. Audit trail included.
DNC Screening: Automatic National DNC Registry check before any call. Configurable per campaign.
AI Disclosure: Auto-inject "This is an AI assistant" disclaimer in first 2 seconds. Configurable per agent.
Call Logging: Every call timestamped, agent ID tracked, recording linkable. Export for legal discovery.
Compliance Reports: Monthly audit reports showing consent coverage, DNC blocks, and call compliance by campaign.

Action Steps for Agencies Affected (This Week)

Day 1: Audit Your Campaigns. List every active campaign. Check whether each contact has written consent on file. Identify gaps. Estimate exposure.

Day 2: Gather Documentation. Pull all consent forms from the past 6 months. Match to phone numbers called. Flag any calls without linked consent. Document DNC screening evidence.

Day 3: Implement Hermes (or upgrade your current platform). Set up consent manager. Import contact list with consent proof. Enable DNC screening and AI disclosure. Run compliance audit report.

Day 4-5: Update Your Process. Require PEWC before any campaign launch. Add AI disclosure to all agent configs. Train team on consent requirements. Document your compliance playbook.

Key Sources and Citations

FCC Declaratory Ruling on TCPA and AI Voices (June 5, 2026): https://www.fcc.gov/document/fcc-confirms-tcpa-applies-ai-technologies-generate-human-voices
Henson Legal: TCPA Compliance Guide 2026: https://www.henson-legal.com/ai-voice-compliance
Ginsburg Law Group: AI Robocalls + TCPA (February 2026): https://ginsburglawgroup.com/2026/02/ai-robocalls-the-tcpa-consent-rules-you-need-to-know/
Colorado AI Act Requirements (Effective June 30, 2026): https://leg.colorado.gov/bills/sb24-205
Class Action Settlements 2025-2026 Analysis (Q1 2026): https://www.classactionexcitement.com/2026-ai-voice-litigation-update

FAQs

Does TCPA really apply to AI voices? Yes, confirmed by FCC June 5, 2026. AI-generated voices are "artificial or pre-recorded voices" under 47 USC 227(b)(1)(B), triggering TCPA consent requirements.

Can I use Established Business Relationship to skip consent? No. EBR exempts manual sales calls from Do-Not-Call rules, but does NOT exempt AI voice calls from TCPA consent. You still need PEWC.

What counts as prior express written consent? Email, SMS, form submission, or signed document where the recipient explicitly agrees to receive calls from your AI agent. "Yes" in chat doesn't count. Must be in writing with clear identification of the calling party.

What are the penalties for non-compliance? $500 per call for negligent violations, $1,500 per call for willful violations, no aggregate cap. A 5,000-call non-compliant campaign can cost $2.5M to $7.5M in penalties.

Do I need a lawyer to be compliant? Recommended. But Hermes includes compliance controls (consent manager, DNC screening, AI disclosure, call logging) that handle 80% of the burden. Get legal review for your consent form and campaign templates.

Originally published at buildwithhermes.com/blog/tcpa-compliance-rules-2026-06-08.