Every lab partnership we can find so far is with Western software. The surfaces nobody covers are the business systems India runs on: no public sandbox, no traces on the internet, and no reason for a Western team to look.
Every lab partnership we can find so far is with Western software. The gap here is not positioning, it is that the data does not exist yet, and it sits in exactly these surfaces. Choose one to see where an agent actually fails on it.
Two and a half million businesses run on it. Desktop-first, licensed per machine. The integration surface is a local HTTP gateway on port 9000 that speaks XML, and since Release 7.0 (December 2025) also native JSON. What it still has no version of is a cloud endpoint: the process has to be running on a Windows machine, with the company file open, for anything to answer.
GSTR-1 filing and 2B reconciliation. Session-bound, captcha-wrapped, and built for a human at a keyboard. A model that has read every GST document still cannot complete a filing.
The accounting stack of the Indian small business. Zoho has a real API and Busy mostly does not, which makes them a useful pair: the same task, two very different action spaces.
MCA21, EPFO, ClearTax. Filing systems with no Western equivalent and no public sandbox, which means almost no training data exists for them anywhere.
The method is domain independent. Browser and desktop workflows, internal tools, anything where a person completes a task and the result can be checked against state rather than words.
San Francisco can build this. It is choosing not to.
Which one do you need?Nothing here is hard to pay for. A studio with money can hire, buy a Tally licence, and be running in a quarter. What it cannot buy quickly is the reason to bother.
A replica of Slack fetches around $300,000 and every lab already knows it wants one. A Tally voucher task is a $200 – 2,000 line item in a market where almost nobody has heard of Tally. Same effort, three orders of magnitude less obvious. That is the whole gap, and it is an attention gap, not a skill gap.
The second reason is structural. Every lab partnership we can find so far is with a company that owns its own surface — Benchling, Shopify, Stripe, Salesforce, Slack. Tally is not going to sign that deal, so there is no partner to route around, and no vendor gets a warm introduction to the problem.
Which means the moat is time, not talent. Tally integrators already exist in India and quote ₹50,000 – 2,00,000 for a connector. The window is about two years, and it closes the moment someone with a team decides this is worth a quarter. Our only real defence is publishing audited environments before that happens.