BankerBro

Trust

How BankAI works

Last updated: 1 July 2026

BankAI is the AI assistant inside BankerBro. It is helpful for explaining a circular or drafting an email, but it is not a substitute for your bank's approved policy. Here is exactly what is under the hood.

This page is maintained by the BankerBro team. It reflects current app-visible controls and operator practices. It is not a legal certification or third-party audit. For questions, email hello@bankerbro.com.

The model

BankAI uses Google's Gemini family of large language models, called via a secure gateway. We do not host our own model and we do not train a model on your conversations.

What it is given

  • Your message.
  • A system prompt that tells the model it is a helper for Indian bankers, to be concise, to cite the RBI/SEBI/NPCI/GST source name when it makes a factual claim, and to say “I don't know” instead of making things up.
  • The recent messages in the current thread, so it can follow the conversation.

What it is NOT given

  • Your bank's internal systems or documents.
  • Other users' conversations.
  • A live feed of your account, transactions or KYC data.

Where the “trained on banking” claim comes from

We do not fine-tune the model on customer or bank data. The phrase means BankAI's system prompt, retrieval sources and knowledge-hub grounding are all banking-specific — RBI/SEBI/NPCI/GST/IT-Department circulars and our own knowledge-hub articles. The underlying model is a general-purpose Gemini model.

Limits — please read

  • BankAI can be wrong. Always check a regulatory answer against the original circular.
  • It is not legal, tax or financial advice.
  • Do not paste customer data, PAN, Aadhaar, account numbers or your bank's internal material into the chat.

Data & retention

Conversation threads are stored in your account so you can revisit them. You can delete a thread at any time; deleting the thread removes the messages from our database. We do not sell chat data and we do not share it with third parties beyond the model provider needed to generate the response.