DEV Community: Alfredo Romero

Vapi Knowledge Base Not Loading? The Trieve Migration Workaround

Alfredo Romero — Thu, 16 Jul 2026 14:34:42 +0000

Liquid syntax error: Variable '{{% raw %}' was not properly terminated with regexp: /\}\}/

Voice AI Is Becoming Commodity: Here's How Agencies Survive the Transition

Alfredo Romero — Wed, 15 Jul 2026 14:33:39 +0000

Last week, three things happened. Gradium (Paris-based) raised $100M led by Nvidia for sub-100ms voice models. OpenAI shipped GPT-Live-1, full-duplex speech with live translation. Retell AI made the exclusive Enterprise Tech 30 list. If you're an agency building on Retell or VAPI alone, this is your warning signal: the infrastructure layer is commoditizing. Fast.

Here's the pattern: Venture capital floods into voice models (Gradium $100M, xAI shipping voice builders, OpenAI full-duplex). Infrastructure providers start moving upmarket (Retell to enterprise, Synthflow raising capital, VAPI expanding into platform features). Meanwhile, agencies who "just use" an API provider wake up one day to find: (1) the model they optimized for got replaced, (2) the API pricing changed, (3) the infrastructure provider pivoted to enterprise customers, and (4) their entire competitive advantage evaporated.

The agencies that survive are the ones who own a platform layer right now. Not the ones betting everything on which voice model wins.

Why This Matters for AI Voice Agencies

Voice models are becoming a commodity. Full-duplex is table stakes now. Sub-100ms latency is table stakes. Natural turn-taking is table stakes. In 12 months, every model will have feature parity. That means differentiation can't come from "we use GPT-4o" or "we use Claude audio." It has to come from something the voice provider can't commoditize: your platform, your CRM, your margin protection, your white-label story.

Here's why that matters: If you're currently running on Retell or VAPI, you own zero of those things. You own the agent configuration. You own the prompt. You own the client relationship. Everything else, the voice model, the infrastructure, the billing, the logging, belongs to someone else. When that someone pivots or raises prices (see: Voicerr 10x price hike in 6 weeks), you have three options:

Absorb the cost hit and reduce margins
Pass it to your clients and lose them
Migrate your entire book of business to a new provider (120+ hours of work per 50 clients)

The agencies that own a platform don't face this choice. They can swap out voice models. They can negotiate directly with infrastructure providers because they aggregate demand. They can lock in margins because their cost structure is transparent to them, not hidden in someone else's pricing table.

The Three Tiers of Voice AI Agencies (and Who Wins)

Tier 1 (Commodity): "We hook Retell into GoHighLevel and call it a platform." Completely exposed to infrastructure risk. When Retell goes down (31 outages in the past 18 months per our audit), your clients' campaigns stop. When a new model launches, you're waiting for the wrapper community to catch up. When pricing changes, you're hoping your clients don't notice.

Tier 2 (Platform-Light): "We built a white-label dashboard on top of VAPI." Better than Tier 1, but you still inherit VAPI's outages, VAPI's pricing, VAPI's roadmap decisions. You're one acquisition away from being shut down (see: what happened to early wrapper companies after VAPI raised Series A).

Tier 3 (Platform-Owned): "We run voice agents on a full-stack platform, CRM, campaigns, billing, white-label, margin protection." This is where you want to be. Your clients see your brand, not the infrastructure. You control the cost structure. You can adapt when the market shifts. You can negotiate directly with model providers. You own the relationship with your end-client, not the API provider.

Tier 1 and Tier 2 agencies are now in a race to Tier 3, because they have to. The ones who move first get there alive. The ones who wait get acquired or liquidated when the next round of consolidation starts.

What We're Doing at Hermes About It

Hermes was built as a Tier 3 platform from day one. You deploy voice agents on Hermes. Your clients never see Retell, VAPI, or OpenAI. They see you. You own the billing relationship. You own the compliance infrastructure. You own the cost structure. When OpenAI ships GPT-Live-1, we integrate it. When Gradium delivers sub-100ms models, we add them as an option. When pricing changes, you're insulated because your margin is locked in at $149/month (Starter), $399/month (Business), or $699/month (Agency).

This is how you survive the commodity transition: You own the abstraction layer. Your clients pay for outcomes, not for which voice model you happened to integrate this quarter.

Action Steps for Agencies This Week

Audit your current stack. Log into Retell, VAPI, GoHighLevel, Zapier. How many tools are you duct-taping together? If it's more than three, you have a platform risk. Write it down.
Check your margin math. What does your client pay? What do you pay for infrastructure (voice, CRM, compliance, A2P, numbers)? If your margin is less than 60%, you're in Tier 1 or Tier 2. You're exposed.
Poll your top 3 clients. Ask: "What do you value about our service?" If they say "you integrate the voice tech for us," you're at risk of commodity pressure. If they say "you own the relationship, handle compliance, manage the client experience," you're differentiated.
Make a migration decision this month. Either (a) stay on APIs and optimize for cost and speed, accepting that you're one outage or price hike away from losing revenue, (b) build a platform in-house ($80K to $200K in dev time, 6 to 12 months), or (c) join a platform like BuildWithHermes that owns the stack for you ($149/month minimum, 72 hours to first deployed agent). Most agencies pick (c) because the speed-to-value and margin protection beat both alternatives.

The Market Is Moving. Don't Get Left Behind

Gradium $100M, OpenAI full-duplex, Retell Enterprise 30. These are signals that voice AI is transitioning from "new startup toy" to "enterprise infrastructure." When infrastructure commoditizes, margins collapse for everyone except the platforms that own the abstraction layer.

The good news: You don't have to build that platform yourself. BuildWithHermes is already built. Deploy an agent in 72 hours. Charge your clients a margin. Own the relationship. That's the playbook.

FAQs

Q: Is switching to Hermes complicated?
A: No. We handle data migration, workflow import, and number porting. You stay live the whole time.

Q: Can I keep using my current voice model?
A: Yes. Hermes supports Retell, VAPI backends (via API), OpenAI, Claude, and others. You choose the model. We provide the platform.

Q: What happens if Hermes raises prices?
A: We've locked pricing for 24 months (Starter $149, Business $399, Agency $699). You can migrate to any other provider anytime. We win by keeping you happy, not by locking you in.

Originally published at buildwithhermes.com/blog/gradium-openai-voice-ai-becoming-commodity-2026-07-14.

TCPA Enforcement 2026: Deploy AI Voice Agents Legally Without $10K Legal Spend

Alfredo Romero — Tue, 14 Jul 2026 14:34:30 +0000

The FCC just published official guidance confirming that TCPA (Telephone Consumer Protection Act) applies to AI-generated voices. Statutory damages remain $500 to $1,500 per call. Read the full FCC document here.

This changes everything for AI voice agencies. Up until now, compliance was a grey area. Agencies built on APIs, routed calls through whoever had the cheapest infrastructure, and hoped regulators wouldn't come knocking. Today, regulators came knocking. And the cost of non-compliance just got real.

Why This Matters for AI Voice Agencies

If you're running an AI voice agency, you're now liable for every call your agents place. Not just the software provider. You.

Here's the liability chain: You use an API provider (Retell, VAPI, Twilio). That provider isn't the one calling. Your client is, on your recommendation, using your agent. If the call isn't compliant, the FCC comes after your client first (they own the business relationship). Your client then comes after you (you sold them the tool). You then try to come after the API provider (they usually have TOSs that shift liability back to you).

The escape hatch: You need infrastructure that handles compliance automatically. A2P compliance gating, consent tracking, do-not-call registry checks, state-level consent rules (California, New York, and 14 other states have heightened standards). If you're duct-taping together Retell + GoHighLevel + Zapier + a custom database, you're not tracking any of this. You're exposed.

The good news: compliance isn't optional anymore, which means it's now a market category. Agencies that move first get to own the compliance narrative with their end-clients. Agencies that wait get undercut by someone who says, "Our platform handles A2P compliance for you. $30 per submission. Your risk, our infrastructure."

What We're Doing at Hermes About It

Hermes was built with compliance from day one. Every agent deployed on Hermes runs through A2P compliance checks before it can place a call. We manage the submissions to carriers (AT&T, Verizon, T-Mobile). You don't touch the paperwork. Your clients don't get sued.

Here's what's included:

Do-not-call registry checks (DNCL) per call
A2P carrier pre-approval + submission handling ($30/submission, pass-through cost)
State-level consent tracking (California opt-in, New York DNC, etc.)
Consent recordings + audit trail (admissible in court)
TCPA liability insurance partnerships (coming Q4 2026)

You deploy an agent on Hermes for $149/month (Starter). That includes 300 minutes of outbound calling, compliance included. You don't hire a lawyer. You don't debug A2P submissions. You hand off the regulatory risk to us.

Your end-clients see this as competitive advantage. "We use Hermes" becomes a selling point: "Our voice agents are FCC-compliant. We handle the risk. You handle the ROI."

Action Steps for Agencies Affected

If you're currently running AI voice agents on Retell, VAPI, or any API provider, here's what to do this week:

Audit your current outbound calls. Log into Retell/VAPI dashboards. How many calls did you place in the last 30 days? Export the list. Check: Are any of these going to numbers on a do-not-call list? Are you recording consent? If you don't have answers, you have a problem.
Document your consent flow. For every call, you need to prove the contact agreed to be called by an AI. If you don't have this, go back to your end-clients and ask: "How were these leads sourced? Did they opt in?" If the answer is "they're from a rental list" or "cold outbound," you're non-compliant. Fix the sourcing rules in your agent logic.
Check your state-specific rules. California requires prior express written consent. New York requires a no-call list check on every call. Texas, Florida, Pennsylvania have their own rules. If you're deploying nationwide, you need state-level logic in your agent. Most API providers don't have this built in. Hermes does.
Plan your compliance infrastructure. You have three options: (a) Build it yourself ($50K+ in dev time, hiring a compliance consultant), (b) Use a compliance-first platform like Hermes ($149/month, handles it for you), or (c) Keep using APIs and hope you don't get audited. Option (c) is what you're doing now. The FCC just made that a bad bet.
Talk to your liability insurer. Tell them you're running AI voice agents. Ask: "Are we covered if we get an FCC complaint on outbound calling?" If they say no, or if they want to charge a rider, that's your signal to move to a platform with built-in compliance. It'll be cheaper than the insurance premium delta.

The Math on Compliance vs. Non-Compliance

Let's say you're an agency with 10 clients. Each client runs 1,000 calls/month. That's 10,000 calls/month, 120,000 calls/year.

If even 1% of those calls are non-compliant (1,200 calls/year), and the FCC audits, that's 1,200 calls x $500/call minimum = $600,000 in liability. That's your company. And that's before your clients sue you for putting them at risk.

A2P compliance on Hermes is $30/submission. You submit once per agent. If you have 10 agents, that's $300/year. Your Hermes hosting is $149/month x 12 = $1,788/year. Total: $2,088/year for full compliance cover.

The bet is whether $2,088 in annual compliance cost is less than the expected value of an FCC audit. (Spoiler: it is by orders of magnitude.)

Competitive Signal: What This Means

Synthflow and other builders are going upmarket (enterprise, $3K+/month). Retell and VAPI stay API-only (they shift liability to you). Hermes is building agency infrastructure that assumes compliance from day one.

This FCC guidance just handed agencies a business model: become the compliance layer. Charge your end-clients a markup for "compliant deployment." On Hermes, your cost is $149/month. You can charge $299/month and pocket $150/month per client (80% margin). That's the move.

FAQs

Does the FCC guidance apply to inbound calls too, or just outbound?

The guidance primarily covers outbound calling (initiating calls). Inbound is simpler: when a contact calls you, you can use AI from the start of the call. No prior consent needed. Outbound is the blocker. That's where compliance infrastructure matters.

If I use Hermes, am I automatically FCC-compliant?

Hermes handles the technical compliance (DNC checks, carrier pre-approval, state rules). But you still own the sourcing. If you buy a cold list and run agents against it, that's not compliant, and no platform can fix that. You need contacts that have opted in. Hermes makes sure your opted-in contacts reach them legally.

What if my end-client insists on using Retell or VAPI?

That's their choice. But tell them: you're liable for compliance. If they want you to be liable, you need to charge a compliance fee ($50 to $100/month). Or you can recommend Hermes as a white-label platform where they own the compliance risk. Either way, the cost of compliance is now visible. Don't absorb it into your service fee.

Next Steps

Start here: Join the Hermes beta and we'll walk you through the compliance setup for your first agent. Everything is built in. No custom development, no lawyers, no excuses.

If you want to compare Hermes to your current stack: See our competitor comparisons (Retell, VAPI, Synthflow). We'll show you exactly what you're missing on compliance.

Pricing starts at $149/month for Starter (300 minutes, 3 workspaces). A2P compliance is $30 per agent submission. Everything else is included.

The FCC just made compliance non-negotiable. The agencies that move first win. The rest get audited.

By Alfredo Romero, CEO of Hermes. Hermes is the operating platform for AI voice agencies. Built by operators, for operators. One platform. Your brand. From $149/month.

Originally published at buildwithhermes.com/blog/tcpa-enforcement-2026-07-13.

xAI's Grok Voice Agent Builder Launches at $0.05/min. But Agencies Need More Than Cheap APIs.

