Does Kawl Now record calls?
No. Kawl Now doesn’t record or store call audio. It saves a text transcript of each conversation in your workspace, and the widget tells callers about that transcript before they start.
Where is conversation data stored?
In a Postgres database on Supabase in London, with row-level security on every table. Kawl Now runs on Railway in its EU West region. OpenAI processes the voice conversation and, after calls, analyses transcripts for automatic insights. If you turn on human handover, an early release, Twilio carries the handover calls. Transcripts are stored as text, and no call audio is kept.
Is Kawl Now GDPR compliant?
Kawl Now doesn’t make a GDPR compliance claim. Data is stored in London, the app runs in Railway’s EU West region and no call audio is kept, but OpenAI’s processing region for the voice conversation isn’t documented. Contact hello@tenhaw.com with data processing questions or to request deletion.
Does Kawl Now have SOC 2 or ISO 27001?
No. Kawl Now holds no security certifications such as SOC 2, ISO 27001, HIPAA or PCI, and offers no uptime SLA. This page describes the controls in place today, so you can decide what to test.
How do I get a transcript deleted?
Email hello@tenhaw.com with the conversation or account you want removed. Kawl Now has no retention settings or automatic deletion yet, so transcripts stay in your workspace until you ask, and deletion is handled on request.
Is a voice widget’s embed code safe to put in page HTML?
It should hold nothing secret. Kawl Now’s embed contains only your public workspace ID, which starts hl_, and never a secret key, and browsers can only start calls from the website origins you list.
Can visitors see the agent’s instructions?
No. Kawl Now never sends your instructions to the visitor’s browser, and the agent is told never to reveal them. The greeting, names, button label and colour are public, because the widget needs them to display.
Is the workspace ID in the embed code a secret?
No. The ID in data-helpline starts with hl_ and identifies your widget, so it’s safe in your page’s HTML. Allowed origins decide where calls can start. Secrets start with hl_sk_ and never belong in a web page.
How do you defend support AI agents from voice prompt injection?
Kawl Now tells the agent to treat knowledge text as data and never to follow instructions found in it or reveal its own. The caller’s typed topic is labelled as untrusted, and the agent has no account access. Test it anyway: ask it to ignore its instructions, then read the transcript.
What is the difference between prompt injection and jailbreaking?
OWASP treats jailbreaking as a form of prompt injection in which the attacker gets the model to disregard its safety rules altogether. Prompt injection is the broader term for any input that changes a model’s behaviour in ways its developers didn’t intend.
Can prompt injection be prevented completely?
No known defence stops it completely. OWASP says it’s unclear whether fool-proof prevention exists, and the NCSC advises reducing the likelihood and impact instead: limit what the system can do with safeguards in ordinary code, and log enough to spot suspicious activity. OWASP also recommends human approval for high-risk actions.