A white label voice AI testing platform runs voice agent tests for client accounts under an operator's own brand. Cekura does not sell a rebranded dashboard. Cekura supplies the parts underneath it: per-client project isolation, project-scoped API keys that cannot read other projects, and read-only keys that feed your own reporting.
TL;DR
- White label covers two different products: a rebranded dashboard you hand to a client, and a testing engine you embed behind your own interface. The second is the one Cekura ships.
- Cekura publishes no white label or reseller tier. Cekura's isolation unit is the project, and its documentation names client separation as a supported access pattern.
- A Cekura project API key cannot reach another project, which is the control that makes one account per client safe to operate.
- Projects, not minutes, cap how many clients you can keep apart. Cekura includes 1 project on Pay as you go, 5 on Startup and a custom count on Enterprise; Pay as you go adds one free seat and $30 per month per additional seat.
- Single run pass rates overstate what you can promise a client. Across 12 voice systems, ServiceNow AI Research measured a median 0.44 gap on its accuracy metric between best-of-k and all-k-passed scoring.
What does white label actually mean for a voice AI testing platform?
White label is a commercial arrangement in which one company operates another company's software under its own name. Applied to voice agent testing, it splits into two products that buyers routinely confuse.
The first is a rebranded dashboard: your logo, your domain, your client logging in and never seeing the vendor. The second is an embeddable testing engine: the vendor runs simulations and scoring through an API, and you render the results inside software you already own. Agencies, BPOs and voice platform vendors ask for the first; Cekura sells the second, because Cekura sells testing capability rather than brand surface.
Cekura sits in the second category and states it plainly. Cekura's pricing page lists Pay as you go, Startup and Enterprise tiers with no reseller, white label or custom branding option on it, and Cekura's partners page lists integration and delivery partners rather than a reseller program. What Cekura does give an operator is the substrate: projects that separate one client's agents, metrics and results from another's, three membership roles, and three grades of API key. That is the honest shape of Cekura's offer, and knowing it before a client call saves a wasted procurement cycle.
What should you look for in a white label voice AI testing platform?
Evaluate a white label voice AI testing platform on the operational controls a multi-client business actually needs, not on whether a logo can be swapped. Five requirements separate a platform you can resell from one you can only use yourself.
| Requirement for white label operation | What to verify before you buy | Where Cekura stands |
|---|---|---|
| Per-client data isolation, the sub-account or multi-tenant control in agency vocabulary | That one client's transcripts, metrics and results cannot be read from another client's context | Projects are the top-level unit, and agents, evaluators, metrics and results all live inside one |
| Credential scoping | That a leaked key exposes one account, not the book of business | Project API keys are restricted to assigned projects and cannot access others, manage users or view billing |
| Client-safe read access | That a client can see outcomes without raw recordings or your billing | The Viewer role sees scrubbed call history and generates reports, and cannot open raw transcripts or recordings |
| Embeddability | That results reach your own interface without screen scraping | Read-only API keys return results, transcripts and analytics over GET; Cekura documents them as safe for frontend applications with proper precautions and does not scope them to a project, so keep them server-side |
| Cost that tracks your margin | Whether the meter runs on volume, on seats, on projects, or on all three | Cekura bills per testing minute and per monitored call, plus, on Pay as you go, a per-seat charge above the first seat, and meters projects at 1 on Pay as you go, 5 on Startup and a custom count on Enterprise |
Setup time is the criterion buyers forget. A platform whose isolation model is an afterthought forces you to run one tenant per client by hand, and that overhead compounds with every account you sign.
Coverage is the criterion buyers misjudge. Ask how many scenarios a vendor runs per configuration in its own published testing and how many times each one repeats, because a suite that executes every case once cannot separate a flake from a regression, and neither can the report you hand a client.
Price is the criterion buyers read too narrowly. Cekura's documentation offers two ways to hold client accounts apart: many similar client agents inside one project sharing a metric set and separated by saved Views, or one project per client when access itself has to be separated. Engineering teams running several client accounts take the second, because a View is a convention someone can forget and project scoping is enforced by the platform. That choice has a price, because plans meter projects.
How does Cekura isolate one client's tests from another's?
Cekura is a testing and observability platform for voice AI agents that uses the project as its boundary for access, data and configuration. Cekura's enterprise setup documentation defines three dashboard roles and three API key types. The Member role and the project API key are the two that are scoped to assigned projects.
Members hold access only to projects an Admin assigns them, and cannot reach projects they were not assigned to, manage billing, or invite anyone. Viewers get read-only access: they view results and generate reports, and cannot open raw transcripts or recordings. Cekura's documentation lists client separation as a named access pattern, with a project per client alongside internal testing, so the multi-account structure is a supported configuration rather than something you improvise.