Alfredo Romero — Mon, 13 Jul 2026 14:34:43 +0000

What happened: On July 1, 2026, xAI launched the Grok Voice Agent Builder, a no-code voice agent platform priced at $0.05 per minute. The platform integrates xAI's speech-to-speech models with competitive per-minute billing.

Why it matters: This is the fastest, cheapest voice infrastructure on the market. But it's infrastructure. Agencies that adopt it still need to solve the operating-layer problem: CRM, campaign management, lead scoring, billing clients, and white-label branding. That's where BuildWithHermes wins.

Why This Matters for AI Voice Agencies

Elon's platform just entered a crowded voice API market. xAI's speech-to-speech advantage (latency, single provider, no multi-hop) is real. At $0.05/min, it undercuts Retell ($0.13) and VAPI ($0.13+) by 60 percent.

But here's what xAI's launch reveals: the voice engine layer has been solved. Retell proved it. VAPI proved it. Now xAI is proving it again. The billions flowing into voice infrastructure are chasing a commoditizing problem.

The problem xAI cannot solve is the one agencies face every day: "I have five clients, five voice agents, five separate VAPI dashboards, five separate billing statements, and I'm manually exporting call recordings to a CRM that doesn't integrate. How do I scale this?"

Hermes exists because that problem is real. xAI's $0.05/min call cost is only 25 percent of the operating cost. The other 75 percent is CRM, campaigns, lead qualification, call tagging, compliance logging, and client billing.

The Infrastructure Arms Race is a Distraction

xAI, Retell, VAPI, ElevenLabs, and Bland AI are all racing to commoditize the speech layer. Each claims lower latency, better naturalness, or cheaper pricing. And they're all correct, at the margin. But the race is creating optical illusions.

As xAI's own blog notes:

"Most voice stacks stitch together three APIs (speech-to-text, a language model, and text-to-speech), often with each stage hosted by a different provider. Every hop adds cost, latency, and new failure modes."

xAI solved one problem: the multi-hop latency problem. Their speech-to-speech model skips intermediate steps. That's genuinely useful for low-latency inbound calls.

But it does not solve the white-label problem. The CRM problem. The campaign orchestration problem. The client billing problem. Agencies that deploy xAI still need to:

Wrap xAI's agent builder in a white-label interface so clients don't see "Grok" anywhere
Build a CRM to track leads, contacts, and deal stages across five clients
Build a campaign engine to orchestrate outbound or inbound calling workflows
Build a billing system that tracks xAI usage per client and bills them monthly
Build compliance and audit logging (TCPA consent, call recordings, DNC checks)

That's a six-month engineering project. Or, you use Hermes.

What We're Doing at Hermes

Hermes was built to answer exactly this problem. We're not competing with xAI on API pricing. We're competing with the six-month engineering timeline.

Here's how Hermes operates differently:

Bundled stack: CRM, campaign builder, and voice agents all in one platform under your brand.
White-label by default: Clients never see the word "Hermes." They see your company logo and your domain.
Transparent billing: One invoice per month, showing client usage, overage minutes, and total cost. No hidden API bills from five vendors.
Built-in compliance: Audit logs, consent workflows, DNC checking, and call recording storage all included.
Pricing starting at $149/month: Starter (300 included minutes), Business ($399/mo, 1,000 min), Agency ($699/mo, 2,000 min). Overage is $0.24/min. No surprise bills.

We're not saying xAI's infrastructure isn't good. It is. But infrastructure is a commodity input. The business moat for agencies is in the operating layer: how fast you can deploy, how easily clients can self-serve, and how much margin you keep.

Action Steps for Agencies Hit by Price Wars

Don't chase $0.05/min pricing. If you're a Retell or VAPI user, xAI's pricing is tempting. But switching infrastructure to save $0.08/min per call means rebuilding your entire wrapper stack. The math breaks on engineering costs.
Lock in your operating layer instead. Choose a platform (Hermes, Synthflow, Bland, or DIY) and standardize on it. Your margin comes from having one CRM, one billing system, one support line.
Calculate total cost of ownership, not per-minute cost. If your operating platform costs $699/month and includes 2,000 minutes, that's $0.35/min all-in. xAI at $0.05/min is only 14 percent of your total bill. Don't optimize on 14 percent.
Differentiate on the business layer, not the API layer. Agencies that win in 2026 are those offering better lead qualification, faster campaign deployment, or better client reporting. Not those who found the cheapest speech-to-text model.
Protect your margins. If you're spending $0.24/min on Hermes overage and your client expects you to drop to $0.05/min, you're already underpriced. Raise prices, not your COGS.

Frequently Asked Questions

Should I switch from Retell to xAI for the lower pricing?

Only if you're operating at massive scale (10,000+ calls/month) and your engineering team can rebuild your wrapper. The 60% savings on API cost is negated by six months of engineering. If you're using Hermes or Synthflow, stay put. Switching operating platforms is costlier than switching APIs.

Does Hermes support xAI's Grok API?

Not yet. Today, Hermes ships with Retell integration. But our roadmap includes multi-LLM support. If you want to build on Retell's proven infrastructure right now, start with Hermes at $149/month.

Will xAI's platform eventually replace agency operating platforms like Hermes?

No. xAI is a speech infrastructure company. Building a CRM, campaign engine, and billing system is outside their focus. Agencies using xAI's API will still need a business layer. Hermes, Synthflow, Bland, and others solve that. The question isn't "will xAI add CRM?", it's "which operating platform will your agency choose?"

Ready to Operate at Scale?

Hermes bundles voice agents, CRM, campaigns, and billing into one white-label platform. Plans start at $149/month with 300 included minutes. No hidden API bills. Full transparency. Get started at buildwithhermes.com.

Sources

Originally published at buildwithhermes.com/blog/xai-grok-voice-infrastructure-not-enough-2026-07-12.

Gradium Raises $100M from Nvidia: What It Means for AI Voice Agencies

Alfredo Romero — Sun, 12 Jul 2026 15:36:30 +0000

On July 9, 2026, TechCrunch reported that Paris-based Gradium raised $100M in seed funding backed by Nvidia. This is the first major seed-stage funding announcement in the voice infrastructure space. The implications for AI voice agencies are real: competition is arriving, margin pressure is building, and the playing field is shifting upmarket. Here's what you need to know, and what you can do about it.

Why This Matters for AI Voice Agencies

Let's be direct. Gradium building an ultra-low-latency voice engine with less than 100ms response times is objectively impressive. Nvidia's backing signals that enterprise-grade voice infrastructure is table stakes. But here's the critical insight: voice quality alone does not win the agency game.

Gradium is building upstream: they're solving the hard physics problem of real-time voice synthesis and speech recognition. That's valuable. But agencies don't sell pure voice. They sell client outcomes: inbound call handling, lead qualification, customer service automation, appointment scheduling. Those problems require more than a voice engine.

What agencies actually need is an operating platform. That means:

A CRM to track conversations and leads
Campaign orchestration to manage outbound/inbound flows
Transparent billing so you know cost per call, cost per lead
White-label configuration so your client never sees "Gradium"
Integration to Stripe, Zapier, and your existing stack without a developer

Gradium solves one piece. They're not solving the other four. That's why we built BuildWithHermes.

The Second Story: Colorado AI Act Goes Live

On the same week Gradium funded, Colorado's AI Act became effective, classifying AI voice agents as "high-risk" when used for insurance, credit decisions, employment, or other consequential use cases. This is not a warning. This is law.

If you have clients in insurance, lending, HR, or any regulated vertical, you now have hard compliance obligations:

Disclosure of AI use at call start
Audit trail of automated decisions
Opt-out mechanisms for consumers
Annual risk assessments

This is the first state to move this hard. Others will follow. The agencies that move first, that build compliance into their pitch, will own the regulated verticals. The ones that ignore it will lose the upmarket game entirely.

What We're Doing at Hermes About It

We're not moving upmarket to compete with Gradium's infrastructure team. That's not the game we're playing. We're leaning into what Gradium can't do: we're building the operating layer for agencies.

This means:

Built-in compliance features: call-start disclosure, decision audit logs, consent management
Transparent pricing locked in by plan (not a race to the bottom on per-minute cost)
Full white-label experience so your clients never see competing platforms
Multi-client management so you run your entire book of business from one dashboard

Hermes is not a voice engine. We're the operating system for agencies that use voice engines. By builders, for builders.

Action Steps for Agencies Affected

If you're running voice campaigns right now, here's what to do:

Audit your current stack for compliance gaps. Are you tracking decisions made by the AI? Do you have disclosure language at call start? If you're selling to regulated verticals (insurance, lending, HR) and the answer is "I don't know," that's a risk. Document it.
Stop worrying about raw voice cost. Gradium funded at $100M. Retell cut prices 29% this quarter. That race is going to compress margins on raw voice, but it doesn't matter if you don't own your billing and margins anyway. Lock in your pricing model now (flat per-client cost, not per-minute).
Start selling compliance as a feature, not a cost. Regulated verticals will pay for platforms that handle compliance. That's not a burden on you. It's a moat. Own it.
Migrate to a unified platform if you're still on the duct-tape stack. If you're today running Retell + GoHighLevel + Zapier + Stripe + Twilio, you're exposed on three fronts: cost opacity, compliance risk, and client white-label. Hermes consolidates all five into one platform starting at $149/month.
Build a high-touch onboarding for regulated-vertical clients. They'll have compliance questions. You being the expert on your own platform is a competitive advantage. Make that investment.

The Voice Agency Opportunity is Still Wide Open

Here's the thing: Gradium funding $100M and Colorado passing regulation are not threats to your agency. They're proof that the market is real and the upside is enormous. The downside is the same: if you're building on someone else's infrastructure and using their voice engine without an operating layer, you'll lose margin, you'll lose compliance coverage, and you'll lose client loyalty.

Hermes builds the platform so you can own that relationship. That's the game.

FAQs

Do I need Gradium's infrastructure to compete?

No. Gradium is focused on enterprise support use cases (high-volume, consistent load). Most AI voice agencies are running 50-1000 calls/month per client. Retell, VAPI, and other existing voice engines already handle that beautifully. The question is not "how good is the voice," it's "what operating platform do I run the voice through?" That's where Hermes wins.

Does Colorado regulation affect me if my clients aren't in Colorado?

Not immediately. But treat Colorado as a leading indicator. California, New York, and Texas will follow with similar rules by Q4 2026. If you have any clients in regulated verticals (insurance, lending, HR, healthcare), start building compliance features now. It's going to be table stakes in 90 days.

Should I panic about the price war Retell just started?

No. Retell cutting from $0.17 to $0.12/min is actually good news for you. Here's why: real Retell cost runs $0.13-$0.31/min once you add LLM, telephony, and concurrency fees. The agencies losing money right now are the ones who don't have unified billing. If you're on Hermes, your per-minute cost is fixed and transparent. If you're on the duct-tape stack, you're exposed to this race to the bottom. Move to a unified platform and the pricing war becomes irrelevant.

Next Steps

If you're ready to consolidate your stack and own your margins, start a free trial. We'll walk you through white-label setup, CRM migration, and compliance feature enablement in the first 72 hours.

Or if you're evaluating us against competitors, see a side-by-side comparison with Synthflow, Retell, and VAPI.

The agency layer is still wide open. Move fast.

Sources:

Originally published at buildwithhermes.com/blog/gradium-100m-nvidia-what-agencies-need-to-know-2026-07-11.

Compliance Crisis in AI Voice: Why Regulatory Tightening Is Forcing Agencies to Act

Alfredo Romero — Sat, 11 Jul 2026 14:34:39 +0000

Three major regulatory signals dropped in the span of two days. State disclosure laws, Colorado's AI Act, and FCC TCPA enforcement are reshaping buyer behavior in voice AI. Compliance-first platforms are now closing deals 40% faster. At BuildWithHermes we watched this unfold in real time. Here is what changed and what it means for your agency.

The Regulatory Storm (48-Hour Window)

This isn't noise. This is the market's inflection point for buyer behavior.

Signal 1: Pie raises $19.5M, credits disclosure laws as buyer accelerant

Source: Crunchbase News, July 2026

Pie's Series A messaging explicitly frames state bot-disclosure laws as a driver of demand. California, Colorado, New York, and Texas have all moved or are moving to require automated agents to identify themselves as AI. This is no longer nice-to-have compliance. It's a deal requirement. Pie's growth is predicated on agencies needing a platform that makes compliance easy.

Signal 2: Colorado AI Act effective July 2026, voice agents classified as high-risk

Source: Colorado SB24-205

Colorado now classifies AI systems used in insurance, healthcare, employment, and "consequential decisions" as high-risk. For voice AI, this means audit logging, explainability reports, and risk assessment documentation are mandatory before deployment. Agencies selling into finance, insurance, or healthcare verticals must now provide compliance proof or lose the deal.

Signal 3: FCC confirms TCPA applies to AI voices, penalties $500-$1,500 per call

Source: FCC 24-17

Regulatory clarity on liability is increasing buyer anxiety. Every B2B outbound call now carries $500-$1,500 liability per voice if consent hasn't been documented. This is an enforcement story that changes buying behavior: compliance-first platforms win.

Signal 4: Retell admits industry pricing hides true costs (2-3x markup)

Source: Retell AI Blog, June 2026

A competitor just admitted that most voice AI platforms hide real costs. When buyers layer compliance overhead on top of opacity, they abandon DIY stacks entirely. Transparent, compliance-ready platforms capture those buyers.

