Resources · FAQ
Questions buyers actually ask.
If yours isn't here, email [email protected] — a founder answers.
Product
Who is Sparring for?+
Two audiences on one platform. Practice is for people who have difficult conversations for a living — frontline support, sales, people managers. Arena is for teams that ship customer-facing AI agents and need to know how those agents behave under pressure, not just whether they answer correctly.
How realistic are the counterparts?+
Each has a visible stance, a hidden motive revealed only under a specific condition, assigned pressure tactics, and a temperature that moves in small steps only when earned. We play-test the whole library against the live engine before release and publish the flag rate. Read How it works.
Will the counterpart ever just agree because I was polite?+
No. Politeness earns attention, not agreement. Movement requires the specific thing the scenario defines — acknowledging the real issue, asking the right question, holding a line. The debrief shows you which.
What languages are supported?+
English is the base for product, content and support. Simplified Chinese ships for the foundation library and the interface. Learners choose per session. Further locales follow customer demand.
Is there voice?+
Text today. Voice practice is on the roadmap for 2027; the engine and judge are modality-independent.
Can we create our own scenarios?+
Yes — Studio drafts a scenario from your policy, SOP or notes; you review and publish privately to your tenant.
Data, models and security
Which AI models do you use?+
Frontier models from Anthropic through a pluggable gateway; GCP-resident inference via Vertex AI for tenants who choose it. You pay per session regardless of model. We don't expose model choice as a pricing lever because we change it when a better one ships.
Is our data used to train models?+
No. Never — not ours, not any provider's. Transcripts, scenarios and uploads are used only to deliver your reports and are retained per your policy.
Where is data stored?+
Google Cloud, asia-east1 by default; EU and US regions on request for Platform customers. With GCP-resident inference, transcripts never leave Google Cloud.
How long is data kept?+
Default 365 days for transcripts and debriefs, configurable per organisation. Learners can delete sessions at any time. Usage and audit records are retained for billing and accountability.
Do you have SOC 2?+
Not yet. SOC 2 Type I is planned for H1 2027, a third-party penetration test for Q1 2027. We publish the current status honestly on the Security page and answer questionnaires in writing.
What happens if a learner says something alarming?+
The simulation stops before any model is called, crisis resources appear, the session is locked, and — if you configure it — a designated contact is notified with the session id and category only. Detection is deterministic in English and Chinese.
Pricing and procurement
How is it priced?+
Per session: $4 per practice session, $2.50 per Arena session. Platform (dashboards, SSO, Studio, residency, API) is an annual tier from $12,000. Volume and seat bundles available. See Pricing.
Can we buy through Google Cloud Marketplace?+
Yes. Usage bills to your Google Cloud invoice and counts toward your commitment; Platform and volume terms come as private offers. See Marketplace.
Is there a free trial?+
A free pilot: 50 sessions over 14 days with a named contact and one scenario built from your material. No card required.
Who is the legal entity?+
Carnegie Intelligent Technology Limited (卡耐基智能科技有限公司), company no. 80704751, Hong Kong. Registered office: Room 5017, 5/F, Yau Lee Centre, 45 Hoi Yuen Road, Kwun Tong, Kowloon, Hong Kong.
Can we sign our own paper?+
For Platform, yes. Practice and Arena use our standard terms; our DPA with SCCs applies to all plans.
Ethics and HR
Can managers use scores in performance reviews?+
No. Our Acceptable Use Policy prohibits using practice data for hiring, firing, promotion or compensation decisions. Scores exist to help people improve.
Do learners know what managers can see?+
Yes, on every briefing screen. The organisation chooses a transparency setting; the learner sees it stated.
Can staff be evaluated without knowing?+
No. Covert evaluation is prohibited by the AUP and the product surfaces visibility to the learner by design.
Is this a replacement for human coaching?+
It's the practice that makes human coaching land. Debriefs give a manager or coach the quoted lines to talk about; the heatmap tells them what to coach next.
See it with your own scenarios.
Pilots start with your real policies, your real objections, your real customers — not generic demos.