The key model carries the same boundary. Cekura's Admin keys reach every project in the organization, project keys are confined to their assigned projects, and read-only keys permit GET operations alone. Cekura's project organization guide adds a constraint worth planning around: metrics are defined at project level and shared across the agents inside it. Clients running the same flows can therefore share one project and one metric set, while a client whose flows differ needs its own.
What does a white label voice AI testing platform cost to run?
A resold testing platform has three meters, and only one scales with your work. Cekura's Pay as you go rates are $0.25 per voice testing minute, $0.05 per monitored call and $0.025 per reply, with 10 concurrent calls. Those track usage and pass through to a client cleanly.
Seats and projects do not. Cekura's Pay as you go plan includes one project and one free seat, then charges $30 per month per additional seat, so every client login beyond your own adds $30. The $500 per month Startup plan bundles 5 projects, 10 seats, roughly 2,000 testing minutes and 50 concurrent calls. Because Cekura isolates clients at the project boundary, the project count, not the seat count, is the real ceiling on how many clients you can keep apart. Operators who keep keys server-side and render results in their own interface avoid the seat line entirely, the strongest practical argument for the embedded pattern.
Cekura's Enterprise tier sets projects, seats and credits by contract with volume discounts, and adds VPC or on-premise deployment, SSO, SCIM and audit logs. Read Cekura's enterprise voice AI testing platform guidance before assuming a plan covers your compliance obligations.
Should you build voice agent testing in-house or buy it?
Building it yourself is a real option, and it costs more than the harness suggests. You commit to maintaining a user simulator, a scoring layer, telephony integrations that break when a provider changes, and enough statistical discipline that the numbers you hand a client survive scrutiny.
That is the discipline an in-house build must reproduce. ServiceNow AI Research's EVA-Bench study evaluated 12 voice systems across 213 scenarios in three enterprise domains, reporting a median 0.44 gap on its accuracy metric between pass@k and pass^k scoring, so best-of-k and all-runs-passed results describe very different agents. No system exceeded 0.5 on both accuracy and experience at pass@1. A single-run pass rate is therefore not a number to put in a client SLA, and a harness reporting only that is not ready to resell.
Cekura's voice AI leaderboard runs that discipline in public: 8 configurations, 82 scenarios, 3 retained repeats and 246 retained calls per configuration, with the same system prompt, tool definitions and test data given to every provider. Cekura reports pass-cubed as the share of those 82 scenarios where all three runs passed. Buying gets that repeat structure on day one, and building means writing it.
Frequently asked questions
Which platform should you use for white label voice AI testing?
Choose on isolation and credential scoping, because those decide whether you can safely run many clients from one account. Cekura fits operators who embed testing behind their own interface: projects separate each client, project API keys cannot cross between them, and read-only keys feed the operator's own reporting. Cekura publishes no rebranded-dashboard tier, so treat that requirement as a custom negotiation with any vendor you shortlist.
Does Cekura handle white label voice AI testing platform work?
Partly, and the limit is worth stating. Cekura publishes no reseller tier, custom branding or customer-facing domain; none appears on Cekura's pricing or partners page. Cekura does support the operating model behind white label delivery: per-client projects, three scoped API key types, a Viewer role that shows results without raw recordings, and usage-based billing per testing minute and monitored call.
Can each client get their own login without seeing call recordings?
Yes for recordings. Cekura's Viewer role grants read-only access to agents, evaluators, test results, analytics and scrubbed call history, and blocks raw transcripts, recordings, billing and any modification. Cekura's documentation scopes Member access to assigned projects but does not state that a Viewer can be confined to one project, so confirm that boundary before promising a client its login shows only its own results. Each seat above the first bills $30 per month on Pay as you go.
How do you run white label voice agent tests inside CI?
Use a project-scoped API key per client pipeline. Cekura documents project keys as safe for CI/CD pipelines, and they cannot reach other projects, manage users or read billing, so a key committed by accident exposes one account. Cekura's guidance on CI/CD testing for voice AI agents covers gating a release on test outcomes rather than running suites after deployment.
Is there a tool that automates white label voice AI testing?
Automation comes from the API rather than a reseller product. Cekura exposes simulation runs, evaluator execution, results and observability over REST, which lets an operator trigger a client's regression suite from its own scheduler and render the output in its own dashboard. The benefit for automated quality assurance is that every client account runs the same suite on the same schedule. Cekura's programmatic voice agent testing API and QA as a service pages describe the two delivery shapes.