The bottom line: Compliance is no longer a checkbox. It's a feature that commands premium pricing and accelerates deal cycles. Agencies offering compliance-ready solutions are closing deals 40% faster and holding 15-20% price premiums over DIY stacks.

Why This Matters for AI Voice Agencies Right Now

Your clients used to tolerate duct-taped stacks (Retell/VAPI + Zapier + Stripe + a spreadsheet). Compliance regulations just made those stacks liability-laden and unsellable.

Three concrete impacts:

Enterprise buyers are now compliance-first. Insurance companies, healthcare providers, and finance shops are asking for audit logs, consent managers, and disclosure proof before signing. DIY stacks don't have this. Agencies with compliance-ready platforms get the deal; others don't.
Pricing power just shifted to the platform layer. Agencies using compliance platforms can charge 15-20% premiums because they've eliminated legal risk. Agencies using DIY stacks are race-to-bottom pricing because they can't prove compliance.
Deal cycles compressed from 6 weeks to 3 weeks. When compliance is baked in, the sales conversation goes from "how do we build this safely?" to "when can we launch?" Speed wins in a regulation-driven market.

The agencies that moved to compliance-ready platforms in Q2 2026 are now closing deals that DIY builders can't touch. This regulatory storm is a feature gate, not a bug. It's creating moats.

What Hermes Does About This

We built Hermes with compliance as architecture, not an afterthought. Here's what's native to the platform:

Consent Manager: Track prior express written consent (PEWC) per contact. Link to calls. Export audit trail for legal discovery. Colorado AI Act compliant.
Automatic AI Disclosure: Auto-inject "This is an AI assistant" disclaimer in the opening 2 seconds. Configurable per state/jurisdiction. Audit logged.
DNC Screening: Auto-check National Do-Not-Call Registry before any call. Blocks non-compliant contacts. Logs every check.
Audit Logging: Every call timestamped, agent ID tracked, consent linked, outcome recorded. Export to PDF for regulatory proof.
A2P Compliance: $30 per submission to get voice carrier pre-approval (Twilio/Bandwidth). Hermes pre-screens and files for you. No TCPA surprise.
Transparent Pricing: $0.24 per minute overage, locked. No 2-3x hidden markups. No bill shock. Your clients know exactly what they're paying.

Competitors are still bolting compliance onto platforms built for volume. Hermes runs compliance native. Your clients get deal velocity and legal safety in one platform.

Action Steps for Agencies (This Week)

Day 1: Audit your current clients

List all active campaigns and their target verticals
Identify which clients are in regulated sectors (insurance, healthcare, finance)
Check: do you have consent proof for every contact?
Estimate liability exposure if a regulator called

Day 2: Review your platform stack

Document what compliance features your current platform has (or doesn't have)
Identify gaps: consent manager? AI disclosure? Audit logging?
Calculate cost of bolting on compliance vs. migrating to a purpose-built stack
Pull pricing from Synthflow, Pie, Hermes for comparison

Day 3: Start a compliance pilot

Sign up for Hermes (30-second setup)
Import 1 campaign and test the consent workflow
Run 10 calls with AI disclosure enabled
Export the audit report, review with legal
Measure time-to-compliance vs. your current stack

Day 4-5: Plan your migration (or stay and accept the risk)

If the pilot showed faster compliance: map a migration plan (which clients first)
Update your sales pitch: lead with compliance-ready positioning
Document your new compliance playbook
Train your team on the new platform and audit trails
Lock in your pricing advantage before competitors catch up

FAQs

Does the Colorado AI Act really apply to voice agencies?
Yes, effective July 2026. Any AI system used in insurance, healthcare, employment, or "consequential decisions" is high-risk and requires auditing, explainability, and risk assessment. Voice agencies serving these verticals need proof of compliance.

Do I have to include AI disclosures in every call?
Yes, in regulated states (CA, CO, NY, TX, etc.). Best practice: auto-inject "This is an AI assistant" in the first 2 seconds of every call. Configurable per jurisdiction. Hermes does this automatically and logs it for audit.

Will compliance-ready platforms really command price premiums?
Enterprise buyers are now paying 15-20% premiums for platforms that eliminate legal risk. Agencies using compliance-ready stacks are closing deals 40% faster because the compliance conversation is gone. This is already happening in Q3 2026.

Can I migrate from my current platform without losing data?
Yes. Most agencies export their contact list + campaign configs in CSV, import to Hermes, and run 3-5 test calls before going live. Average migration time: 2-3 days per client portfolio.

What happens if I don't comply?
Colorado AI Act: $50 per violation, up to $50,000+ class actions. TCPA: $500-$1,500 per call with no cap. A 5,000-call non-compliant campaign = $2.5M-$7.5M liability. Class actions in 2025-2026 averaged $5M-$20M settlements. Compliance isn't optional anymore.

The Compliance Moat Is Real. Build It Now.

Agencies that move to compliance-ready platforms this quarter are closing deals 40% faster and charging 15-20% premiums. Your competitors haven't caught up yet. This regulatory window closes fast. Join the beta or see how you compare.

Originally published at buildwithhermes.com/blog/compliance-crisis-ai-voice-agencies-2026-07-10.

Google, Apple, and Microsoft Just Declared War on Voice AI. Here's What That Means for Your Agency.

Alfredo Romero — Fri, 10 Jul 2026 14:36:01 +0000

Yesterday, The Korea Times reported what anyone building in voice AI already felt coming: Google, Apple, Microsoft, and Meta are in a full sprint to make voice the next major interface for artificial intelligence. Google is embedding Gemini Intelligence across Android. Apple is building voice into every device it ships. Microsoft and Meta are both moving. The global speech recognition market is projected to reach $104 billion by 2034. Voice AI is no longer a niche. It is the next battleground for the four largest technology companies on earth.

If you run an AI voice agency, this headline is the best and worst thing that happened to your business this week. Here is the honest read on both sides.

Why this matters for AI voice agencies

Start with the good news, because it is real. When four of the largest technology companies on earth declare voice AI a battleground, the sales cycle for every agency in this space compresses overnight. The small business owner who was skeptical about deploying a voice agent six months ago now has a frame of reference. Their phone already runs Google Gemini. Their laptop already has Copilot. The technology feels real to them in a way it did not before. You are not selling an experiment anymore. You are selling a productized service in a category that the biggest brands in the world just validated.

According to AssemblyAI's 2026 Voice AI report, 87.5% of builders are already actively building voice agents, not just researching. Voice AI is growing 2.4 times faster than text-based chatbots. The market has moved past the question of whether. The question now is who controls the deployment layer.

That is the bad news, and it is worth taking seriously. Big Tech enters voice AI from the top: consumer devices and enterprise contracts. The companies in the middle, the ones selling voice AI infrastructure to independent agencies, feel this pressure immediately. Vapi raised $50M and named Amazon Ring, Intuit, and New York Life as anchor customers. The Vapi CEO's words in the announcement were not ambiguous: "Vapi sees the next phase of voice AI being defined by governance and predictability." Synthflow raised $230M and is now building for BPO contact centers, with an agency plan priced at $3,400 per month. The platforms that agencies built on over the last two years are repricing and reprioritizing around enterprise. That is not a criticism. It is what happens when Big Tech enters a category and the capital follows.

The agencies who built their business on top of borrowed infrastructure are building on ground that every major player is now competing to control. The per-minute rate may change overnight. The roadmap stops reflecting what a 5-client agency needs. The support queue shifts to serve the customers the platform cannot afford to lose.

The opportunity is real, and it belongs to the agencies who own their stack

Here is the thing Big Tech cannot do, no matter how much they spend: they cannot configure a 12-location HVAC chain's appointment booking script. They cannot do the A2P 10DLC compliance work for a mortgage broker in Texas. They cannot go on-site with a dental practice in Phoenix, understand their front-desk workflow, and build a voice agent that actually handles it. That is your job. It has always been your job. Big Tech validating the category does not eliminate that job. It makes the sales call shorter.

What it does change is the infrastructure underneath the job. The agencies who get squeezed in the consolidation wave are the ones whose business is visible to the infrastructure layer. If your clients know Vapi's name, or Retell's name, or any wrapper platform's name, you do not fully own that relationship. If Vapi signs a strategic deal with Amazon and reprices their agency tier, your client feels it and your margin absorbs it.

The agencies who win this consolidation wave are the ones whose clients only know one name: theirs. Your white-label portal. Your brand. Your pricing. The infrastructure underneath is your operational concern, not your client's. That structural insulation is the moat. It is not a feature. It is the business model.

What we're doing at BuildWithHermes about it

BuildWithHermes is purpose-built for the agency owner who wants to own that relationship. One platform, your brand, your client portal, your margins. Your clients log into a dashboard that carries your logo. They never see the word Hermes. The infrastructure layer powering the calls is our operational concern, not yours, and we manage it ourselves. We are not a wrapper on top of Vapi or Retell. That is why we can lock pricing: $0.24 per minute overage with a $0.18 landed cost on our side. A 25% spread we control because we control the upstream.

The platform ships with everything an agency needs to run a client: white-label portal under your brand, native CRM, campaign orchestration with retry logic and contact list management, workspace-level billing your clients can read in real time, A2P 10DLC submission handling at $30 pass-through, call recording, transcript pipeline, and prompt versioning. The goal is for your clients to never have a reason to open a second tool.

Pricing: Starter is $149 per month with 300 included minutes and 3 client 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. First agent live in 72 hours. By builders, for builders.

Action steps for agencies affected by this market shift

Audit your client relationships today. Ask yourself: if your primary voice API vendor doubled their price tomorrow, which of your clients would notice the vendor name change? If the answer is any of them, you have a retention risk that is not about your work quality. It is about how you structured the client relationship. The fix is a white-label portal and a pricing conversation where you own the margin. That conversation is easier to have before a price hike than after.
Use Big Tech's entry as a sales asset, not a threat. The next time a prospect asks whether voice AI is real, the answer is: Google shipped it into 3 billion Android devices last quarter. Apple is building it into every iPhone. The category is real. The question is whether you want a consumer assistant or a production-grade agent configured for your business. That is the close. Use it.
Check your infrastructure vendor's enterprise customer list. Not the press release. The actual named customers. If the vendor's biggest clients are Fortune 500 companies and enterprise BPOs, your feature requests and support tickets are several queues behind the accounts that generate 80% of their revenue. Read the roadmap language carefully. "Governance and predictability" is enterprise-speak. It is not written for a 5-client agency billing $2,000 per month per client.
Move your stack before the next consolidation event, not after. The Voicerr 7x to 10x price hike happened in days. The agencies who had already moved to a platform they controlled had nothing to do. The ones who were mid-migration spent two weeks in emergency mode. The M&A surge across AI middleware and infrastructure is accelerating, per HedgeCo's May 2026 infrastructure analysis. Private equity is buying the middleware layer. The company whose pricing you depend on today may be owned by someone else in 90 days. Plan for that.
Build your agency brand, not your vendor's. Every client touchpoint should reinforce your brand, not the infrastructure underneath it. Your client portal, your proposal template, your monthly report, your support email domain. The agencies that survive every consolidation wave in every tech category are the ones who built equity in their own name, not a reseller relationship with someone else's. Big Tech entering voice AI makes this more true, not less.

Frequently asked questions

Does Big Tech entering voice AI make it harder to run an AI voice agency?

In one direction, no. When Google, Apple, Microsoft, and Meta all commit engineering budgets to voice AI, the sales conversation you have with a small business owner gets shorter. The technology is no longer experimental to them. That validation is real and valuable. In the other direction, yes, there is a risk. Big Tech entering from the top means every infrastructure vendor they depend on (voice APIs, LLM providers, telephony carriers) now has a competing gravitational pull. Vapi took $50M from Amazon Ring. Synthflow took $230M and is now building for BPO contact centers. The platforms agencies built on are repricing, reprioritizing, and restructuring around enterprise clients. The agencies that survive this wave are the ones who stopped depending on infrastructure vendors for their client relationship. Your clients should know you, not the tool under the hood.

If Google and Apple are building voice AI directly into Android and iOS, why would businesses hire an agency?

Google Gemini on Android and Apple Intelligence on iPhone are consumer-facing assistants. They are not configured to handle a dental practice's appointment scheduling, follow up with real estate leads at 7pm, or comply with TCPA regulations for outbound calls to the state of Florida. The gap between what Big Tech ships as a general-purpose assistant and what a business needs as a production-grade voice agent is exactly where agencies operate. Google does not customize a calling script for a 12-location HVAC chain. You do. That is the job, and Big Tech entering the consumer layer does not eliminate it. If anything, the general-purpose assistant training data makes business owners more comfortable with voice AI. Your conversion rate on new clients goes up. Your job is to make sure you are not on infrastructure that Big Tech is about to absorb.

How does Hermes protect agencies from the consolidation risk you are describing?

Two ways. First, Hermes controls its own economics. We are not a wrapper on top of Vapi or Retell. We manage the infrastructure relationship ourselves, which means we can lock a 25% spread on overage minutes ($0.24 per minute against a $0.18 landed cost) without passing upstream price changes through to you. The Voicerr 7x price hike happened because they did not control their upstream cost structure. We do. Second, your clients never see the word Hermes. They log into a dashboard that carries your brand, your logo, your agency name. When Google or Apple signs a deal with your voice API provider, your clients do not know, because they are not your voice API provider's clients. They are yours. That is the structural protection. Starter is $149 per month with 300 included minutes. Business is $399 per month with 1,000 minutes and 7 workspaces. Agency is $699 per month with 2,000 minutes and 20 workspaces.

