VritantaNextGen
Request a call
VritantaNextGen
Request a call

Roadmap

What is live, what we are building, and what comes later.

AI now answers your customers and does parts of your work, but sometimes it is wrong and nobody notices.

We check your AI, fix what hurts, and keep it healthy, and a named person signs every finding.

We also build the business software you run on, where AI only drafts and you approve.

Request a call

Where everything stands

Here is what you can use today, what we are making now, and what waits until it is paid for.

Open for bookings The Clinic: AI Checkup, Treatments (the fix) and the Care Plan

Live
AI Checkup · INDRA · The Lab · Vritanta AI · TaraLand · Homestay hub · Agent testing (Shiva): Live, on its own site (opens another site)
Building
Conversation reading, inside the Checkup · INDRA for gyms
Designed
Automatic weekly re-tests, for Care Plan clients · Your own software, built on INDRA · Router · Helper network
Later
Arena

Live means you can use it or buy it today, because it is already built and open. Building means we have approved it and are making it now, so you cannot use it yet.

Why we are building this

Businesses are putting AI to work faster than anyone can check it, and the mistakes are quiet ones.

The words we use

Agent
An agent is an AI program that carries out tasks for you, such as answering customers or sending invoices.
Signed record
A signed record is a note of what happened, sealed with a digital signature so any later change shows.
INDRA
INDRA is our own business software, and its live demo uses example businesses with made-up data.
Shiva
Shiva tests AI agents before they go live, and it now has its own website, linked here.

What goes wrong

  • A chatbot quotes a wrong price or invents a policy, and a customer holds you to it.
  • An automation stops working on a Tuesday, and nobody finds out until the month-end numbers look strange.
  • An agent reports that a job is done when nothing happened, and we have seen that in our own company.

So we are building a way to test AI before it goes live, watch it, and prove what we found.

What we do
We check the AI that talks to your customers, fix what is broken, and keep watching it afterwards.
What it means for you
You find wrong prices, false promises and silent failures before your customers do, instead of after they complain.
Next step and cost
Request a call and we start; the Checkup costs ₹9,999 (draft price), credited in full if you book a treatment.
How you can check us
Open the free INDRA demo and the example report on the Clinic page, then compare them with what we say.

What we do for you

Here is everything we offer or are making, with its true state, so you can see what is ready today.

  • Live

    Live means you can use it or buy it today, because it is already built and open.

  • Building

    Building means we have approved it and are making it now, so you cannot use it yet.

  • Designed

    Designed means we have written the plan down, but we have not started to build it yet.

  • Later

    Later means we will not start it until the money from our other work pays for it.

  • AI Checkup

    Live

    We study how your business and your AI behave, then write a short report on what to fix first.

    Price: ₹9,999 (draft price)

  • INDRA business software

    Live

    INDRA is our business software for bills, books, customers, staff and your website, where AI drafts and you approve.

    Price: The demo is free to open

  • The Lab, free tools

    Live

    The Lab has free tools for PDFs, images, video and text that run inside your browser, so nothing is uploaded.

    Price: Free

  • TaraLand

    Live

    TaraLand is a calm play-and-learn world for children, with no ads, no accounts and no tracking of any kind.

    Price: Free

  • Vritanta AI

    Live

    Vritanta AI answers questions by voice or text, using only the words on our own published pages.

    Price: Free

  • Conversation reading, inside the Checkup

    Building

    Software reads real conversations with your chatbot and flags where it loops, gives wrong prices, or claims unfinished work.

    Price: Price when it opens

  • Agent testing (Shiva)

    Live

    Agent testing now lives on Shiva's own website; open it there to read what it does today.

    Price: Price on request — ask us

  • INDRA for gyms

    Building

    A version of INDRA for gyms, handling members and batches, where AI drafts reminders and you approve them.

    Price: Price when it opens

  • Automatic weekly re-tests, for Care Plan clients

    Designed

    Every problem we find becomes a test, re-run every week and again whenever your AI provider changes its model.

    Price: Price when it opens

  • Your own software, built on INDRA

    Designed

    We plan to build software shaped to your business on top of INDRA, aiming for weeks rather than months.

    Price: Price when it opens

Try the INDRA live demo to see the Live software with example businesses and made-up data.

Draft price These prices are drafts, not final. Nothing is charged without a written quote that you accept first. Prices in dollars are for clients outside India.

