Sahi’s Dale Vaz wants an AI financial team in every pocket

Kyro is Sahi’s agentic AI assistant, built into its app and web platform. It is designed specifically for investing and trading. That matters because general-purpose chatbots can hallucinate, miss a user’s portfolio context, and offer no clear accountability when financial guidance goes wrong.

Kyro uses a “vertical harness” around an AI model. It draws on a user’s Sahi activity, 28 skills written by Sahi’s research team, and more than 100 tools. For example, it can create alerts, conditional orders, scanners, or trading strategies from plain-language instructions. Calculations come from Sahi’s backend APIs, not the model itself.

Kyro is still not an investment adviser and cannot make buy-or-sell calls. Sahi says it has tested the assistant with more than 11,000 traders. Its longer-term goal is a team of AI agents that helps each Indian investor monitor risk, portfolios, and taxes in their own language.

Sahi tested Kyro with more than 11,000 traders in a closed user group. This gives the assistant a substantial early trial among real users, rather than limiting evaluation to laboratory demonstrations. The figures offer a snapshot of adoption and reported performance during its early testing phase.

During that trial, Kyro generated over 2.7 lakh responses. Sahi reported a 99.2% success rate, meaning the company classified nearly all those responses as successful under its testing measure. The article does not define the precise criteria behind that rate.

User return was also notable. Seventy-eight percent came back after their first interaction. Traders used Kyro for more than task execution: some treated it as a sounding board for losses and discipline problems, while others interacted in Gujarati, Tamil, and Marathi. These are early results from a closed group, not a claim about every investor.

Sahi’s central argument is that fintech solved access more effectively than outcomes. Indians can open trading accounts quickly, but many still struggle to decide what to do with their money. A market loss can leave investors unsure whether to buy, sell, or stop investing altogether.

Vaz links this problem to complexity. Jargon, large amounts of data, and too many tools overwhelm retail investors. They then face two unattractive choices: spend years becoming experts or trust someone who promises guaranteed returns. The gap between wanting to grow wealth and having the capability to act remains wide.

Regulator-led investor education has helped, but it has not closed that gap. Sahi presents purpose-built assistance as a possible bridge. Kyro aims to explain strategies in plain language and use personal context, rather than simply handing investors another stream of generic information.

Kyro reduces incorrect financial information by separating language from computation. The AI model interprets a request, but Sahi’s backend systems perform the calculations and return the result. This matters because general-purpose models can hallucinate, or confidently produce false information without admitting uncertainty.

For example, when a user asks to add four and five, the model calls Sahi’s calculator rather than calculating the answer itself. The same design routes relevant requests to backend APIs. Kyro’s constitution adds more safeguards: responses must be grounded in facts, include checkable evidence, and acknowledge when the system does not know.

These controls reduce, rather than eliminate, the risk of errors. Sahi also says the system has in-country inference, data logging, and full conversation traceability. Users still press the final buy or sell button, so Kyro supports decisions instead of replacing the investor’s responsibility.

Kyro currently helps investors turn intentions into practical tools. Users can set price alerts, create conditional orders, build custom scanners, and describe a trading strategy in plain language. Kyro can then convert that strategy into executable code, making complex workflows easier to create.

Its role is assistance, not direct investment advice. The assistant can use a trader’s Sahi activity and its expert skills to provide relevant help, but it cannot tell the user to buy or sell a specific investment. The user remains responsible for pressing the final transaction button.

Vaz said advisory features would require a separate SEBI investment advisory licence, which Sahi plans to pursue. Until then, Kyro has strict rules against buy-or-sell calls. Sahi is also previewing a Discipline Score that would flag overtrading, oversized positions, and revenge trading rather than simply recommend transactions.

SEBI regulations determine how far an AI financial assistant can go in India. Sahi designed Kyro to comply with those rules, rather than treating it as an unrestricted chatbot. This matters because financial guidance can affect real money, and users need a clear framework for responsibility and protection.

Kyro’s compliance design includes in-country inference, data logging, and full traceability of conversations. It also keeps the user in control: the person presses the final buy or sell button. At present, Kyro has strict rules against making buy-or-sell calls, so it supports research and execution tools without acting as an adviser.

Vaz said advisory features would require a separate SEBI investment advisory licence. Sahi plans to pursue that licence, but the article does not say when it may receive one. For now, regulation defines both Kyro’s boundaries and the safeguards surrounding its operation.

Investors still need to manage diversification, position size, and discipline because AI cannot magically make money for everyone. A chatbot may answer a stock question without knowing that half the user’s portfolio is already in banking stocks. That missing context can leave an investor unnecessarily concentrated.

Position size and behavior matter too. Sahi’s planned Discipline Score will flag oversized positions, overtrading, and revenge trading. These habits can turn a single loss into a larger problem. Kyro can provide context, evidence, and tools, but the user still presses the final buy or sell button.

Vaz’s broader point is that AI should augment investing skills, not replace the individual. Kyro is not currently allowed to make buy-or-sell calls, and its constitution rejects unsupported certainty. Future AI agents may watch risk, portfolios, and taxes, but investors will still need to make responsible decisions.