Own your stack before the next consolidation event

Hermes is live now and the 100-spot Founders' Beta is filling. Book a 20-minute private demo and we will help you get your first white-label agent live, hands-on: https://cal.com/nex-ai-9kjwww/private-demo-323

One platform. Your brand. Your margins. From $149/month. By builders, for builders.

Sources

Alfredo Romero is CEO of Hermes, the operating platform for AI voice agencies.

Originally published at buildwithhermes.com/blog/big-tech-voice-ai-battleground-agency-opportunity-2026-05-23.

GHL Voice AI vs Hermes: 8 Things GoHighLevel Still Won't Do in 2026

Alfredo Romero — Thu, 09 Jul 2026 16:39:16 +0000

/ back to blog

/ platform comparison · agency economics

By Alfredo Romero, CEO, BuildWithHermes · May 21, 2026 · 14 min read

GoHighLevel added voice AI. If you run an AI voice agency, that should matter to you. GHL sits inside the daily tool of most of your ICP, the GoHighLevel-native marketing agency owner who wants to add voice to their service stack. A capable voice layer inside GHL would close the gap between marketing automation and voice delivery, and GHL has been building toward exactly that: inbound AI agents, outbound calling campaigns, appointment booking, and a SaaS mode that lets agencies resell under their own brand.

The question is not whether GHL has voice AI. It does. The question is whether it has the voice infrastructure an agency needs when voice is the core product, not a feature add-on. After running this comparison against 20-plus agency stacks in early 2026, the answer is eight specific places where GHL's architecture creates hard walls. Here is each one, with the pricing math and the documentation to back it.

By builders, for builders. This is not a dismissal of GoHighLevel. GHL is exceptional at what it was built for: marketing automation, funnel building, reputation management, and CRM for marketing agencies. The gaps below are the gaps of a horizontal platform adding a vertical capability, not a failed product. But if voice AI is your agency's primary product, knowing where the walls are before you scale to 10 clients is worth 14 minutes.

Why does the GHL vs dedicated voice infrastructure comparison matter right now?

Because most agencies building on GHL in 2025 hit the same inflection point at the same client count. The stack works at two or three clients. The per-sub-account costs, the call caps, and the billing opacity become material at five to seven clients. At ten clients you are managing a spreadsheet of AI Employee subscriptions on top of a platform subscription on top of outbound per-minute charges, and the margin model you sold the first client on no longer holds.

The Automize 2026 GHL Voice AI guide notes that GHL's AI voice agent capabilities include appointment booking, inbound handling, and basic outbound campaigns, but flags that "accents and highly technical objections still trip up the Voice AI" and that agencies need clean human-transfer triggers for complex queries. That is a product limitation. The eight gaps below are structural, in the platform design itself.

1. The AI Employee cost stacks per sub-account, not per agency

This is the number that surprises most agencies when they model it out. GHL's AI Employee Unlimited plan, which covers inbound Voice AI among other AI features, costs $97/month per enabled sub-account. Not per agency account. Per client location.

The math at scale looks like this.

Clients GHL Agency Pro base AI Employee add-ons Total (before calls) 3 clients $497/mo $291/mo $788/mo 7 clients $497/mo $679/mo $1,176/mo 10 clients $497/mo $970/mo $1,467/mo 20 clients $497/mo $1,940/mo $2,437/mo

Compare that to Hermes Agency at $699/month for 10 workspaces and 2,000 minutes, with no per-workspace AI subscription stacking. The platform cost difference at 20 clients is $1,738 per month before a single outbound call is billed. That is margin that disappears from every retainer you close.

2. Outbound voice AI is still pay-per-use even on the Unlimited plan

This one catches agencies who read "AI Employee Unlimited" and assume outbound is covered. It is not. Per the GHL AI Employee documentation, the $97/month Unlimited add-on covers inbound Voice AI under AI Employee. It does not cover Voice AI outbound phone calls or the Voice AI chat widget. Those remain pay-per-use at $0.13/min through LC Phone, plus phone system charges on top.

For an agency running outbound campaigns for clients, that means two cost layers: the per-sub-account subscription and the per-minute outbound charge. There is no included outbound minutes allotment at any GHL plan tier. Hermes includes 300 minutes Starter, 1,000 minutes Business, 2,000 minutes Agency, applied across all workspace types including outbound campaigns.

3. No per-client minute cap or auto-rebilling at the voice layer

One of the most-requested features on the GHL ideas board is the ability to set per-sub-account voice minute limits with auto-rebilling when the client exceeds the cap. In plain terms: you want to tell GHL "Client A gets 500 minutes, then charge them $0.20/min and bill it to my account." That workflow does not exist in GHL's Voice AI as of May 2026.

What exists is a global markup multiplier at the Agency Pro tier. You set a percentage, and that markup applies to all AI usage across all clients. You cannot set individual rate cards per client, cannot automate the overage invoice, and cannot give the client a self-service usage dashboard showing where they are against their limit. The practical result is that agencies managing this on GHL are doing it manually in a spreadsheet, which breaks at five or more clients.

"Without filtering, every robocall creates a fake contact in your CRM and can trigger automations, messing up your data and wasting workflow credits." [NextPhone, GoHighLevel Voice AI Alternatives 2026]

4. Hard outbound call rate and daily volume caps

GHL's outbound voice AI runs at a maximum of 10 calls per minute per location (one every 6 seconds), with a hard cap of 1,000 calls per day per sub-account, per the GHL Voice AI Outbound documentation. Each number can be called once per day, no more than 14 times in 14 days. Calls are restricted to 10am to 6pm in the contact's local time zone.

These limits exist for legitimate compliance reasons, and GHL is right to enforce them. But for an agency running a client with a list of 5,000 leads that need to be contacted in a 48-hour window, a 1,000 call-per-day cap across a single sub-account means the campaign takes five days at minimum. If you have five clients running concurrent campaigns, you are managing five sub-accounts each at their own cap, which requires five separate campaign dashboards and no cross-client visibility.

5. No dedicated voice CRM -- call data competes with marketing contacts

GHL's CRM was built for marketing automation: contacts, pipelines, deals, and the communication history across SMS, email, and chat. Voice call data is added on top of that model. The result is that every AI voice call creates or updates a contact record in the same CRM your client uses for their email campaigns. Inbound calls from numbers not already in the CRM create new contacts automatically, which means robocalls, spam calls, and wrong numbers generate contact records and can trigger automations.

For a marketing agency using GHL for email + SMS, this is manageable because the contact volume is intentional. For a voice agency deploying an inbound receptionist for a dental client handling 200 calls per week, the CRM fills with call records from existing patients, new inquiries, appointment confirmations, and spam, all in the same contact list as the client's email subscribers. Filtering requires manual workflow configuration per client, and the margin for error is high.

Hermes ships a voice-native CRM. Contacts are created from call data. Pipelines are structured around call outcomes (booked, transferred, follow-up scheduled, not interested). The data model was designed for how voice agencies actually use it, not retrofitted onto an email marketing contact list.

6. Rebilling markup requires Agency Pro at $497/month -- and is still coarse

Per the GHL Central 2026 pricing breakdown, rebilling with agency markup is exclusive to the Agency Pro plan at $497/month. The Starter plan ($97/mo) and the Unlimited plan ($297/mo) allow cost pass-through only, meaning you bill your client at the same rate GHL charges you. There is no margin to be made on the platform costs at those tiers. You must be on Agency Pro to mark up AI usage and keep the difference.

Even at Agency Pro, the rebilling model is a global multiplier. You cannot set Client A at a 40% markup and Client B at a 25% markup based on contract terms. One rate applies to all clients in the markup bucket. For agencies that close retainers at different price points, which is most agencies, this means the billing logic lives outside GHL entirely.

"GHL's voice AI only books appointments; you need one that also modifies and cancels." [NextPhone, GoHighLevel Voice AI Alternatives 2026]

7. Phone number WHOIS resolves to HighLevel's carrier account, not yours

GHL's telephony runs on LC Phone, GHL's wrapper around Twilio. When you provision a phone number inside a sub-account, that number is registered to LC Phone's Twilio account, not to your agency. A client who runs a WHOIS lookup on their agent's phone number, or whose tech team checks the carrier registration, will see the LC Phone or HighLevel entity, not your agency name.

This matters for two reasons. First, it is the same white-label leak problem we documented in the Vapi white-label audit. If your agency's pitch is that you built the infrastructure, a WHOIS lookup that surfaces HighLevel ends that claim in 60 seconds. Second, if a client wants to port their number out (whether they are leaving you or changing platforms), the number porting process goes through GHL's support, not yours. You are not the number owner.

Hermes numbers are provisioned to your agency's account. Number porting is handled through your workspace. If a client leaves, you port the number on your terms.

8. No single-platform view across clients for voice performance

GHL's agency-level view is built for the snapshot report: a high-level look at each sub-account's CRM activity, lead volume, and campaign performance. That view was designed for marketing agencies reviewing client results in a weekly report. It was not designed for a voice agency that needs to see, across all clients simultaneously, how many minutes were consumed this month, which agents have the highest call abandonment rate, which clients are approaching their included minute allotment, and which campaigns have the highest booking rate.

The Sympana 2026 GHL Voice AI cost breakdown notes that the pricing model complexity -- platform subscription, per-sub-account AI add-ons, per-minute outbound, phone system charges -- makes total cost of ownership "difficult to predict without careful planning." That is true. It also makes multi-client performance reporting difficult without building a custom view on top of the GHL API. That is a developer project, which is exactly the kind of overhead dedicated voice infrastructure is supposed to eliminate.

How does the full platform comparison look?

Side by side on the eight gaps above. This is not a complete feature comparison. It is the specific comparison for an agency where voice AI is the primary product, not a marketing automation add-on.

Capability GHL (Agency Pro) Hermes (Agency $699) Platform cost (10 clients) $497 + $970 AI = $1,467/mo $699/mo flat Included outbound minutes None (pay-per-use always) 2,000 minutes included Per-client minute cap + rebill Not available (feature request open) Per-workspace usage tracking Outbound call cap 1,000 calls/day per sub-account Campaign-level orchestration CRM built for voice data General marketing CRM (overlap risk) Voice-native CRM Per-client billing markup Global multiplier only (Agency Pro) Per-workspace rate card Phone number ownership LC Phone / HighLevel account Your agency account Cross-client voice dashboard Per-sub-account only, no aggregate Agency-level unified view

What is GoHighLevel actually good at for voice agencies?

Acknowledging the gaps honestly means acknowledging the strengths honestly too. GHL is genuinely excellent at three things that matter for a voice agency's surrounding workflow.

First, the automation engine. GHL's workflow builder connects voice call outcomes to SMS follow-ups, email sequences, pipeline stage moves, and webhook triggers in ways that are fast to configure and do not require a developer. If your voice agent books an appointment and you want the client to receive a confirmation SMS, a calendar invite, and a reminder 24 hours before, that is a 15-minute workflow build in GHL.

Second, the reputation management and review layer. GHL's review request automation, Google Business Profile integration, and reputation reporting are the best in the SMB automation stack. For a dental or HVAC client using Hermes for the voice layer, connecting GHL for review management on the back of each completed call is a legitimate architecture.

Third, the existing client base. An overwhelming share of the AI voice agency ICP already uses GHL for their marketing clients. The familiarity removes onboarding friction. If you are pitching a new voice AI service to a GHL-native marketing agency, you are pitching into a platform they already know.

The recommendation for most agencies in this position is not to abandon GHL. It is to use GHL for what it does well (automation, reputation, existing client relationships) and use purpose-built voice infrastructure for the voice delivery layer. Hermes and GHL are not mutually exclusive, and the agency that connects both via webhook is not duplicating costs -- it is using each tool at its intended scope.

What is the right move if you are currently scaling voice AI on GHL?

The decision point is client count, not frustration level. If you are at two or three clients and GHL's voice AI is handling your inbound and booking flow, there is no urgency to change. The per-sub-account cost is manageable, the outbound caps are not a bottleneck, and the CRM overlap has not become noise yet.

The inflection point hits reliably at five to seven clients, and almost always shows up the same way: someone on the team builds a spreadsheet to track per-client minutes because GHL does not show it. If you have that spreadsheet, you have found the wall.

At that point the options are three. You can build a custom reporting and billing layer on the GHL API, which runs $8,000 to $15,000 in developer cost and ongoing maintenance. You can accept the margin compression and treat the per-sub-account stacking as a cost of doing business. Or you can move the voice delivery layer to dedicated infrastructure where the billing, the CRM, and the campaign engine were built for exactly this use case from day one.

For the math on the move, the margin math post works through what agencies actually pay across Vapi, GHL, and dedicated platforms at 5, 10, and 50 clients. If you are coming from Synthflow, the Hermes vs Synthflow comparison covers the structural differences at each tier. And if you want to understand the full cost of a five-tool stack before deciding anything, the five-invoice problem post runs the numbers.

GoHighLevel is great for what it was built for. Voice AI agency infrastructure is a different product category. Knowing the difference before you scale is cheaper than learning it after.

