Asking your data in plain language: what a good answer includes

Sushrut Ikhar, Co-founder, Matih Labs —

Asking your data in plain language: what a good answer includes

TL;DR — "Chat with your data" has become a demo, not a product. A good answer is more than a paragraph. With Mitra, you ask a business question about a connected data source or an uploaded file, and the answer comes with a chart and the SQL behind it. You can ask a follow-up in the same conversation. You can ask in many Indian languages, or by voice, and you can ask from WhatsApp, from code, or from an AI tool you already use. This post goes through what each of those means and why we think they all belong in one answer.

Dashboards answer the questions someone predicted last quarter. Plain-language questions answer the one you have today. "Why did sign-ups drop last week?" is not a request for a table. It is a request for something you can read, check and build on.

That is the bar we hold Mitra to. Here is what it looks like in practice.

Where do the answers come from?

From your data. You connect a data source, or you upload a file such as a CSV or an Excel workbook, and then you ask. Both kinds of data work the same way. A founder with a spreadsheet and a team with a connected database use the same box to ask questions.

You write the question the way you would say it to a colleague. "Which products had the biggest drop in sales last month?" There is nothing to learn first.

Why does every answer come with a chart and SQL?

Because a number on its own is hard to use. Mitra gives you two things next to the written answer.

  • A chart. It shows the shape of the result, so a trend looks like a trend and a ranking looks like a ranking.
  • The SQL behind it. If you read SQL, you can check exactly what was asked of the data. If you do not, someone on your team who does can check it for you.

We put the SQL there because we do not think anyone, including us, should be trusted on the output alone. An answer you can inspect is an answer you can correct. The companion post, why a number should come with its SQL, goes into the ways a query can look right and be wrong.

What happens when the first answer is not the one you wanted?

You ask again. Conversations in Mitra keep their context, so a follow-up such as "now only for the West region" or "show it by week" builds on the answer before it. This is how most real analysis goes. The first question is rarely the last one.

When an answer is good, you can turn it into a dashboard. A dashboard built from answers can be presented, kept in versions, downloaded as a PDF, duplicated, and shared with your workspace. Dashboards can also be embedded. A signed embed carries fixed filters for each viewer. A public embed is open to anyone with the link.

Can you ask in your own language?

Yes. Mitra understands many Indian languages, with Sarvam models behind them, and it also takes voice. Most business questions in India are not first thought of in English. Making people translate their question before they ask it is a quiet tax. It is better to remove it.

Where else can you ask?

Where your team already is.

  • WhatsApp. Ask Mitra from a chat, the way you would message a colleague.
  • The public REST API. It lives at /api/v1, so you can ask from your own code.
  • An MCP server. It lets Claude, ChatGPT, Cursor and VS Code ask Mitra directly.
  • An agent skill and open-source SDKs in Python and TypeScript, released under Apache-2.0 at github.com/matih-labs/mitra-sdk.

An owner or admin creates API keys and MCP connectors in Settings > API & MCP. The API overview shows how to start, and the MCP overview covers the AI tools.

Who is it for, and how is it shared?

Mitra works in workspaces. A workspace has members with roles: owner, admin, member or viewer. Every action draws from the workspace's credits. You sign in with an email code, a phone OTP, a passkey or SSO.

FAQ

Do I need to know SQL?

No. You ask in plain language. The SQL is shown so that you, or someone who reads it, can check the answer.

What can I ask questions about?

Data sources you connect, and files you upload, including CSV and Excel.

Can I ask in Hindi or another Indian language?

Yes. Mitra works in many Indian languages, using Sarvam, and accepts voice.

Can I ask from WhatsApp or from my own app?

Yes to both. Mitra is available on WhatsApp, and the public REST API at /api/v1, the MCP server and the Python and TypeScript SDKs let you ask from your own software or from an AI tool.

Who can create API keys?

Workspace owners and admins, in Settings > API & MCP.