What we do
We offer checks and care for your AI, and we make business software where AI only drafts.
What it means for you
You can see at a glance what you can use today and what is still being made.
Next step and cost
Request a call about a Live item; the Checkup shows ₹9,999 and the Care Plan says ask us.
How you can check us
Each card shows its status and its price, and the Live ones link to something you can open right now.

How a Checkup works, step by step

A Checkup is a short, fixed job, and this is each step of it, with who does what.

The AI Checkup, step by step

The Checkup has seven steps, from the first call to your decision, and each step names who does it. The part where software reads your chatbot's conversations is still being built, and data terms are not written yet.

  1. A personLiveThe callYou tell us what your AI does and what worries you, and we ask how your work really runs.
  2. A personBeing writtenYou say what we may seeWe explain what we take and why, and you agree in writing before we look at anything.
  3. SystemBuildingPrivate details come outNames, phone numbers and health details are removed before anything is stored, so our notes never hold them.
  4. Agent draftBuildingSoftware suggests problemsOur software suggests possible problems, such as loops, wrong prices and false claims of finished work.
  5. A personLiveA named person checks each oneA named person checks every suggestion by hand, and nothing reaches you until that person has signed it.
  6. A personLiveYour reportYour report says what we found, what it means for you, what to do next, and how to check us.
  7. A personLiveYou decideNothing changes in your systems without your yes, and you can undo any change we make with you.

Source: our offer list and draft prices on the Clinic page

What we do
We run a Checkup in seven steps, and a named person answers for what the report says.
What it means for you
You get a short, plain report you can act on, and your systems stay untouched until you agree.
Next step and cost
Request a call and we start; the Checkup costs ₹9,999 (draft price), credited in full if you book a treatment.
How you can check us
Compare the example report on the Clinic page with the steps here, and ask us anything on the call.

Our promises, and what we never do

These are the promises we make to every client, and each one is something you can hold us to.

  • You see the full price in rupees before you book, and no fees are added later.
  • A named person signs every finding, because AI helps us look but a person answers for the result.
  • We tell you what we could not check and what we could not fix, so you never guess.
  • Every report carries proof you can check yourself, which we explain in the section on how you can check us.
  • Your data stays yours: we collect only what the work needs, and we never sell or share it.
  • If a problem we find could protect other businesses, we share only a made-up copy, and only with your OK.
  • AI only drafts, people approve, and you can undo any change that we make together with you.
  • We mark every example as Example, and we show no fake reviews, no invented clients and no made-up numbers.
  • If you do not understand something we wrote, that is our mistake, and we will rewrite it.

We have not written our file-deletion terms yet, so ask on the call and we will write them down.

What we will never do

There are also four things we will never do, and no reason or request will ever change that.

  • We never send a message, publish anything, make a payment, or change your systems without a person saying yes.
  • We never test a real system unless you have given us written permission to test that system.
  • We never sell or share your data, and we never show a price that grows at checkout.
  • We never put ads, tracking or accounts into TaraLand, which is our calm play-and-learn world for children.
What we do
We keep a short list of promises, and we write them down so that you can hold us to them.
What it means for you
You know the price, who is responsible, and what we will refuse to do before you hand us anything.
Next step and cost
Read the care page for what is confirmed today; every price stays a draft until we confirm it.
How you can check us
Read them against our care page, which shows only the promises we have already written and confirmed.

How you can check us

We do not ask you to trust us, so each report is built to let you check our work yourself.

  1. A signed record

    A signed record lists what we looked at and what we found, and you keep its final signature. After you hold that signature, even we cannot quietly change the record without the change showing.

  2. A before-and-after

    Each fix comes with a before-and-after, so you can see exactly what changed in your own systems.

  3. A test you can run again

    Each problem we find becomes a test, and you can run it again whenever you want to check.

  4. A named person who signs

    A named person signs every finding, so a human being answers for it and not only a program.

The honest limits

This is the limit of what our records can prove, and we state it plainly.

“anything you already hold, or older than the latest third-party timestamp, cannot be changed without detection.”
  • A third-party timestamp is a dated stamp from an independent service that shows when a record already existed.
  • A newer record, or one you have not yet received, could still be changed by us before it is stamped.
  • The tools that make all this strongest belong to our agent testing work, which is still being built.
What we do
We build each report so that a stranger could check it, not only the person who wrote it.
What it means for you
You can verify our work without our help, and that means you never have to rely on our word.
Next step and cost
Ask for the signed record, the before-and-after and the test; they come with the report at no extra charge.
How you can check us
Check this very page: every number has a source at the bottom, and every status says what is built.

