What we build01 of 12Agents18 things we build
Whole jobs wait until someone gets to them.
AI agents take whole jobs off your team, in your software and on the web, and a person approves what leaves.
Tested on twenty of your own past jobs before it goes live, and kept working as AI changes.
Eashan Gupta, founder. You meet him on the first call.
Make your own
Make your own agent.
When an email asks for a price, it drafts the quote from your price list. Above your limit, someone on your team says yes. It goes back as an email, ready to send.
Its first job
- 09:12Email from Mehta Traders: a price for 40 office chairsRead
- 09:13Quote drafted from your price listDrafted
- 09:13Over your limit, so it waitsFor your yes
Sample data · Powered by Agentec
All 18
Every AI agent we build.
Each one opens with what we build, three examples and what we don't promise.
A sample morning: every job the AI agents did, on one record, with the ones a person said yes to.
-
An AI agent for one whole job
Takes one repetitive job off your team and does it every day, asking a person before anything leaves.
What we build
- A map of the job with the person who does it today: every step, input and exception.
- An agent connected to your mail, sheets and business software, with only the access this job needs.
- An approval on your phone for anything that sends, pays or changes a record.
- A daily log of what the agent did, what it skipped and why.
For example, in your trade
- Manufacturing. An AI agent reads each purchase order that arrives by mail, enters it in the ERP and flags any price that doesn't match the contract.
- Moving. An AI agent turns every booking request in the inbox into a job in the transport system and asks the customer for a missing pickup address.
- Law firms. A law firm's AI agent opens a matter for every new instruction, runs the conflict check and asks the partner to confirm.
What we don't promise. It won't handle every exception on day one. New kinds of case go to a person until the agent has seen them.
-
A company assistant your team can message
Looks up orders, cases and stock for staff and raises their requests, from WhatsApp, Slack or Teams.
What we build
- One assistant in your team chat that answers from your files and shows the source.
- Actions it can take for staff: raise a request, update a record, draft a reply.
- Roles, so the warehouse sees stock and finance sees money.
- Approval cards for anything that changes a record or reaches a customer.
For example, in your trade
- Store chains. Store managers ask the assistant for stock at another branch, and it raises the transfer for the area manager to approve.
- Relocation. Consultants ask in the team group what a case needs next, and the assistant answers from the case file with the source.
- Hospitality. Front desk staff ask in Hindi where a guest's late checkout request stands and get the answer from the booking system.
What we don't promise. It only knows what is written down. Where the answer isn't in your files, it says so and asks a person. On WhatsApp it answers from your own records, as a service of your business. A general assistant runs in your own app, the browser, Slack or Teams, because WhatsApp's business rules have barred those since 15 January 2026.
-
Research on demand, with every source
Ask about a market, a company or a rule and get a sourced report, every claim linked.
What we build
- A research agent that plans the questions, searches the web, filings and your own files, and cites every line.
- Report formats in your house style: market scan, company profile, supplier check, regulation note.
- A source check that flags any claim without a link before you see it.
- A library of past reports your team can search and build on.
For example, in your trade
- Manufacturing. