Founder-led, fixed fee, weekly working builds, source code yours. This page is the version you can read at midnight before deciding whether to book thirty minutes: how a project runs, what it costs, which tools we use, and the objections we hear most, answered.
You describe the workflow or the product. We ask about volume, systems, who decides what, and what "working" would mean. If it is not a fit, or a cheaper tool would do, we say so on the call.
You get: A recommended approach, a timeline and a firm price range. Sometimes a reason not to proceed.
A one- to three-page brief: scope, the systems touched, the rules, the measure of success, what is out of scope, the fee and the dates. You sign that, not a proposal deck.
You get: A document both sides can point at when the question "is this included?" comes up.
The decisions that decide whether it scales: the data model, the permission model, the rulebook an agent follows, the baseline we will measure against. Agreed before features are built.
You get: The rulebook or data model, a data-flow diagram, and a working skeleton on staging.
A running build on staging every week that you can use, not a status report. Slack or WhatsApp through the day, a weekly call, Loom for anything that needs showing. Scope changes are logged and costed, never silently absorbed.
You get: Something to use every Friday, and a log of what changed and why.
Production deploy to your accounts. The measure taken again. Documentation, runbooks, prompts and the repository handed over, with a session for your team. Thirty days of support after go-live.
You get: Your repository, your cloud and model accounts, the numbers, and a written plan for what comes next.
| Engagement | Timeline | Shape |
|---|---|---|
| AI Agent Pilot | 2–3 weeks | One workflow, two systems, approval step, measured result, handover. |
| One workflow in production | 4–8 weeks | Integrated, monitored, with a console and a retainer option. |
| Agentic workflow across systems | 6–12 weeks | Planning across several systems, guardrails, memory, review console. |
| MVP or product build | 4–14 weeks | Scoped to the one thing it must prove; Bubble for speed or Next.js and Supabase for scale. |
| Community platform or LMS | 8–14 weeks | Members, marketplace, courses, events and an admin console on one schema. |
Timelines assume one owner on your side who can answer questions within a day. The commonest cause of a slipped date is a system access request that takes three weeks.
Week one of a pilot is the week that decides whether the build is worth doing. This is what happens on each day, what we need from you and for how long, and what exists by the evening that did not in the morning.
From you: 45 minutes on the kick-off, then whoever can grant access. Nothing else.
By evening: A one-page working agreement, an access list with owners, the baseline sheet started.
From you: 90 minutes from the person who runs the desk or the inbox. The most valuable hour of the pilot.
By evening: Rulebook v0.1 and the escalation list, in your sheet, with the gaps marked.
From you: 20 minutes to read the tool list and strike anything that should never be there.
By evening: Tool list signed, data-flow diagram, a skeleton you can watch classify real messages.
From you: 45 minutes to supply the awkward cases from the last month and mark which replies were wrong.
By evening: A test set that runs on demand, the approval queue, templates in review.
From you: 30 minutes, and the decision. The owner on your side and whoever signs.
By evening: A yes, a no, or a changed scope, in writing; and either way the rulebook, the diagram and the baseline are yours.
About five hours of your time across the week, nearly all from the person who runs the workflow today, in the slots above. We work from Nagpur; the sittings land in your morning for the UK and your early morning for the US East Coast, or in Indian working hours for Indian clients. The commonest reason week one slips is Monday’s access list, which is why it is the first thing on it. The week is the first of the AI Agent Pilot; the rulebook it produces is the fill-in template, and the baseline follows what a pilot measures.
