Evidence
What we can prove,
not what sounds good in a story.
Tanyakan.ai is still young. Instead of publishing stories we cannot reference, here are capabilities backed by repeatable tests that you can verify during a trial.
Answering questions from a customer Excel file
Tested against a real customer payroll workbook—38 sheets and tables 99–241 columns wide. Answers were checked against manual calculations from the raw rows.
correct questions, 4 rounds
16/16
One workspace, many branches, no data mixing
Each WhatsApp number can have its own persona, flow, and documents. A building-supply number never answered a clothing-store question from the wrong source, including table calculations.
branches per workspace
No limit
Internal security review with fixes verified
A full internal review found 20 issues—1 critical and 5 high. All were fixed and re-verified. This was an internal review, not a third-party certification.
findings remediated
20/20
The bot admits when it cannot be certain
Given 3 of 47 branches and asked for a total, the bot refused to add partial data and explained that the source was incomplete.
numeric safety tests
3/3