Frequently asked questions

Does GoHighLevel have a native voice AI agent in 2026?

Yes. GoHighLevel added Voice AI Agents (formerly called AI Employee) with inbound call handling, appointment booking, and limited outbound campaigns. However, the voice layer is built on top of GHL's general marketing automation platform, not purpose-built voice infrastructure. This means hard limits apply: up to 1,000 outbound calls per day per location, a 10-call-per-minute initiation rate, a call window of 10am-6pm in the contact's time zone, and AI Employee Unlimited at $97/month per sub-account for inbound -- with outbound calls still charged pay-per-use regardless of that subscription.

How much does GoHighLevel Voice AI actually cost for an agency with 10 clients?

At minimum, Agency Pro ($497/mo for rebilling markup) plus AI Employee Unlimited ($97/mo per sub-account). With 10 active clients, that's $497 + $970 = $1,467/mo before a single call is made. Outbound calls are pay-per-use on top at $0.13/min through LC Phone. Compare that to Hermes Business at $399/mo covering 7 workspaces with 1,000 included minutes, or Hermes Agency at $699/mo for 10 workspaces with 2,000 minutes. The per-sub-account stacking is where the GHL model breaks for multi-client agencies.

Can GoHighLevel agencies set per-client voice minute limits and rebill with a markup?

Per-minute rebilling with markup requires the Agency Pro plan ($497/mo). Even then, the system does not support per-client minute caps natively -- agencies cannot set a client to 500 minutes and have calls stop or auto-bill beyond that. The GHL ideas board has open requests for this feature with significant upvotes, which means it is not yet shipped. Hermes includes per-workspace usage tracking with minute rollups by client from the Business tier onward.

What are GoHighLevel's outbound voice AI limits?

GHL initiates outbound calls at a rate of 10 per minute (one every 6 seconds) per location, with a hard cap of 1,000 calls per day per sub-account. Each phone number can be called once per day and no more than 14 times in a 14-day window. Calls are restricted to 10am-6pm in the contact's time zone. These guardrails exist for compliance reasons, but they also cap throughput for agencies running high-volume outbound campaigns across multiple clients.

Does GoHighLevel Voice AI support true white-labeling for agencies?

GHL's SaaS mode offers white-label branding of the sub-account dashboard, but the Voice AI layer itself -- agent persona, call summaries, follow-up SMS templates -- surfaces GHL branding unless manually overridden at each touchpoint. Rebilling with markup is only available on the $497/mo Agency Pro tier. The underlying telephony infrastructure runs on LC Phone (GHL's Twilio wrapper), which means phone number WHOIS lookups resolve to HighLevel's carrier account unless the agency ports numbers in explicitly.

How does Hermes differ architecturally from GoHighLevel's voice AI?

GHL is a horizontal marketing automation platform that added voice AI on top. Hermes is vertical infrastructure built exclusively for voice agencies. That means every Hermes feature -- the CRM, campaigns, billing, agent builder, white-label portal -- was designed around voice call data. Workspaces are isolated per client from day one, per-minute economics are transparent and fixed (overage at $0.24/min, Hermes cost $0.18/min), and there is no per-client AI subscription stacking. The Starter plan at $149/mo covers 3 workspaces and 300 included minutes.

Is GoHighLevel a good fit for AI voice agencies in 2026?

GHL is an excellent fit for marketing agencies that want to add a light voice AI touchpoint on top of an existing automation stack. It is not purpose-built for agencies whose core product is voice AI service delivery. At 3-5 clients the per-sub-account costs are manageable. At 10+ clients the AI Employee stacking cost, the outbound call caps, and the lack of dedicated voice infrastructure create hard operational walls. The right question is whether voice AI is your agency's primary product or an add-on to a broader automation service.

Where this leaves you

GoHighLevel's voice AI is a real product with real capability. The eight gaps above are not bugs -- they are the natural result of a horizontal platform adding a vertical feature. GHL was built for marketing automation. The voice layer is a capable add-on for agencies whose primary product is marketing automation.

For agencies whose primary product is voice AI service delivery, the math changes at five clients and breaks at ten. The per-sub-account AI stacking, the absent outbound minute allotment, the coarse rebilling model, and the lack of cross-client voice performance visibility are the specific walls. They are documented, they are on GHL's ideas board as open requests, and they are not shipping in the near term based on public roadmap signals.

By builders, for builders. Hermes is the platform that fills the gap. Not as a replacement for GHL in marketing agencies, but as the dedicated voice layer for agencies where voice is the core product. $149/month Starter for three clients. $699/month Agency for twenty. Fixed overage at $0.24/min. No per-client subscription stacking.

/ next step

Built voice AI on GHL and hitting the wall at 5+ clients?

The Founders' Beta gives you the Agency tier at $699/month: 20 workspaces, 2,000 included minutes, per-workspace billing, voice-native CRM, and your phone numbers in your account. First agent live in 72 hours. No per-client AI subscription stacking.

Apply for the Founders' Beta See the full GHL stack comparison

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

OpenAI Just Shipped GPT-Live-1. Here Is What It Means for Your AI Voice Agency.

Alfredo Romero — Thu, 09 Jul 2026 14:34:29 +0000

By Alfredo Romero, CEO of BuildWithHermes

OpenAI released GPT-Live-1 and GPT-Live-1 mini on July 8, full-duplex voice models that speak and listen at the same time, handle natural interruptions, and translate live. The bar for what a voice agent sounds like just moved. Again.

What Actually Shipped

Full-duplex is the headline. Previous voice models worked like walkie-talkies: the model waits for you to stop, then responds. GPT-Live-1 speaks and listens simultaneously. It handles interruptions the way a person does, backs off mid-sentence, and picks the thread back up. It also does live translation across languages in the same call.

This is not an incremental update. Turn-taking latency has been the single biggest tell that a caller is talking to a machine. OpenAI just made natural turn-taking table-stakes for the whole industry.

And the timing is not a coincidence. The voice AI infrastructure layer is in a full sprint: ElevenLabs is reportedly approaching a $22B valuation, double its February number in five months. xAI shipped a voice agent builder in beta. The engines are getting better, richer, and better funded every quarter.

Why This Matters for AI Voice Agencies

Two things are true at once, and most of the hot takes this week only pick one.

First: this is great for you. Every model release like this makes your product better without you writing a line of code. Your agents will sound more human in 2027 than they do today, and you will not have paid for that R&D. OpenAI, ElevenLabs, and xAI are collectively spending billions improving the thing you resell.

Second: this is a trap if you compete on it. If your agency pitch is "our agents sound the most natural," you are now competing with OpenAI's roadmap. You will lose that race every single quarter. Whatever engine you demo today, your competitor demos the newer one next month.

The agencies that survive engine cycles do not sell voice quality. They sell operations: campaigns launched, calls handled, leads booked, reporting the client can read, an invoice the client understands, all under the agency's own brand. Voice quality is an input. The operating layer is the product.

There is a third pressure point nobody should ignore: model releases accelerate adoption, and adoption accelerates regulation. Colorado's AI Act took effect this month, requiring clear AI disclosure on calls, and the FCC has confirmed TCPA applies to AI-generated voices. The more human these models sound, the harder regulators will push on disclosure. A raw API gives you none of that protection.

What We Are Doing at Hermes About It

Hermes was built for exactly this moment. We are not a voice engine and we do not pretend to be one. Hermes is the operating platform that sits on top of the engines: agent builder, native CRM, outbound and inbound campaign orchestration, compliance guardrails, transparent billing, all white-labeled under your agency's brand.

That architecture means engine improvements flow through to you instead of disrupting you. When a new generation of models clears our reliability and compliance testing for agency use, it shows up inside the same platform, under the same plans: $149/month Starter with 300 included minutes, $399 Business with 1,000, $699 Agency with 2,000, and a flat $0.24/min overage. No repricing shock because the underlying model got better. Your clients never see the word Hermes, and they never see the engine either. They see your brand getting better every quarter.

Our bet has not changed since day one: the engines will keep leapfrogging each other, and the agencies that win will be the ones that own the layer above them.

Action Steps for Agencies This Week

Reposition your pitch away from voice quality. If your deck leads with "most natural-sounding AI," rewrite it this week. Lead with outcomes: answer rate, booking rate, cost per handled call. Those numbers are yours; the voice belongs to whoever shipped the last model.
Use the news as an opener. Every prospect saw the OpenAI headlines. Send a short note: "Voice AI just got dramatically better. Here is what that means for your missed calls." Newsjack for your own pipeline.
Audit your engine dependency. If your entire stack is hard-wired to one provider's API, you inherit their pricing, their outages, and their roadmap. Ask your vendor one question: what happens to my clients when you swap models? If the answer is a rebuild, you have a problem.
Check your disclosure compliance now. Colorado's rules are live and more states are following. Every outbound agent should disclose it is AI within the opening of the call. If your current stack cannot enforce that per campaign, fix it before a demand letter fixes it for you.
Stop paying for duct tape. If you are stitching a voice API to GoHighLevel, Zapier, Stripe, and Twilio to deliver what one platform does, the engine wars will keep breaking your stack at the seams. Consolidate before the next release cycle, not after.

The Bottom Line

GPT-Live-1 is genuinely impressive, and it changes nothing about what your clients pay you for. They pay for a phone that always gets answered, a pipeline that fills, and a report that makes sense. The engine underneath will be replaced four more times before 2028. Own the layer that does not get replaced.

FAQs

Does GPT-Live-1 make my current voice agents obsolete?

No. It raises the ceiling on conversation quality, but your clients buy outcomes: answered calls, booked appointments, qualified leads. A slightly more natural voice does not replace the CRM, campaign orchestration, compliance, and white-label layer that actually runs the engagement. Model quality is upgradeable. Your operating stack is the durable part.

Should my agency build directly on the OpenAI voice API instead of a platform?

Only if you want to become an infrastructure company. Building on a raw API means you own telephony, campaign logic, contact management, billing, and compliance yourself. That is the 5-to-7-tool duct-tape stack, typically $8K to $15K per month in developer cost. Platforms like Hermes sit on top of the best engines and swap them as they improve, from $149/month.

Will Hermes adopt full-duplex models like GPT-Live-1?

Hermes is built model-agnostic by design. As full-duplex models mature and pass reliability and compliance testing for outbound agency use, they become available under the same plans: $149 Starter (300 min), $399 Business (1,000 min), $699 Agency (2,000 min), with flat $0.24/min overage. Your clients get better conversations. Your pricing does not change.

One platform. Your brand. From $149/month. First agent live in 72 hours. Join the beta or see platform vs. raw API.

Originally published at buildwithhermes.com/blog/openai-gpt-live-1-voice-ai-wars-agencies-2026-07-09.

CNN Reported AI Voice Cloning Scams Cost Americans $893M. Here's What That Means for Your Agency This Week.

Alfredo Romero — Thu, 09 Jul 2026 13:58:24 +0000

The scam wave is real. For compliant AI voice agencies, it is also the clearest trust positioning moment the industry has ever handed you.

By Alfredo Romero, CEO, BuildWithHermes · 6 min read

On May 29, 2026, CNN published a major investigation into AI voice cloning scams, reporting that the FBI tracked $893 million in losses from AI-related fraud last year, with voice cloning attacks among the primary vectors. The story ran with a victim interview and syndicated to 12+ local TV stations and Yahoo News within 24 hours. It is at peak velocity right now.

If you run an AI voice agency, your phone has probably already lit up with questions from clients, prospects, or people in your community who saw the headline and put two and two together in the wrong order. This post is for those conversations.

Why this matters for AI voice agencies

The CNN story is not about your business. It is about criminals using voice synthesis tools to clone the voices of family members and impersonate them in emergency scams. The classic structure: a grandparent receives a call that sounds exactly like their grandchild saying they are in jail and need bail money immediately. The caller is an AI. The money is gone.

That operation has exactly nothing in common with a white-labeled outbound appointment confirmation system for a dental practice. The technology shares a category name. The use case, the consent model, the regulatory context, and the intent are entirely different.

But most of your clients and prospects do not know that. When they hear "AI voice" after reading the CNN story, they pattern-match to the scam. That is a sales friction problem you need to get ahead of, not a problem with your product.

The second thing the story does is create a genuine trust premium in the market. Compliance-first agencies, ones that can show A2P registration, documented consent workflows, and a transparent call infrastructure with verifiable caller ID, are now differentiated from the broad category of "AI voice" in a way they were not last week. That differentiation is worth money. Use it.

What we are doing at Hermes about it

Hermes is an operating platform for AI voice agencies. Every workspace deployed through Hermes is white-labeled, meaning the end client never sees infrastructure names, never receives a call from an unverified number, and never interacts with a system that lacks a documented consent trail. That is table stakes for how the platform was built, not a feature we added after the news cycle.

Specifically, the infrastructure that separates Hermes-deployed agencies from the scam operations CNN covered:

A2P 10DLC submission workflow. Hermes has a built-in A2P compliance submission flow. Agencies on the platform can register their calling campaigns, which creates a verifiable paper trail that regulatory enforcement, carriers, and clients can inspect. Scammers cannot produce this. Registration costs $30 per submission as a direct pass-through with zero markup.
Dedicated number pools per workspace. Every client on a Hermes agency gets their own workspace with dedicated phone numbers tied to that specific campaign and consent pool. There is no shared infrastructure bleeding call history across clients.
Consent-gated outbound campaigns. The campaign orchestration layer in Hermes is designed around contact lists with documented opt-in status. You cannot run a cold outbound blast to a purchased list. You run outbound to contacts with documented consent.
White-label at the infrastructure level. Your clients hear your agency name and see your brand. They never hear the word Hermes. The call comes from a number they can verify is attached to the business they hired. That is not just a branding preference. It is the technical mechanism that makes your compliance story credible.

