Dhanur AI
Docs menu· Limits and costs

Limits and costs

Live

How pay-as-you-go credits work, what things cost, free credits, top-ups, low-balance alerts, the limits in place, GST invoices, and what is coming next.

Pay as you go

Dhanur AI has no monthly plans or fees. Your AI team uses credits as it works, and you top them up when you need more.

  • 1 credit = ₹1.
  • Free credits: every new workspace gets ₹25 of credits, in both its test and live wallets. You don't need a card to start.
  • Top-ups: from ₹50 to ₹1,00,000. Packs of ₹200, ₹500, ₹1,000 and ₹2,500 have no bonus. ₹5,000 or more adds 5% bonus credits, and ₹10,000 or more adds 10%.
  • GST: 18% is added when you buy credits, never on usage.
  • No expiry: credits stay in your workspace until you use them. The oldest are used first.

Each workspace also includes 2 seats (members plus pending invites) and 1 project. If you need more, email outreach@prodigalai.com. See Pricing for the current prices.

What things cost

Everything your agents do is priced in credits. Prices follow what we pay for AI models and connected apps, with our margin included.

Work Price
A reply from an agent Depends on how smart the agent is (Smart, Smarter or Smartest) and how much it reads. The agent builder shows about what a reply costs. That figure allows for the agent looking something up first, which most replies do.
Other AI work Charged by the text it uses, at the same model rates. This covers Describe-it drafts, Write with AI, test checks, AI table columns and AI steps in your tools.
A connected-app action About ₹1.39 for each 30 seconds it runs
An email an agent sends or receives About ₹0.21
A WhatsApp message an agent sends Meta's fee with our margin: about ₹0.18 for a reply, more for a marketing message. Replies inside the 24-hour window are free until 1 October 2026, then they carry this fee.
Importing website pages, uploading knowledge files, starting tasks from schedules or webhooks, and calls to your own MCP server's tools No charge yet. We'll give you 30 days' notice before it costs anything

A WhatsApp or email conversation costs the AI plus the channel's own fee for each message. A short WhatsApp conversation of four or five replies carries about half a rupee of Meta's fees from October.

Test chats are charged at the same rates. This covers the builder's test chat, evaluations, test bulk runs and AI column fills, because they use the same AI models.

To see what a piece of work used:

  • Each reply in a test chat shows its price.
  • Each task shows its total, and each step shows its own. See Tasks and approvals.
  • Photos and PDFs sent with a message are read by the model too, so they count towards the price of that step, and of later steps while the agent still looks at them.
  • Analytics adds it up by agent and by model.

These prices are what you pay, not what we pay our providers. Plan & usage has the current price list.

When credits run out

  • At zero or below, agents stop taking on new work. A test chat or tool shows "Out of credits — add credits in Settings → Plan & usage."
  • Customers still get an answer. A customer who writes to a live agent (for example through your website chat) is told "The team will get back to you shortly". The task is marked Escalated with the reason, so your team can reply.
  • Work already under way may finish. A reply that was being written when credits ran out can take the balance slightly below zero, by up to ₹20. The shortfall is taken from your next top-up. Anything beyond that is not charged.

Low-balance alerts

Owners and Admins get an email when the balance falls below a threshold, and another when it reaches zero. Each email goes once per drop; after a top-up, the alerts start again.

  • The threshold is 20% of your last top-up, or ₹50 if that is lower.
  • You can change it: on Plan & usage, set your own threshold or switch the emails off.

Spending caps

  • A cap on each task. When a task reaches it, the agent stops and the task shows "This task reached its spending limit, so the agent stopped."
  • A cap on each tool run. A run of one of your tools stops when it reaches its spending limit.
  • A daily cap across the platform. If all AI work on Dhanur AI reaches it in one day, customer-facing AI work pauses until our team has reviewed it. Tasks that are already running can finish their current turn.
  • An autonomy limit on each agent. It caps the steps an agent may take on one message. See Agents.

Plan & usage

In the app, open Settings, then Plan & usage (it is also in the menu under your name). Everyone in the workspace can see it; only Owners and Admins can add credits and change alerts.

It shows:

  • Balance: what you have left;
  • Top-ups: the packs with their bonus, GST and total, and a box for any amount from ₹50;
  • Usage for the last 30 days: Smart, Smarter and Smartest replies, app actions, emails and other AI work;
  • What things cost: the price list;
  • Low-balance alert settings;
  • Billing details: who your receipts and GST invoices are made out to;
  • Purchase history: the status of each payment, with an invoice for each paid one.

Billing details

Before your first purchase, an Owner or Admin adds:

  • the business or person's name;
  • your GSTIN, if your business is registered for GST (optional; add it to claim input tax credit);
  • the address, city and PIN code;
  • the state.

The state is the place of supply. Customers in Delhi, where we are registered, are charged CGST and SGST at 9% each. Customers in every other state are charged IGST at 18%. The total is the same either way. A GSTIN must belong to the state you choose.

You can edit the details at any time. Changes apply to new purchases; documents already issued stay as they are.

Adding credits

  1. Pick a top-up. Choose Add credits on a pack, or enter an amount. The first time, you are asked for your billing details.
  2. Check the total, including GST and any bonus credits. Paying means you agree to our Terms, Cancellation & Refund Policy and Privacy Policy.
  3. Pay on Razorpay's page by UPI, card or net banking. We never see your card or UPI details.
  4. Back in the app, Razorpay's return is checked with Razorpay and the credits are added. If you close the page too early, the credits are added when Razorpay tells us about the payment.

A payment link works for 60 minutes. A payment that didn't go through adds nothing. If money left your account anyway, we add the credits once Razorpay reports the payment, or refund it.

Refunds. A top-up can be refunded if you ask within 7 days of buying it and none of its credits have been used. After that, or once you have used any of it, the top-up isn't refundable, but the credits stay yours for as long as you need them. Failed or duplicate payments are always refunded. See Cancellation & Refunds.

Test mode. While payments run in Razorpay's test mode, the page shows Test mode — no real money moves. Pay with Razorpay's test card or UPI details; nothing is charged. Test credits are kept apart from real ones, and usage while in test mode is taken from test credits.

Tax invoices

Each paid purchase has a Tax invoice you can open from the purchase history, print, or save as a PDF. It shows:

  • Prodigal AI Technologies Private Limited, with its GSTIN, CIN and registered office;
  • your billing details, including your GSTIN if you added one, and the place of supply;
  • the credits with their SAC code, the taxable value, CGST and SGST or IGST, and the total;
  • the Razorpay payment ID and the date.

Bonus credits are free, so they are not on the invoice.

Numbering:

  • Invoice numbers look like DH-2026-00001.
  • Payments made in test mode are numbered separately (TEST-2026-00001). They are marked as test payments and are not valid tax invoices.
  • A document issued before our GSTIN was on record is titled Payment receipt instead.

Coming next

Next. Everything in this section is being built now.

  • A monthly spending limit for your workspace, on top of the alerts.
  • Credit notes for refunds.

Later

Planned. Spending limits for each project, and using your own AI provider keys so that model use is billed by your provider.

Last updated 24 September 2026

Something unclear or wrong? Tell us