An agent is software that talks to your customers and reads your systems, and both change. Rules need updating when the business does, facts go stale, a model version retires, a provider changes its API, and someone has to read the log. This page prices that work in three tiers, says what each includes in hours and in outcomes, and is honest about the businesses that are better off without one.
One agent, live, that must keep working: monitoring, fixes, facts and rulebook changes.
The agent keeps working, and someone who is not you reads the log every week.
One or two agents you want to get better each month: Care plus a monthly improvement sprint and model upgrades.
The agent gets measurably better every month, on the metric you chose.
Several agents or a customer-facing one at volume: Improve plus a weekly ops review with your team, incident response and the quarterly re-measure.
Several agents, or a customer-facing one at volume, run like a service with a named person behind it.
Fixed fee against a scoped brief, agreed on a free 30-minute call. Changes are logged and costed, never silently absorbed. Source code, prompts and accounts are yours. Prices are proposals until confirmed on the call.
Model usage, hosting and messaging fees are billed to your own accounts, so you see every rupee and nothing is marked up. At small-business volumes they are tens of dollars a month; at a customer-facing agent handling thousands of conversations they are hundreds. The bands below are the same ones the cost estimator quotes.
| Volume | Runs a month | USD / month | INR / month |
|---|---|---|---|
| Low volume | Under 500 runs a month | $30–$120 | ₹2,500–₹10,000 |
| Mid volume | 500–5,000 runs a month | $120–$600 | ₹10,000–₹50,000 |
| High volume | 5,000+ runs a month | $600–$2,500 | ₹50,000–₹2,00,000 |
Automated monitoring: errors, escalation spikes, cost anomalies, a dead integration. Alerts come to us first.
The ten-minute agent review from the template library, done by us: five transcripts read, every escalation read, one facts or rulebook change made, opt-outs verified.
Improve and Operate: the improvement sprint shipped and the one-page report sent, with the metric against baseline.
Operate: the re-measure, two weeks on the baseline sheet, compared with the pilot numbers, and the decision whether to promote, extend or retire.
The weekly review and the re-measure use the templates in the template library and the method on what a pilot measures, so you can run both yourself if you ever stop.
A draft-only agent on one channel, owned by someone in your team who reads the log on Monday mornings, can run for a year without us. The handover is written for exactly that: prompts, tool code, the runbook, the approved-facts list and the account access are yours, and the weekly review template is on this site. Thirty days of support are included after every build regardless.
Buy Care when nobody in the business will read the log, or when the agent writes to a system of record. Buy Improve when the agent is customer-facing and you want the metric to keep moving. Buy Operate when there is more than one agent, or when an outage would cost you a day of bookings. If none of those is true, say so on the call and we will tell you to keep the money.
The pilot that comes first · All the pricing bands · What the handover includes
No. Thirty days of support are included after every pilot and every build, and the handover is written so the agent runs without us. Roughly half of small-business agents need nothing more than that. The retainer is for the cases where the agent is customer-facing, writes to systems of record, or where nobody in the business will read the log.
Engineering time on your agent: changes, fixes, upgrades, the weekly review, the reports. Monitoring that runs on its own does not count. Unused hours do not roll over, and if a month needs more than the tier includes we say so before doing the work and quote it as a scoped change, logged and costed.
Monthly, with a month’s notice either way. Stopping means the agent keeps running on your accounts with your code; we hand back anything of ours still in the loop and leave the runbook current. Nothing is held hostage.
Because the work does not scale with conversations; it scales with change. A quiet agent whose business changes every month needs more engineering than a busy one whose rules are stable. Hours are the honest unit, and the running costs that do scale with volume are billed to you directly at cost.
On Care we update and test within the included hours when it happens. On Improve and Operate we evaluate the replacement against your real transcripts before switching, and the monthly report says what changed and why. This is the single most common reason an unmaintained agent stops working, usually about a year in.
The AI Agent Pilot is a one-off $4,500 (₹3,50,000); builds are fixed fees against a brief. The retainer is the only recurring fee, and it is optional. All the bands are on the how-we-work page.
Which agent, on which channel, how many conversations a month, and who reads the log today. We reply with the tier that fits, or with "you do not need one".