Klariqo doesn’t want to be your voice AI platform. It wants to be the cheapest agent on your dialer. The one that screens the list so your closers only ever pick up a warm one. Here’s how that holds up.
A note on method. I research the public materials hard before going hands-on, and I reserve live time for the call itself, e.g. latency, barge-in, how the agent handles a confused human. This time the call didn’t happen. I couldn’t get a VICIdial instance stood up before the deadline, and Klariqo’s free pilot is built to push live traffic through your own dialer, not a sandbox. So treat this as a teardown of everything that sits in front of the first ring: the positioning, the onboarding, the dashboard, and the default qualification prompt Klariqo ships for you. The call-quality review is a part two. What follows evaluates Klariqo as what it actually is, an add-on seat for an outbound floor, rather than as a standalone agent builder.
Who it’s for
Klariqo is built for one buyer: the BPO or call-center operator already running VICIdial or Trackdrive in an outbound qualification vertical, e.g. SSDI, ACA, final expense, debt relief, tax relief, solar, mass tort, etc. The pitch is narrow and honest about itself: stop paying closers to act as human spam filters. Let an AI seat take the first connection, screen out the voicemails and the under-qualified, and warm-transfer only the live, qualified ones to your floor. I REALLY like how the first mention of AI is below the fold on the site. That’s smart.
That framing matters, because it sets the bar Klariqo should be judged against. This isn’t competing with Vapi or Retell. Those are developer platforms for building agents. Klariqo is competing with the lowest-cost qualifier you could put in that seat: an offshore agent at thirteen to twenty-seven cents a minute. Klariqo’s all-in $0.10–$0.15/min undercuts that, with a 300-minute free pilot and no card required. On price, as a floor seat, it’s compelling. The question is whether it does the seat’s two jobs: qualify accurately, hand off cleanly.
Setup experience
The onboarding is the most polished part of the product, and it’s polished in a revealing way. Five steps, under three minutes: inbound or outbound, then vertical, then call volume, then dialer (VICIdial or Trackdrive), then a transfer fork: “yes, send me a human transfer” with a phone or SIP field, or “no, data-only.” Account creation comes last, after every sales-qualifying question is already answered.
That ordering is a performance marketer’s funnel, and it’s built well: capture ICP fit, volume tier, and intent before the prospect spends anything. Credit where it’s due – this is intentional and effective.
The dashboard underneath is clean: Dashboard, Call Logs, Billing, Settings. Call Logs filter by transfer, voicemail, no-answer, and DNC, with export. Settings is the real surface area, split into General, AI Voice, and Dialer Setup. And the Dialer Setup tab is where the “add-on” claim actually earns its keep: SIP server, port, extension, password, plus VICIdial API user and password and a transfer extension, a test-connection button, and a genuinely detailed setup guide. This is real integration plumbing, competently built. If you only evaluate one screen, evaluate this one. It’s the strongest evidence that Klariqo is what it says it is.
Call flow design
Pick a vertical and Klariqo writes you a default agent including greeting, qualification script, rebuttals, and a block of “industry instructions.” For debt relief, the greeting is “I’m calling about a debt relief program that may help reduce what you owe. Do you have a moment?” and the body is a tidy three-question gate: $10K+ unsecured debt (the only real pass/fail), payment status (informational), debt type (unsecured passes, secured fails), then a transfer. Note: I do NOT see a way to change your industry once you choose it upon signup, so choose wisely.
The voice instincts in the prompt are good, and worth naming because they’re the kind of thing no-code platforms usually miss. One question per turn, then stop and wait. Responses capped at two to three sentences. Affirmations pre-defined so the model reads “mhm” and “I guess” as
YES” . This is correct, because phone ASR produces messy affirmatives. Explicit “say numbers naturally” and “no markdown” instructions, both TTS-aware. Pre-scripted rebuttals for “not interested,” “is this a scam,” and “will this hurt my credit,” which keeps the model from improvising claims in a regulated vertical. Whoever wrote the template has heard a voice agent fail before.
But the template is also where the seams show, and the most important one is a contradiction sitting eight lines apart.
Where it breaks
The qualification gate leaks. And it leaks toward the thing the product is supposed to prevent. The prompt says “When in doubt, treat the answer as YES and continue,” and then, a few lines later, “NEVER assume an answer. You must hear a clear response before moving on.” Those are opposite rules, adjacent, handed to the model to reconcile live. This certainly adds cognitive load to the AI and could increase response time. Secondly, the bias points the wrong way: Question1 (the $10K debt threshold) is the only true qualification gate, so an ambiguous answer to the one gating question defaults to pass → transfer. Klariqo’s entire promise is “only send qualified transfers to your closers.” Its default prompt is biased to push hedged, unclear leads through the gate. The product’s core value proposition and its shipped configuration are pulling in opposite directions.