Ranges, not quotes. The number firms up on the scope call once we know which systems, how much autonomy and how much data work is involved. Every engagement is a fixed fee against the written brief.
| Engagement | What it covers | USD | INR (+GST) | Timeline |
|---|---|---|---|---|
| AI Agent Pilot | One workflow, up to two systems, an approval step, running on your data with your team using it. | $4,500 | ₹3,50,000 | 2–3 weeks |
| One workflow, in production | A single workflow automated end to end and wired into the systems you already run — inbox, CRM, WhatsApp, spreadsheets. | $8,000–$42,000 | ₹6,50,000–₹35,00,000 | 4–8 weeks |
| Agentic workflow | An agent that plans across several systems, exercises judgment and acts, with guardrails, logging and a review console. | $15,000–$60,000 | ₹12,00,000–₹50,00,000 | 6–12 weeks |
| Agent platform or product | Several agents sharing memory, permissions and a console, or an AI feature inside a product you sell. | $60,000–$150,000 | ₹50,00,000–₹1,25,00,000 | 3–6 months |
| Under 500 runs a month | $30–$120 | ₹2,500–₹10,000 |
|---|---|---|
| 500–5,000 runs a month | $120–$600 | ₹10,000–₹50,000 |
| 5,000+ runs a month | $600–$2,500 | ₹50,000–₹2,00,000 |
Model usage plus hosting, billed to your own accounts. Messaging carrier fees are extra. A retainer for changes and monitoring is optional and scoped on the call.
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.
Product and MVP builds are priced the same way. For reference, a scoped MVP runs from about $8,000 for a single-module build to $15,000–$45,000 for a multi-module platform; see the MVP development page for the breakdown.
Get a range for your case in two minutes · What an AI agent costs in India in 2026
Boring on purpose. Every piece is something your next developer will already know, and every account is yours.
Ketan Khamgaonkar (IIT Roorkee) scopes, architects and leads every build, with the SigmaticAI team. No account manager relaying messages.
We work from Nagpur, India, with 9:00–13:00 US Eastern and 9:00–17:00 UK covered. Slack or WhatsApp through the day, a weekly call, Loom when something needs showing.
Your repository, your cloud and model accounts, your Stripe. IP assigned in the contract. Nothing about the handover depends on us still being in the picture.
If a SaaS tool does your job, or a Zap would do, or the process does not exist yet, we say so on the scope call. A pilot that should stop at week one, stops.
Agents draft, prepare and act within rules. Anything that writes to a system of record or reaches a customer passes an approval step until the log earns it more autonomy.
The brief fixes the fee. Changes are written down and costed before they are built, so the invoice never surprises either side.
Read the case studies: The Raising Club, a five-module marketplace in production in the US, first commit 29 May 2026 and live 2 July 2026; Accsys, offline-first accounting with GST reporting. Read the reviews on Google and Clutch, quoted verbatim on the service pages. Try the demo agents. Then book thirty minutes and ask anything.
Because you should know the number before you start, and because an hourly meter rewards slowness. The fee is fixed against a written brief; changes are logged and costed. We carry the risk of estimating badly.
It usually does, a little. A change is written down, costed and agreed before it is built. Small ones are often absorbed; we say so when they are. What never happens is a surprise on the invoice.
You do. Everything is built in your repository and your cloud and model accounts from day one, and IP is assigned in the contract. The handover is documentation and a session, not a transfer.
Thirty days of support are included. After that, a monthly retainer covers monitoring, rule changes, model upgrades and new integrations, scoped on the call. Plenty of clients run without one; the handover is written for that.
It stays in your systems and your region. Agents see the conversation or document they are working on, not the archive. The model provider does not train on your data under the API terms we use. We write the data-flow diagram into the handover so you can show it to anyone who asks, and design to the DPDP Act, GDPR or HIPAA as your project requires, with the caveat that we say upfront what we have and have not built under each.
Most of our clients are in the US, the UK and India. Billing in USD or INR, contracts in English, overlap hours as above.
The pilot exists for exactly this question: one workflow, two to three weeks, a fixed fee. If the workflow happens a handful of times a week, we will tell you a person with a checklist is cheaper. If it happens dozens of times a day, it is worth a conversation.
One person who owns the workflow and can spend two or three hours a week on it, access to the systems involved (read-only to start), and honest answers about how the work is done today. The rest is ours.
Bring whatever you have: a sketch, a spreadsheet, a competitor you like. You leave with an approach, a timeline and a price range.