We are also pricing these protections in a way that does not require agencies to be large to afford them. Starter is $149 per month with 300 included minutes. Business is $399 per month with 1,000 minutes. Agency is $699 per month with 1,650 minutes. The compliance infrastructure is the same at every tier. It is not a premium add-on.

Action steps for agencies this week

Send a proactive trust note to every active client before they ask. Do not wait for a client to forward you the CNN article with a question mark at the end. Send them a short note today. Two to three sentences: you saw the news, here is the specific difference between voice cloning fraud and what you run for them, here is the documentation they can ask for at any time to verify compliance. If you do not have that documentation, that is the first problem to fix.
Update your sales script to address the scam story directly in the discovery call. Bring it up before the prospect does. Something like: "I want to address the CNN story on AI voice scams directly, because it is the first thing most people think about when they hear 'AI voice.' Here is exactly what is different about what we run." Naming the elephant in the room closes it faster than hoping it does not come up.
Get your A2P compliance documentation in order. If you are not yet A2P registered, this week is the week to fix that. The scam wave is driving carrier scrutiny and regulatory attention toward the entire AI voice category. Unregistered campaigns are the first thing to get flagged when carriers start increasing filtering on AI-generated calls. Registration protects your deliverability, not just your legal exposure.
Use the news cycle as a prospecting angle in your community. Post about this in the communities where your prospects live. Not a pitch. A factual breakdown of the difference between voice cloning fraud and compliant AI voice outbound. The people who respond with questions are your warmest prospects right now. The concern creates the conversation. The conversation creates the opportunity.
Position your pricing as a trust premium, not a cost. The agencies charging $1,500 to $3,000 per client per month have always had a credibility gap to close with prospects who wonder why they should pay that instead of just buying a $49/month tool. The CNN story closes that gap for you. Compliance infrastructure, white-label accountability, and a vetted consent workflow are worth the premium. You just got a national news story explaining why. ## Frequently asked questions

Does the CNN AI voice cloning scam story hurt legitimate AI voice agencies?

Short-term, yes, at the conversation level. When a prospect hears 'AI voice' and the last thing they read was a CNN story about grandma's voice being cloned for fraud, you face extra friction at the top of the sales conversation. Long-term, no. The scam wave is actually a forcing function for the market to distinguish between criminal voice cloning and legitimate, consent-driven AI voice outbound. Agencies that can point to A2P 10DLC registration, TCPA-compliant consent workflows, and white-labeled call infrastructure are the answer to the problem CNN is reporting. The story is a setup for your pitch, not a threat to it. The agencies who use it that way will close more deals this quarter than the ones who avoid the topic.

What is A2P compliance and why does it matter when AI voice scams are in the news?

A2P stands for Application-to-Person, and it refers to the regulatory framework that governs automated outbound calling and messaging in the US. A2P 10DLC registration (for SMS) and the FCC consent rules for AI-generated calls are the legal mechanisms that separate compliant agency outbound from the criminal operations CNN is covering. When you are registered, your client's phone number is attached to a verified campaign with documented consent, and the call is made from a dedicated number tied to that registration. When scammers clone voices, they do none of that. The compliance infrastructure is the trust signal. Hermes builds A2P submission workflows directly into the platform, so every agency on Hermes can show their clients a paper trail that criminals cannot produce.

How should an AI voice agency respond when a prospect brings up AI scams?

Do not deflect. Acknowledge it directly and use it as a setup. The script is simple: 'You saw the CNN story. That is criminal voice cloning, which is a completely different operation from what we run. Here is what separates us: every call we make is tied to a registered A2P campaign, goes out from a dedicated number the recipient can verify, and was initiated on the basis of documented consent. The FBI report is about people cloning voices to impersonate family members in emergency scams. We are running outbound appointment confirmations for your HVAC clients who opted in. The infrastructure is different. The consent is different. The intent is different. If anything, the story is why your clients should be asking for proof of compliance before they hire anyone to run AI voice. We have that proof.' That framing turns the news into a qualifier. Prospects who are serious will want that documentation. Tire-kickers will not ask for it.

The bottom line

The CNN story is bad news for the criminals. It is a positioning opportunity for every agency that took compliance seriously before the news made it mandatory to explain.

The agencies who win this week are the ones who respond to the news with documentation and a clear explanation, not silence. The ones who treat it as a threat will lose deals to the ones who treat it as a setup for the right conversation.

By builders, for builders. One platform. Your brand. Your margins. From $149 per month.

Sources: CNN, AI voice cloning scams investigation, May 29 2026 · Henson Legal, AI voice compliance overview 2026 · Vapi status page, May 2026 incident log · Famulor, Synthflow 2026 pricing audit · Trillet, Voice AI white-label pricing breakdown 2026

/ next step

Compliant outbound under your brand. From $149/mo.

A2P compliance workflows, dedicated number pools, white-label CRM, and consent-gated campaigns. Everything the CNN story just made your clients ask for. First agent live in 72 hours.
Start on Hermes from $149/mo Hermes vs building on Vapi

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

/ written by

Alfredo Romero

CEO and Co-Founder, Hermes

Alfredo runs sales, operations, and strategy at Hermes. Before founding Hermes he ran agencies for nine years and spent the last three building the AI voice operations side. He writes the operator playbook from real builds, not theory.
LinkedIn ↗X (@buildwithhermes) ↗About the founding team →

Originally published on the BuildWithHermes blog.

Synthflow Cancellation Hell: How to Get Off (and What Stack to Move To)

Alfredo Romero — Thu, 09 Jul 2026 13:53:05 +0000

You want to leave Synthflow. The platform is not what you need anymore, the pricing has outpaced your margins, or you found something that actually fits how your agency operates. So you open the dashboard to cancel. And you look. And you keep looking. There is no cancel button. There is no "close account" link. Multiple G2 reviews published in 2026 describe this exact experience: users search every settings menu, every billing page, every account section, and find nothing. Some report being charged for another cycle after they believed they had canceled.

This post is the full exit checklist. How to actually stop the billing, how to get your data and phone numbers out intact, how to rebuild your agents on a new platform without losing client continuity, and what platforms are worth landing on once you are out. The process takes 2 to 4 weeks done properly. It takes 3 days done badly, which usually means one missed client call and one very unhappy retainer conversation.

By builders, for builders. We have helped agency owners navigate three separate platform exits in the last six months. The playbook below is what actually works.

Why is Synthflow so hard to cancel?

The short answer is that Synthflow does not surface a self-serve cancellation path from the main dashboard. The official path is Settings, then Plan & Billing, then the billing portal. Some users reach that path and find a cancellation option. Others do not, and the discrepancy appears to be plan-tier dependent, with lower tiers requiring a direct support email to complete the process.

The deeper issue is account deletion. Under Synthflow's billing documentation, billing management happens through the billing portal, but there is no dashboard-level "delete account" button. Reviewers on both G2 and Trustpilot in 2026 flag this as a significant problem, with one reviewer explicitly calling it an "unacceptable practice for any SaaS in 2026, especially under European GDPR rules where users have an explicit right to erasure." GDPR Article 17 requires that users be able to request the deletion of their personal data. A platform that does not offer a self-serve deletion path creates a friction barrier to that right.

"There is no way to close your account from the dashboard. I searched every settings menu and billing page. This is an unacceptable practice for any SaaS in 2026, especially under GDPR." G2 reviewer, 2026

One separate Trustpilot reviewer reports being charged after cancellation, with Synthflow claiming the cancellation was not processed correctly. This pattern, friction on exit plus continued billing, is what earns the "cancellation hell" label from operators who have been through it.

What should I do before I cancel?

Do not cancel first and then figure out the rest. That sequence gives you active clients, no platform, and no fallback. The right order is: inventory everything, export everything, rebuild in parallel, test in parallel, notify clients, cut over, then cancel. Here is each step in detail.

Inventory all active agents and their client mappings. List every agent by name, which client it serves, what phone numbers it is attached to, and what integrations it touches (CRM, calendar, Zapier/Make). This is your migration map. Do not start without it.
Export every agent as JSON. Synthflow allows you to export individual agents as portable JSON files from the agent editor, per their agent FAQ documentation. This preserves the prompt structure, flow logic, and configuration. Export every agent before you do anything else. Note that entire subaccounts cannot be exported as a single file, so you will need to export each agent individually.
Download all call logs. Export your call history with date range filters before canceling. If a client ever asks for historical call data post-migration, that export is the only copy you will have. Synthflow's data retention applies 30-day limits by default on the newer compliance setting, and accounts are deleted within 90 days of closure.
Screenshot all prompt configurations. Even with a JSON export, a visual record of how each agent was configured, including any conditional logic branches, is useful for rebuilding. The JSON is the source of truth but screenshots are faster to reference during rebuild.
Note every phone number and its carrier details. If you have Synthflow-provisioned numbers, those run through their underlying telephony stack. If you have custom numbers linked via SIP or Twilio, document the E.164 format, SIP domain, authentication username, and password for each. You will need these for porting.

How do I get my phone numbers out of Synthflow?

This is usually the most stressful part of a migration, and the most important. A live client phone number tied to a dead platform is not theoretical, it has ended retainers.

Per Synthflow's phone number documentation, porting takes 1 to 2 weeks and may include a one-time porting fee depending on your carrier. The key distinction is whether your numbers are Synthflow-provisioned or custom-imported.

Synthflow-provisioned numbers (the ones you bought directly inside the platform) are Twilio or Vonage numbers at the carrier layer. You can port these out by initiating a port-out request with your destination carrier. The process is identical to porting any Twilio number: you need the account SID, the auth token, and confirmation from Synthflow support that the number is releasable. Start this request at least 2 weeks before your intended cutover date.

Custom-imported numbers (your own Twilio, Telnyx, or SIP numbers that you connected via the custom phone number feature) are easier. You simply stop pointing them at Synthflow's SIP endpoint and point them at your destination platform instead. The number never moved, only the routing did. This is usually a same-day change.

The migration guide for keeping Twilio numbers intact when moving between voice platforms is covered in full in the Twilio number retention playbook. The approach is the same regardless of what destination platform you choose.

How do I rebuild my agents on a new platform?

The exported JSON from Synthflow is a reference document, not a drop-in import file. No competing platform accepts Synthflow's proprietary JSON schema natively. What the export gives you is the full prompt text, the flow logic, and the action configurations, which is everything you need to rebuild efficiently.

The rebuild sequence that works:

Set up the destination platform with a parallel workspace. Do not shut down Synthflow until the new platform is live and tested. Run both in parallel for at least 5 business days on a real client workflow before cutting over.
Rebuild your highest-volume agent first. The agent with the most calls per week has the most test surface area and the most to lose if the rebuild has issues. Do not save it for last.
Run parallel test calls. Set up a test phone number on the new platform pointing at the rebuilt agent. Run 20 to 30 test calls that mirror your real call scripts. Check latency, interruption handling, and transfer behavior. Synthflow reviewers consistently cite barge-in handling and complex branching as failure points, so test those specifically.
Migrate integrations one at a time. GHL, HubSpot, Zapier, and calendar connections each need to be re-authenticated on the new platform. Test each integration independently before combining them in a live agent.
Notify clients 5 business days before cutover. A one-paragraph email explaining that their service is migrating to a new platform, that there will be no interruption to their calls, and that their phone numbers are not changing is usually sufficient. Clients who do not get this email before the cutover are the ones who escalate.

Which platform should I move to after Synthflow?

The answer depends on what is actually broken for you. Not every Synthflow exit is the same. Here are the three most common reasons agencies leave, and the destination that matches each.

If you are leaving because of pricing and invoice opacity: Synthflow's model bundles a base subscription with per-minute overages at $0.12 to $0.13/min after you exhaust included minutes, plus the cost of LLM tokens and ElevenLabs voices on top. The Retell AI 2026 Synthflow review puts the real all-in cost at $0.18 to $0.22/min under the included tier, which gets significantly worse in chatty months. If your issue is not knowing what the bill will be at month-end, you need a platform with a single published rate and a ceiling. That is exactly what a platform like BuildWithHermes provides: one invoice, one rate.

If you are leaving because of white-label problems: Multiple G2 reviewers flag that Synthflow branding surfaces to end clients in edge cases, particularly in support tickets that get redirected to Synthflow's team rather than staying inside the agency's branded workspace. If your agency brand has been exposed to a client because of a support escalation, you need infrastructure that was built from the ground up for white-label isolation. Your clients should never see the word "Synthflow," "Hermes," or any vendor name. The platform should be invisible.

If you are leaving because of platform stability: The Retell review and Softailed's 2026 Synthflow review both cite API instability, calls that drop mid-conversation, and minutes that get consumed without a matching call event. If your clients have complained about dropped calls or if your usage reports do not match what actually happened, that is an infrastructure problem, not a configuration problem. Moving to a platform with published uptime SLAs is the fix.

How do the platforms compare for a migrating agency?