How the pieces fit together

Here is the whole platform on one page, from your business at the top to the fixes at the bottom.

The platform, stage by stage

Six stages run from your business, through agent testing and one signed record, to a person and the fix. Only your business at work is live today; the middle stages are being built, and the helpers are designed.

  1. SystemLiveYour business at workYour business runs on INDRA or on a bot of its own, and that is where the work starts.
  2. SystemBuildingAgent testingBefore an agent goes live, we test it safely with fake companies and fake keys, never real systems.
  3. Signed receiptBuildingOne signed recordEverything the testing does goes into one signed, linked record that nobody can quietly edit.
  4. Agent draftBuildingSoftware reads and draftsOur software reads real conversations and the record, then drafts a finding that a person must confirm.
  5. A personBuildingA person confirms, then we treatA named person confirms each finding, and only then do we fix the problem with your approval.
  6. A personDesignedHelpersInvited helpers test and fix problems for pay, working only on made-up examples, never on your data.

Source: our platform plan, version 1.1 of 27 October 2026, with every status as it stands today

The parts and where each stands

The platform’s parts, what each does, its status and its price
PartWhat it doesStatusPrice
RouterIt decides which AI model answers each request, using your written rules first and a fallback model if one fails.DesignedPrice when it opens
Helper networkIt is an invite-only group of testers and fixers, working on made-up scenarios only and paid for confirmed work.DesignedPrice when it opens
ArenaIt would publish signed rankings of agents on Indian business tasks, with error bars so that nobody is misled.LaterPrice when it opens

Anything here that is not live says Price when it opens, because we will not guess prices for unbuilt work.

What we do
We build in a set order, starting with the Clinic, then agent testing, and then the rest.
What it means for you
Each piece you can use today is marked Live, and the rest show honestly how far along they are.
Next step and cost
Apart from the Care Plan, everything not yet live says Price when it opens, so you can plan without surprises.
How you can check us
You can check Live items yourself: open the INDRA demo, the Lab, TaraLand and Vritanta AI right now.

Why we can build this

We think four things set us apart, and each is something you can look up, not take on trust.

  1. We already own the hard parts

    Our signing code already seals each log entry with a digital signature and links it to the one before it. So far only our own test agents use it, and no customer runs on it yet.

  2. We are our own first patient

    We ran our own company on AI agents first, and our About page tells what went wrong.

  3. A person signs, and we speak to owners

    A named person signs our work, and we speak to business owners in plain words, not only to engineers.

  4. Defences for behaviour, not only attacks

    Shared defences exist for attacks, but we found none for loops, false done claims, wrong prices or broken hand-offs. Immunity and Agent Threat Rules are two public projects that share defences against attacks; we link both in the sources.

Sources: Immunity protocol; Agent Threat Rules (Help Net Security, 3 June 2026).

What we do
We learned by running our own company on AI agents, and we show our failures before asking for trust.
What it means for you
You work with people who have already made these mistakes themselves and who know what to look for.
Next step and cost
Request a call and ask about our failures; if it fits, the Checkup is ₹9,999 (draft price).
How you can check us
Our own story, with its numbers and their dates, is on the About page for you to read.

The science we stand on

We build on published research, and we name the study so that you can read it and judge for yourself.

Accuracy can fall in three months

One study found GPT-4 got a prime-number question right 84% of the time in March 2023, but 51% in June.

  1. March 202384%
  2. June 202351%

Source: How is ChatGPT's behavior changing over time? arXiv 2307.09009

What we do
We read the research on how AI fails, and we turn each finding into a rule we can test.
What it means for you
You get methods that were chosen because evidence supports them, not because they sound clever or modern.
Next step and cost
Reading is free, so follow any link that interests you and bring your questions to the call.
How you can check us
Each row names its study and links to it, and the full list is under Sources at the end.

How we compete with big companies, honestly

Big companies have more money and data than we do, so we say plainly where we win and lose.

Where we win

  • We build for the owner of a small business, using words that owner can read without an engineer.
  • Our work is governed, signed and made for India first, with a named person signing each result.
  • We build from real failures, starting with our own, instead of from a list of features.

