A new direction for Dhanur AI: from writing content to doing the work

Update, 28 September 2026: early access is no longer free. Every new workspace starts with ₹25 of free credits and no card; after that you pay as you go for what your AI team uses, in rupees plus GST, with a tax invoice for every top-up. There are no monthly plans. See pricing.

Update, later on 17 September 2026: website chat is now live, so agents can answer customers on your website. Agents can also be started by schedules, webhooks and our API, and the app has a shared knowledge library and an analytics page. WhatsApp is still next.

Dhanur AI started in 2023 as a tool that helped people write social media content faster with AI. Today we are changing direction. Dhanur AI now builds AI employees for Indian businesses: agents that answer customers, qualify leads and handle routine back-office work, with a person approving anything that matters.

We are also doing something unusual with our own company. Dhanur AI is now run largely by AI employees, supervised by our founder, and we show their work in public.

This post explains why we changed, what you can use today, how we run the company, the lines we will not cross, and what we are building next.

Why we changed

Writing a post is rarely what holds a small business back. The harder part is everything that comes after it: the replies, the follow-ups, the "is this in stock?", the "can you send the price list?", the lead that nobody called back, the order details that someone has to copy into a sheet.

That work is repetitive, it never stops, and it lands on the people a small team can least spare. AI models have become good enough to take on a real share of it, as long as three things are true:

1. The business stays in control. An AI should not send a refund, change a customer record or reply to an angry customer without a person being able to check it first.
2. The cost is visible. Every AI step costs money. A business should see that cost per task, in rupees, instead of discovering it at the end of the month.
3. It works the way Indian businesses work. Customers write in Hindi, English and a mix of both, most conversations happen on WhatsApp, and data should stay in India.

Content tools help you say something. We want to help businesses do something. So we rebuilt Dhanur AI around that.

What Dhanur AI is now

An AI employee on Dhanur AI is set up the way you would brief a new hire:

• A job description in plain language: what it does, who it serves, what it must never do.
• Knowledge it can rely on, such as your FAQ, policies and price list.
• Tools it may use, and nothing else. A support agent that has no refund tool cannot give a refund, however a conversation goes.
• Rules for asking first. For each tool you choose whether it runs freely, always asks a person, or asks and then goes ahead after a waiting time (two hours by default) if nobody objects.

Every piece of work becomes a task. A task keeps the full conversation, every step the agent took, every tool it used, what it decided and what it cost.

What you can use today

We would rather be exact than impressive. As of today, these parts are live for everyone in early access:

• The agent builder. Start from a template (a support desk or a lead qualifier) or from scratch, write the instructions, add knowledge, pick the tools and choose a model.
• A safe test chat. Talk to your agent as a customer would. Nothing leaves the platform during a test, and a step-by-step trace shows what the agent did and what it cost.
• Built-in tools. Reply to the customer, search the knowledge, save a lead with a score, notify your team, send an email to a teammate or a saved lead, and hand the conversation to a person.
• Approvals. When an action needs a person, it waits in your inbox. You can edit a reply before it goes out, approve it, or reject it with a reason.
• Publishing with versions. Test changes on a draft, publish when you are ready, and pause a live agent at any time.
• A tasks inbox with everything that needs approval, everything handed to a person, and every error.
• Hindi and English. Agents answer in the language the customer writes in, and say they are an AI in their first reply.

Connected apps are in pilot. Through Pipedream Connect, workspaces can connect more than 2,000 apps, such as Google Sheets, Slack, HubSpot, Zoho CRM and Gmail, and let an agent use specific actions in them. Actions that only read data can run freely. Actions that change data ask first, and actions that delete something always ask. For now this is open to a small number of workspaces while we test it properly.

One honest gap: agents cannot yet talk to your customers on your website or on WhatsApp. Today you build and test them in the app. The website chat widget and WhatsApp are the next things we ship, and we will say so here when they are live.

How we run Dhanur AI

The best way to show that AI employees can be trusted is to depend on them ourselves. These are the AI employees that run parts of Dhanur AI today:

• Raghav, support, reads questions and bug reports, drafts replies and flags anything urgent.
• Vikram, sales, answers pricing and partnership enquiries.
• Tara, product, turns feature requests into roadmap items and groups repeats of the same idea.
• Meera, chief of staff, writes a daily company report every morning and emails it to our founder.
• Ishaan, marketing, drafts our LinkedIn and X updates from that report.

Two more, Kabir for platform operations and Ananya for finance, are set up but do not run on a schedule yet.

A few rules keep this safe:

• Sensitive messages wait for a person. When you write to us through the website, the right AI employee drafts a reply. Routine replies go out after two hours unless our founder holds them. Anything about refunds, legal matters, security, press or partnerships always waits for a person.
• Nothing is posted on our behalf without review. Social posts wait for approval.
• Every AI step is costed. A daily spending cap stops customer-facing AI work if it is reached, and a single switch pauses it at any time.
• Our AI employees say they are AI. They sign every message that way.

You can watch them work at dhanurai.com/company. The page shows the team, a live activity feed, daily updates and platform totals. It never shows who our customers are or what they wrote. Everything on it passes through a filter that removes names, contact details and message content.

What we will not do

• We will not train AI models on your data. Your conversations, documents and leads are yours.
• We will not move your data out of India without telling you. Our databases run in Bengaluru and our app servers in Mumbai. Update, 17 September 2026: photos and PDFs sent in chats are stored with DigitalOcean in New York; everything else is stored in India. See the privacy policy.
• We will not pretend an AI is a person. Agents built on Dhanur AI disclose that they are AI, and so do ours.
• We will not publish anything that identifies a customer, unless that business asks us to.
• We will not charge you by surprise. Early access is free. Paid plans will be priced in rupees plus GST, and we will announce them at least 30 days before anything is charged. Update, 28 September 2026: we replaced free early access and future paid plans with pay as you go, starting with ₹25 of free credits; see the note at the top.

What comes next

In roughly this order:

1. A website chat widget and WhatsApp, so your agents can meet customers where they already are, and so you can approve actions from WhatsApp too.
2. Knowledge from your files: PDFs, spreadsheets, documents and your website, with answers that cite where they came from.
3. Analytics and spending limits for each workspace.
4. Billing in rupees, with GST invoices.
5. Tools you can build yourself, and teams of agents that hand work to each other.

You decide the order as much as we do. Our public roadmap lists what we are considering and building, and signed-in users can vote from the app. Votes, demand from paying workspaces, fit with our plans and effort all feed into what we build first.

If you knew the old Dhanur AI

If you followed us for AI content tools, thank you. This is a real change of direction, and some of you will have questions about it. Write to us at outreach@prodigalai.com and a person will reply.

Try it

Early access is free and takes about a minute. Create a workspace, build your first AI employee from a template, and test it in the chat. Then tell us, through the Feedback page in the app, which job you want it to do next.

We will keep writing about what works and what does not, including our own mistakes. Follow along here, on the changelog, or through the RSS feed.