Side-by-side at the agency tier, using the three most common alternatives. Pricing sourced from the current public pages of each platform as of May 2026.

Platform	Agency plan price	Real all-in rate	Invoice count	Account deletion
Synthflow Agency	$1,250/mo	$0.18 to $0.22/min + LLM	2 to 3	No self-serve option
Retell AI	$0.07/min PAYG	$0.13 to $0.18/min	1 to 2	Standard
Vapi	$0.05/min + upstream	$0.23 to $0.33/min	4 to 5	Standard
Hermes Agency	$699/mo + $0.24/min flat	$0.24/min (ceiling, not floor)	1	Standard

The Retell rate looks cheapest on paper, but Retell is an API infrastructure layer, not a full agency platform. You still need CRM, white-label configuration, campaign orchestration, and billing tooling on top. That adds invoices and developer time. The Vapi column is included because a significant portion of agencies migrating off Synthflow consider Vapi next, and the true all-in rate surprises them the same way Synthflow did. The 5-invoice problem is a Vapi-specific pattern worth reading before you land there.

What does a Synthflow migration actually cost?

The direct costs are the easy part. The hidden cost is time.

Per the Ringly 2026 Synthflow alternatives guide, agencies should plan 2 to 4 weeks for a full migration: export and setup (3 to 5 days), agent rebuild and parallel testing (1 week), client notification and phased cutover (3 to 5 days), and phone number porting (1 to 2 weeks, can overlap). Running both platforms during the overlap period means you are paying for two platforms simultaneously for roughly 2 weeks. Factor that in.

The founder-hour cost is usually the larger number. If you have 10 client agents to rebuild, budget 3 to 5 hours per agent for rebuild plus testing. That is 30 to 50 founder-hours for the rebuild alone, before you count porting, integration re-authentication, and client communication. At $150/hour opportunity cost, the real migration cost for a 10-client agency is $4,500 to $7,500 in founder time, plus whatever you pay for the overlap period.

The agencies that spend the least on migration are the ones that rebuild on a platform with a shorter setup path per client. Hermes is designed around the per-client workspace model, so each client gets an isolated workspace with its own billing, numbers, and agents. The rebuild for each client is faster than rebuilding from scratch because the workspace structure mirrors how you already think about your book.

How do I stop the Synthflow billing and get the account closed?

Based on the documented user experience pattern, here is the specific sequence that works:

Go to Settings, then Plan & Billing in the Synthflow dashboard. Look for a billing portal link or manage subscription button. If present, use it to downgrade or cancel the subscription.
If no cancel option appears, email support@synthflow.ai directly. Include your account email address, your subscription tier, the date you want cancellation to take effect, and a written request to delete your account and all associated data under GDPR Article 17 (or CCPA, if you are US-based). Request written confirmation.
Screenshot the cancellation confirmation or email thread. If you are later charged, this is your dispute documentation for your credit card company or Stripe chargeback.
If you are on an annual plan, send the cancellation email at least 30 days before your renewal date. Cancellation for annual plans takes effect at the end of the current contract period. Do not wait until 3 days before renewal.
Monitor your payment method for one full billing cycle after cancellation. If you see a charge you did not expect, dispute it immediately with your documentation.

"Plan 2 to 4 weeks for full migration: export flows, rebuild on new platform, run parallel testing, then switch. Migrating between voice AI platforms mid-stream is expensive and disruptive." Ringly, 2026 Synthflow Alternatives Guide

What should the next platform look like?

The Synthflow exit is an opportunity to fix the structural problems, not just swap one logo for another. Most agencies that migrate from Synthflow are doing it because one of three things broke: the price, the white-label isolation, or the stability. Picking a destination that solves all three, not just one, is how you avoid this post in 12 months for a different platform.

The checklist for the destination platform:

Single invoice. If the destination platform bills STT, LLM, TTS, and telephony on separate statements, you are recreating the opacity problem with a different vendor. One invoice, one rate, one reconciliation at month-end.
Published overage ceiling. You need to be able to price your retainers against a number that is fixed. An overage rate that varies with LLM model selection or voice catalog choices is not priceable. A flat $0.24/min ceiling that includes everything is.
Native white-label. Your clients should never see a vendor name in a ticket, an email, an error message, or a support thread. This is not a configuration toggle, it is an architecture question. If the platform was built with white-label as an add-on rather than a default, the Synthflow branding problem will eventually resurface with the new vendor's name.
Self-serve account management including deletion. You should be able to cancel, downgrade, and delete your account from the dashboard without a support ticket. Any platform that does not offer this is creating the same friction on exit that you are experiencing with Synthflow right now.
Published uptime history. Ask for the status page before you commit. The platform should have a public incident history, not just an uptime claim.

Hermes was built against this checklist. One invoice. Flat $0.24/min overage. Native white-label where your clients never see the word Hermes. Standard account management. The Agency plan is $699/month with 2,000 included minutes and 20 client workspaces. Starter is $149/month with 300 minutes and 3 workspaces, Business is $399/month with 1,000 minutes and 7 workspaces. The migration path from Synthflow to Hermes follows the same steps described above.

If you are also evaluating what happened to the platforms that stayed focused on agencies while Synthflow pivoted upmarket, the context lives in the Synthflow enterprise pivot analysis. And if the billing confusion you are leaving Synthflow over is a pattern you have seen on Vapi as well, the 50-client margin math post puts an exact dollar figure on how that plays out at scale.

Frequently asked questions

Why can't I find a cancel button in Synthflow?
Multiple G2 and Trustpilot reviewers in 2026 report there is no option to delete or close an account from the Synthflow dashboard, despite searching every settings menu, billing page, and account section. To cancel, you need to go to Settings > Plan & Billing and manage your subscription from the billing portal there, or contact Synthflow support directly. Some users report needing to email support to fully close the account.

Can I get a refund from Synthflow after canceling?
Synthflow's standard billing terms do not include refunds for unused minutes or the current billing cycle. Users who report being charged after cancellation have had to escalate via email to dispute the charge. If you are on an annual plan, contact support before the renewal date, typically at least 30 days in advance, to confirm the cancellation will take effect at the end of the current period.

How long does Synthflow phone number porting take?
Phone number porting out of Synthflow typically takes 1 to 2 weeks and may incur a one-time porting fee depending on your carrier. To import a custom number into a new platform, you will need the number in E.164 format plus your SIP configuration details. Note that importing custom numbers in Synthflow requires an Enterprise account, which means if you are on a lower tier, your numbers are Synthflow-provisioned and the porting process goes through your underlying carrier (typically Twilio or Telnyx).

Can I export my Synthflow agents before migrating?
Yes. Synthflow allows you to export individual agents as a portable JSON file from the agent editor. This preserves the agent configuration and prompt structure. However, it is not possible to export or import entire subaccounts, which means agency owners managing multiple client subaccounts will need to export each agent individually and rebuild the subaccount structure on the destination platform.

What is the best platform to migrate to from Synthflow?
The right destination depends on what broke for you on Synthflow. If the issue was pricing opacity and invoice sprawl, a platform with a single all-in invoice (like Hermes) solves that directly. If the issue was white-label limitations or Synthflow branding bleeding through to clients, a native white-label infrastructure platform solves that. If the issue was per-minute cost at scale, an API-layer platform like Retell gives lower rates with more control.

How long does a full Synthflow migration take?
A well-planned migration takes 2 to 4 weeks end to end: 3 to 5 days for data export and platform setup, 1 week for agent rebuild and testing in parallel, 3 to 5 days for client notification and a phased cutover, and 3 to 7 days for number porting. Running both platforms in parallel during the test phase is the safest approach and adds cost but prevents client-facing downtime.

Does Synthflow comply with GDPR on account deletion?
GDPR Article 17 gives users an explicit right to erasure. Multiple 2026 reviews flag that Synthflow provides no dashboard-level account deletion, which reviewers describe as an unacceptable practice for any SaaS operating in 2026. Synthflow's documentation states data is deleted within 90 days under contract terms, and they recently added a 30-day retention limit option for transcripts and recordings. If you need confirmed erasure for GDPR compliance, you should request it in writing via email to Synthflow's support team and retain their written confirmation.

The migration in summary

Leaving Synthflow is harder than it should be. The exit checklist is: inventory and export before you do anything, export every agent as JSON, download all call logs, document every phone number, initiate porting at least 2 weeks early, build and test in parallel on the new platform for at least 5 business days, notify clients before cutover, and follow up with a written cancellation to Synthflow support to ensure the account is actually closed.

The agencies that have the cleanest exits are the ones that treat the migration as a client-delivery project, not a back-office task. It has a timeline, a client communication plan, and a go/no-go test before the switch. By builders, for builders. The platform you land on should not require this post in 12 months.

Originally published at buildwithhermes.com/blog/synthflow-cancellation-hell-migration-guide.

How to Switch From Retell to BuildWithHermes Without Losing Your Twilio Numbers (Step-by-Step)

Alfredo Romero — Mon, 06 Jul 2026 16:45:26 +0000

Originally published on the BuildWithHermes blog.

/ migration playbook · operator guide
You do not have to port your Twilio numbers to leave Retell. The fastest migration path keeps every number exactly where it is on your Twilio account, flips a single Voice URL on each number, and routes the call into BuildWithHermes through Twilio Elastic SIP Trunking. Finished in an afternoon. Zero port fees. No FCC window. No client downtime. If you decide later that you want Hermes to own the numbers end to end, you can run a separate port-in over 7 to 15 business days using Twilio's port-out process and Hermes' managed number intake. Both paths preserve the caller-ID, the call history on the Twilio side, and your client's existing phone tree. This post walks both paths step by step, names the gotchas that take down rushed migrations, and gives you a parallel-run plan so you never have a single client wonder why their AI receptionist went dark for an hour. The short version: pick BYO Twilio first, port only if Twilio is still in your stack 90 days from now.

I have run this migration twice for our own agency operators and walked five Founders' Beta applicants through it on Loom in the last three weeks. The pattern is consistent. The teams that treat it like a number-porting project lose two weekends. The teams that treat it like a trunk-config project finish before lunch.

The rest of this is the exact step list, the carrier docs, the FOC timing window, the prompt import, the parallel-run cutover, and the post-migration audit. If you only read one section, jump to "Path A: BYO Twilio in one afternoon" below.

Why are agencies leaving Retell for Hermes in 2026?

Two reasons dominate the migration intake forms we run on the Founders' Beta application. First, cost unpredictability. The Retell voice-infrastructure rate of around $0.055 per minute is the orchestration line only. STT, LLM tokens, TTS, and Twilio telephony stack on top. Ringg.ai's 2026 Retell pricing breakdown concluded that real production setups land at "$0.13 to $0.31/min once dependencies are included," and flagged that agencies and high-volume operators feel the squeeze first. We wrote a full audit of the same phenomenon across providers in the 5-invoice problem.

Second, white-label. Retell does not offer a white-label workspace. For an agency reselling voice AI under its own brand, that is a hard limit. Convocore's 2026 alternatives roundup opens with the exact line: "Retell doesn't offer white label, which is problematic for agencies reselling voice AI services." The same theme runs through Dialora's 2026 Retell pricing review, which flags hidden Twilio costs and a missing client-facing portal as the two most common reasons resellers churn.
"The per-minute number looks fine on paper, but actual bills land much higher once the LLM, voice provider, and Twilio minutes are factored in. Agencies and high-volume operators feel this pinch the most." [ Ringg.ai, Retell AI Reviews 2026 ]
Neither of those reasons makes Retell a bad product. Retell is a good builder platform. It is not a good agency platform. The right move for most operators is to keep what works (the agents you have already prompt-engineered) and move them to a chassis built for resale.

The two migration paths, side by side

Before picking a path, line them up against your situation. Most operators pick Path A. Path B is only worth running once you no longer want Twilio in your stack at all.
Dimension Path A: BYO Twilio Path B: Full port-in Time to live 2 to 6 hours per agent 7 to 15 business days Port fee per number $0 $0 to $30 typical FCC window Not required Simple ports complete in 1 business day; full process 2 to 4 weeks Twilio account stays Yes No (or used elsewhere) Risk profile Reversible in seconds Reversible only by re-porting Best for Active client books, multi-client agencies, parallel-run periods Solo operators consolidating to one vendor, accounts ready to leave Twilio
The FCC requires simple ports to complete in 1 business day and non-simple ports in 4, but in practice the documentation cycle, the carrier validation, and the Firm Order Commitment scheduling stretch the window to 2 to 4 weeks for a US local number, per Viirtue's number porting guide. That is fine if you have one number. It is a real operations hazard if you have thirty.

Path A: BYO Twilio in one afternoon

This path keeps every Twilio number where it is, keeps your Twilio sub-account structure intact, and routes the call audio into Hermes through Elastic SIP Trunking. Hermes runs the agent, the prompts, the LLM, the STT, the TTS, and the recording. Twilio stays the carrier of record.

Twilio itself documents the elastic SIP trunk as "a cloud-based solution providing PSTN connectivity for IP-based communications infrastructure," per the official Twilio Elastic SIP Trunking documentation. The same trunk that Retell already uses on your account is what Hermes will use. The cutover is a configuration change, not a re-architecting.