Where we lose

  • Giants offer far more AI models than we do, and they get new models on the first day.
  • They see far more traffic, so they learn more, and they give you one bill and staff around the clock.
  • If you only need your data kept inside India, the large cloud companies AWS and Azure beat us.

What we will never do

  • We will never fight on price, and we will never give away AI usage to win a customer.
  • We will never build a catalogue of five hundred models, and we will never claim fully automatic results.

The tools that watch AI were bought

Tools that watch how AI programs behave have been bought up: Cisco bought Galileo, and ClickHouse bought Langfuse.

Dynatrace agreed in August 2026 to buy Arize, and Palo Alto Networks announced it would buy Portkey.

As far as we found, they serve engineers, and none serves the owner of a WhatsApp bot.

Sources: Cisco announces the intent to acquire Galileo; Two acquisitions deep: what happens to open-source LLM observability (Hackernoon, a secondary report on ClickHouse and Langfuse); Dynatrace to acquire Arize (press release); Palo Alto Networks to acquire Portkey (India Entrepreneur).

What we do
We stay small and narrow on purpose, and we aim at one job the giants have left alone.
What it means for you
You get a supplier that tells you straight where a bigger company would serve you better.
Next step and cost
Sending the request form costs nothing, and if a bigger company fits you better, we will say so.
How you can check us
Check any claim we make against the vendors' own pages, which we list under Sources.

How our own test platform works (technical)

Under our products is one rule: the software drafts, a person approves, and there is a record. We built a research platform to test that rule, and we keep it internal.

Agent mesh (Elixir/OTP, a way of running many small programs that restart themselves when one crashes).
Each agent runs as a supervised process: if one crashes, its supervisor restarts it and the others carry on. It runs on our own server, for our own test agents. No customer uses it.
Signing kernel (Rust).
It signs each agent action with an ed25519 key, a standard digital signature, and links it to the one before, so a changed or missing record shows. So far it has signed only synthetic test agents' actions.
In our products.
INDRA carries the same idea in its own code: every approved action leaves a signed, linked record you can check in its demo.
  1. SystemAn agent drafts the action
  2. A personA person approves it
  3. SystemThe kernel signs it with ed25519
  4. Signed receiptAnyone with the public key can check it

Internal research platform; no customer runs on it.

Sources

Every figure here comes from a source below, and our researchers opened and read each outside source first.

  1. Cisco announces the intent to acquire Galileo
  2. Two acquisitions deep: what happens to open-source LLM observability (Hackernoon, a secondary report on ClickHouse and Langfuse)
  3. Dynatrace to acquire Arize (press release)
  4. Palo Alto Networks to acquire Portkey (India Entrepreneur)
  5. our own agent journal, as of 26 July 2026
  6. Immunity protocol
  7. Agent Threat Rules (Help Net Security, 3 June 2026)
  8. Why Do Multi-Agent LLM Systems Fail? (MAST), arXiv 2503.13657
  9. Which Agent Causes Task Failures and When? (Who&When), arXiv 2505.00212
  10. TRAIL: tracing agent errors, arXiv 2505.08638
  11. Where LLM Agents Fail and How They Can Learn From Failures (AgentDebug), arXiv 2509.25370
  12. tau-bench, arXiv 2406.12045
  13. How is ChatGPT's behavior changing over time? arXiv 2307.09009
  14. Anthropic, a postmortem of three recent issues (September 2025)
  15. FrugalGPT, arXiv 2305.05176
  16. RouteLLM, LMSYS blog (1 July 2024)
  17. LLMRouterBench, arXiv 2601.07206
  18. Dynamic LLM Routers are Often Misguided, arXiv 2610.02762 (preprint)
  19. The Leaderboard Illusion, arXiv 2504.20879
  20. OWASP Top 10 for Agentic Applications for 2026
  21. MITRE ATLAS explained (Speakeasy)
  22. AgentDojo, arXiv 2406.13352
  23. InjecAgent, arXiv 2403.02691
  24. The Tail at Scale, a secondary write-up of the paper by Dean and Barroso
  25. our offer list and draft prices on the Clinic page

Talk to us about your AI.

Tell us what your AI does and what worries you, and we will say plainly how we can help.
Request a call
AI Checkup ₹9,999, a draft price, credited in full if you book a treatment within 30 days.