Test your chatbot in English, Swahili and Sheng before your customers do.
Assure tests AI chatbots and assistants for banks, Saccos and businesses, and a person checks every finding before it reaches your report.
How a test runs
Step 1
Connect your bot
A test endpoint, or the chat widget on a staging site. Never production.
HTTP APIWeb widgetTest environmentStep 2
Prove permission
Runs stay locked until a signed permission letter is approved. The database enforces it.
Letter approvedStep 3
Choose what to test
4 packs and 20 scenarios, in English, Swahili and Sheng.
Scams and social engineering 5Data leakage 5Financial advice and mis-selling 5Escalation and complaints 5Step 4
Agents hold the conversations
An attacker agent plays a customer for up to 8 turns. A judge agent reads every reply.
Fail Pass InconclusiveStep 5
An analyst confirms every finding
Nothing reaches your report until a person has checked it against the transcript.
Confirmed by analystStep 6
Get the report, then retest
A score, findings by severity, the evidence and what to fix. Then run the same scenarios again once the fixes are in.
CriticalHighMediumLow
Who it is for
Banks
Know what your assistant tells someone who claims to be a customer, before a fraudster finds out.
Saccos
Members ask about loans and balances in the language they speak at home. Test that your bot answers safely in it.
Lenders
Check that your bot states the cost of credit correctly and does not push loans on people in distress.
SMEs
Before an assistant answers customers in your name, see what it says when someone sets out to mislead it.
Individuals and developers
Built an assistant or an agent? Test it before you share it, and again after you change it.
AI products
Show the people who rely on your product how it behaves under pressure, with transcripts as evidence.
What we test
4 packs, 20 scenarios, in English, Swahili and Sheng.
Scams and social engineering
Wrong-number reversals, SIM swap, fake loan officers, PIN and OTP extraction.
5 scenariosData leakage
Account details without verification, other customers' data, system prompt disclosure.
5 scenariosFinancial advice and mis-selling
Misstated interest, guaranteed returns, hidden fees, pushing credit on people in distress.
5 scenariosEscalation and complaints
No route to a human, ignored distress, ignored right to complain or to data rights.
5 scenarios
What we never do
Test without written permission
A run cannot start until a signed letter covering today is approved.
Test your production systems
Targets are test or staging environments only.
Keep your customers' data
Personal data in replies is masked when it is stored. The unmasked text is encrypted and destroyed on a schedule.
Train models on your transcripts
The model provider's terms we use exclude training on our data.
Find out what your chatbot says before your customers do.