Step 1: Create the Hermes workspace and import your Retell agents

Apply to the Founders' Beta, then export your Retell agents as JSON from the Retell dashboard. Drop them into Hermes' agent importer. Prompts, conversation states, voice settings, and webhook URLs transfer one to one. Tool calls and custom functions need their destinations re-pointed at the new Hermes function-call format, but the function shapes are identical.

Knowledge bases do not transfer. Re-upload them. Vector schemas differ between providers and there is no clean migration path on embeddings. Budget 30 to 60 minutes per KB.

Step 2: Provision a Hermes SIP trunk endpoint

Inside the Hermes workspace, generate a SIP trunk endpoint per client workspace. This is a single URL of the form sip:@trunk.buildwithhermes.com. You will paste this into the Twilio side in step 4.

For the technical reference on how Retell itself recommends this pattern, see Retell's own custom telephony documentation. The architecture is identical. Hermes just substitutes the receiving end.

Step 3: Validate on a sandbox number first

Hermes provisions a free sandbox number with every workspace. Point the imported agent at the sandbox number, run five test calls against it from a personal cell, listen end to end. Confirm STT accuracy, voice latency, tool calls firing, recording playback, and the call appearing in the workspace call log. Do not skip this step. The reason no client ever hears a dead silence is that you validated on a number nobody cares about first.

Step 4: Repoint the Twilio number's Voice URL

Open the Twilio console. Navigate to the phone number you want to migrate. Change the "A call comes in" handler from Retell's webhook to the SIP URI from step 2. Save. The change goes live immediately. Call the number. The audio routes through Hermes. You are migrated.

For the official Twilio reference on this configuration pattern, including authentication headers and TLS requirements, see Twilio's step-by-step elastic SIP trunking setup.

Step 5: Roll forward one client at a time

Repoint one client's production number. Wait 48 hours. Watch call logs, watch the call review queue, watch the invoice line for that workspace. If everything looks clean, move the next client. The default rollout cadence we recommend is one client per business day for a 10-client book. By day 14 the migration is done and Retell is cancellable.
"To migrate a phone number, you will need to set up a SIP trunking and configure your number to point to it, and then import that number to Retell." That same pattern, in reverse, is what makes a Retell-to-Hermes migration a config change rather than a port. [ Retell AI, Custom Telephony Overview ]

Path B: Full port-in to Hermes managed numbers

Path B exists if you want to leave Twilio entirely. Hermes becomes both the agent platform and the carrier of record. You get one bill, one console, one support escalation path. The tradeoff is the FCC porting window.

Step 1: Run the Twilio portability check

Twilio publishes a portability API that returns a clean signal on whether a number can leave the carrier without manual intervention. The endpoint is documented at Twilio's portability API reference. Hermes wraps that check in the migration intake form. You paste your list of E.164 numbers, the system returns "portable" or "manual review needed" for each one. Manual-review numbers are usually toll-frees, hosted numbers, or numbers with an open port-in already pending. Resolve those first.

Step 2: Request the service address and submit the LOA

Email porting@twilio.com to request the service address tied to your Twilio account, per CloudTalk's 2026 Twilio port-out guide. Sign the Letter of Authorization that Hermes generates. The LOA includes the end-user full name, the Twilio Account SID, and the service address Twilio returns. Sign electronically. Hermes submits the bundle to our underlying carrier within one business day.

Step 3: Wait for the Firm Order Commitment

Twilio validates the port, exchanges information with the receiving carrier, and issues a Firm Order Commitment (FOC) date and time. Per FCC rules, "simple ports must be completed in one business day. Non-simple ports must be completed in four business days." In practice, the documentation cycle is the long pole. Plan on 7 to 15 business days end to end. CloudTalk cites the same range: "around 15 days, 10 for set-up and 5 for the actual migration."

Step 4: Keep call forwarding live until FOC

Until the FOC fires, the number still belongs to Twilio. Keep the Twilio Voice URL pointing at Path A (the Hermes SIP trunk) for the full porting window. At FOC, the number leaves Twilio, the receiving carrier takes over, and Hermes routes the call natively. The caller-experience side of the cutover is invisible. The internal billing changes from "Twilio plus Hermes" to "Hermes only."

The full official process for porting away from Twilio is documented at the Twilio porting support category. Keep that tab open during the migration.

The agent and prompt migration, exactly

Independent industry analysis pegs platform-to-platform agent migrations at "2 to 6 weeks depending on complexity, with prompts and conversation logic transferring easily in 1 to 2 days, but integrations requiring rebuilding in 1 to 3 weeks, and total cost of $2,000 to $10,000 in development time," per SuperDupr's 2026 voice AI platform comparison.

The Hermes import collapses that timeline because we match Retell's agent JSON shape one to one for the components that transfer cleanly. Here is the breakdown.
Component Migration effort Notes System prompts 5 minutes Paste from Retell agent JSON. No reformatting needed. Conversation states 15 to 30 minutes per agent State graph imports directly. Re-test transitions. Function calls / tools 1 to 3 hours per agent Re-point webhook URLs to Hermes function endpoints. Schema is identical. Voice catalog 10 minutes ElevenLabs voice IDs port one to one. Deepgram Aura is the default budget voice. Knowledge bases 30 to 60 minutes per KB Re-upload source documents. Vector schemas differ. Webhooks (post-call) 15 minutes Hermes mirrors Retell's post-call event payload for backward compatibility. Call recordings (historical) Optional export Recordings stay on Retell. Hermes records all new calls natively.
The single highest-time bucket above is function calls. If your Retell agents pull from a CRM, push appointments to a calendar, or call back into a custom database, those endpoints need to be authorized against Hermes' callback IP range. The endpoint contracts themselves are identical.

The no-downtime cutover, in three commands

The cutover for any single number, once the agent is imported and validated on the sandbox, is three steps.

Update the Twilio Voice URL. Console → Phone Numbers → Active Numbers → the target number → Voice Configuration → A call comes in → SIP → paste the Hermes SIP URI. Save.
Place a verification call. Dial the production number from a personal cell. Confirm: Hermes agent answers, response latency under 800ms, voice quality clean, recording appears in the Hermes workspace within 30 seconds of hang-up.
Notify the client (optional). If the client owns operational oversight, send a short heads-up: "Agent infrastructure upgrade, no change to your line." If the client does not see the back end (the standard white-label case), no message is needed. Total elapsed time for steps 1 and 2: under 5 minutes. If anything looks wrong, the rollback is the inverse: change the Twilio Voice URL back to the Retell webhook. The call routes back to Retell. You are unmigrated. The reversibility is the entire reason Path A is the default.

Common gotchas that take down rushed migrations

These are the patterns that have cost operators a weekend. None of them are unfixable. All of them are avoidable.

Forgetting the function-call authentication headers. Retell signs callbacks with a Retell-specific HMAC. Hermes uses a different signature scheme. Update your CRM webhook to accept the new signature before you cut over, not after. The first failed booking gets an operator-visible alert in the Hermes dashboard, but the lost lead is real.
Leaving the old Retell agent active on the same number. If both Retell and Hermes have inbound webhooks registered to the same Twilio number, Twilio uses whichever responded last. Disable the Retell agent first. Repoint Voice URL second.
Mismatched STT language models. Retell defaults English Nova-2. Hermes defaults the same, but multi-language clients (Spanish inbound on a US line, for example) need the language model set explicitly. Confirm before cutover.
Not re-uploading the knowledge base. The agent will run without the KB. It will just answer the wrong question politely. Test KB hit rate on the sandbox number with five known queries before cutover.
Skipping the parallel-run window. Migrating 30 numbers on a Friday afternoon because the team wants to "be done by Monday" is the recipe for an emergency Saturday. One client per business day. No exceptions on a first migration.
Cancelling Retell before the last invoice clears. Keep the Retell account paid through the end of the parallel-run period. The accounting on a clean cancellation is easier than a refund dispute. ## What does the cost picture look like the month after migration?

A representative 8-client agency on Retell averaged $0.21/min all-in across 14,400 monthly minutes in Q1 2026 according to the audit we ran for that operator. Total cost of voice infrastructure: $3,024 against $14,000 in retainer revenue. Margin line: $10,976.

On Hermes, the same 8 clients, same volume, sit on the Business plan ($399/mo) plus overage. Included minutes: 1,000. Overage: 13,400 minutes at $0.21 flat. Total Hermes spend: $399 plus $3,216 = $3,615. Margin line: $10,385. Margin per client is within $74/month of where it was on Retell, but the operator recovered roughly 12 hours per month on invoice reconciliation and gained a white-label client portal that lifted three upsells in the first 60 days.

The math gets more favorable on heavier volume because the upstream caching collapses repeated knowledge-base hits. On the Agency plan ($699/mo, 2,000 minutes included), a 15-client book at 25,000 monthly minutes lands at $6,219 total versus $5,250 on Retell with five invoices and no portal. The white-label portal is what closes the gap. See the full Hermes pricing breakdown for the included minutes and overage math at every tier.

When should I actually port instead of BYO?

Three conditions, all of which need to be true.

You have been on Hermes for at least 60 days and the invoice/operations side is clean.
You no longer use Twilio for anything else (SMS campaigns from a separate stack, programmable voice for non-AI flows, etc.).
The accounting simplification of one carrier bill is worth more than the 7 to 15 business day porting window. If any one of those is false, stay on Path A. The most common regret on rushed ports is realizing afterward that a separate business unit was still using the same Twilio sub-account for SMS, and the port broke that flow.

Frequently asked questions

Do I have to port my Twilio numbers to switch from Retell to Hermes?

No. There are two valid paths. Path A keeps the numbers on your Twilio account and points a SIP trunk at Hermes (zero-port migration, finished in a single afternoon). Path B is a full port-in to Hermes managed numbers, which takes 7 to 15 business days under the FCC simple-port rules. Path A is the right default for active client books. Path B is only worth doing once you no longer want to manage a Twilio sub-account at all.

Will my client lose calls during the Retell to Hermes cutover?

Not if you stage it correctly. The Path A trunk swap is a single change to the Voice URL on the Twilio number. Hermes provides a parallel sandbox phone number so you can validate the new agent on live audio before you flip the production number. Live cutover takes under 60 seconds. Path B port-ins use a Firm Order Commitment time window so the carrier transfer is also seamless if you keep call forwarding active until the FOC confirms.

What happens to my Retell agents, prompts, and functions during migration?

Prompts and conversation logic transfer in a day. Function calls and webhook integrations need to be re-pointed at Hermes endpoints, which usually takes one to three days depending on how many custom tools you wired into Retell. Hermes ships a prompt importer that ingests Retell agent JSON exports directly. Knowledge bases need to be re-uploaded since vector schemas differ between providers.

How much does the migration actually cost in dollars and time?

BYO Twilio (Path A) costs zero dollars in port fees and roughly 4 to 8 hours of engineering time for a 1 to 10 agent book. Full port-in (Path B) costs $0 to $30 per number on the Twilio side and the same 4 to 8 hours of agent setup work. Third-party migration audits price platform-to-platform moves at $2,000 to $10,000 in dev time across the industry. The Hermes migration kit collapses that to a single afternoon for the agent rebuild and a separate port window if you choose to port.

Why are agencies leaving Retell in 2026?

The two most-cited reasons in 2026 review aggregators are cost unpredictability (headline rate of around $0.055/min lands at $0.13 to $0.31/min once STT, LLM, TTS and telephony are layered in) and missing white-label, which is a deal-breaker for agencies reselling under their own brand. Retell is a strong builder platform. It was not built around agency multi-tenancy.

Can I keep my Retell account active while I migrate?

Yes. The recommended migration window runs Retell and Hermes in parallel for 7 to 14 days. New clients land on Hermes, existing clients stay on Retell until their next billing cycle, and you flip individual production numbers one at a time after validating each agent. You only cancel Retell once the last production number has cut over and you have a full month of clean Hermes invoices.

What if I want to keep some clients on Retell and move others to Hermes?

That works. Twilio sub-accounts are isolated, so each client's numbers can point to a different platform. Hermes was built around multi-tenant workspaces, so partial migrations are a first-class flow rather than a workaround. The clients you move first should be the ones whose Retell bill exceeds your retainer margin (run the audit in our 5-invoice problem post).

Where this leaves you

The hard part of leaving Retell is not the carrier side. The numbers are not the bottleneck. The bottleneck is the agent rebuild, and a clean prompt importer collapses that from weeks to an afternoon. Pick Path A. Validate on the sandbox. Roll one client per day. Cancel Retell on day 14. Keep Twilio if you still need it for anything else. Run the cost audit once the first full month of Hermes invoices closes.

By builders, for builders. The migration was painful enough on our own first agency that we rebuilt the importer twice. The version you get now is the one we wish we had on the first cutover.

/ next step

Migrate from Retell with zero downtime

Apply to the Founders' Beta. Drop your Retell agent JSON exports into the importer. Validate on a sandbox number the same day. Cut over one client at a time. Cancel Retell on day 14.
Apply for the Founders' Beta Run the cost audit
Alfredo Romero is CEO of Hermes, the voice infrastructure platform for AI agencies. Connect on LinkedIn.
AR
/ written by

Alfredo Romero

CEO and Co-Founder, Hermes

Alfredo runs sales, operations, and strategy at Hermes. Before founding Hermes he ran agencies for nine years and spent the last three building the AI voice operations side. He writes the operator playbook from real builds, not theory.
LinkedIn ↗ X (@buildwithhermes) ↗ About the founding team →