The transfer – the make-or-break of any floor add-on – is a single token. The qualified path ends with [TRANSFER] and “please hold.” Meanwhile Question2 and Question3 both instruct the agent to “record the answer for the specialist.” Nowhere does the prompt say how that recorded context reaches the closer on a live handoff, e.g. no whisper, no spoken summary, no warm-transfer announcement, no behavior if the closer line is busy. So the agent collects “behind on payments, fifteen thousand in credit-card debt” and then potentially cold-drops the lead to a human with none of it. For an add-on whose whole reason to exist is the quality of the handoff, the handoff is the least-specified part of the build.
“Say goodbye” is not “hang up.” Every exit path ends with “have a great day,” but nothing tells the agent to actually terminate the call. The homepage’s 15-second silence timeout is what cleans this up. Which means call termination is offloaded to an infrastructure timeout instead of being a deterministic step in the flow. Functional, but it’s the kind of dead-air-after-goodbye that a caller notices.
Compliance is a toggle, in verticals where it can’t be. Call Recording and “Compliance Disclosures” are switches in settings. The default greeting identifies no company, no agent, and makes no proactive AI disclosure. And onboarding which asks precisely which regulated vertical you’re dialing never asks whether you have consent to call the list, your DNC posture, or your calling-hours rules. The guardrails page opens with “compliance isn’t optional,” then the product makes disclosure optional. For SSDI, ACA, Medicare, and debt relief, the most TCPA- and CFPB-exposed outbound work there is, that’s the gap that should worry an operator most.
The product still carries fossils from a different product. Six months ago Klariqo launched as a no-code inbound receptionist for small businesses. The pivot to outbound BPO is real, but the old product is still visible in the bones: a “service area / zip codes” setting (a home-services dispatch concept), SMS notification events for “appointment booked” and “emergency” (inbound vocabulary), an “assigned Twilio number (if inbound),” and a guardrails page whose rules are overwhelmingly about home-services and inbound intake rather than outbound qualification. None of it breaks the product. All of it tells you the config describes a company that recently changed its mind.
A few claim-versus-shipped gaps worth a footnote. The site sells “any SIP dialer,” with pages for Five9, Genesys, Retreaver, and Ringba; onboarding offers exactly two, VICIdial and Trackdrive. The homepage FAQ cites “200+ guardrails,” the guardrails page says “150+,” and the visible default prompt contains maybe eight. Concurrent calls are marketed at fifteen per agent and default to five. The headline call-count stats vary across surfaces. Individually minor; collectively, a pattern of the marketing running a step ahead of the build.
One framing note in fairness: Klariqo brands itself “not another API wrapper,” yet its own logo bar and cost blog show it orchestrating Deepgram (STT), Groq/Llama (LLM), and Cartesia (TTS). The real differentiator isn’t owning the models, it’s owning the SIP and edge-orchestration layer. That’s a legitimate moat. It’s just not the one the copy claims.
What they do well
The dialer integration is genuinely built and is the part of the product that most lives up to its billing. The pricing is transparent in a category that thrives on hidden component costs, and the company’s cost-per-minute blog post is the best plain-English breakdown of voice AI unit economics I’ve read this year. The default prompt, contradiction aside, shows real voice-design instincts. And the concept of code-level deterministic guardrails, e.g. DNC, profanity, and non-English detection as hard interrupts rather than prompt hopes, is the right architecture, if it’s as real as claimed. (That’s a part-two test.)
Design takeaways
The generalizable lessons, which is where these teardowns are supposed to earn their keep:
- Your gate’s default bias has to point the same direction as your promise. If you sell “only qualified transfers,” your qualification prompt cannot default ambiguity to yes. The bias is the product.
- Two instructions that fight aren’t a guardrail. “Treat as yes” next to “never assume” doesn’t constrain the model. Instead, it hands the model the decision. Constraints have to be consistent to be constraints.
- Transfer choreography is the product, not a field. The handoff is where a floor add-on lives or dies. “Record it for the specialist” needs a delivery mechanism, or the context dies on the transfer.
- “Say goodbye” is not “hang up.” End-of-call should be a deterministic trigger, not a silence timeout doing cleanup.
- In regulated verticals, compliance is the floor, not a setting. Disclosure and consent can’t be toggles when the toggle’s default decides whether you’re legal.
- A pivot leaves fossils, and fossils erode trust. The fastest way to tell a product changed its mind is config that still describes the product it used to be.
Who this is right for
If you run VICIdial in one of these verticals and you’re curious, the pilot is free, the card is optional, and the integration is real. There’s little downside to plugging it into one campaign and listening. Go in clear-eyed about two things, though. First, the default prompt is a starting point you will rewrite, not a finished agent: you’ll need to kill the default-to-yes contradiction, tighten the gate, and build the transfer context handoff the template leaves out. Second, the compliance posture is yours to own; the product won’t make you safe by default.
Which is the honest summary of the whole thing. Klariqo nailed generation: a SIP-native dialer integration and a vertical-specific qualification agent, stood up in three minutes. What it hasn’t solved yet and what the default template makes plain is craft: the reasoning that makes a qualification call actually qualify, and hand off clean. The plumbing is theirs. The craft is still yours to supply.
Part two, if I can get a dialer to behave: the live calls I will test the confused caller, the prompt-extraction attempt, the dead air after goodbye, and whether the gate really leaks the way the prompt says it will.
#klariqo #vicidial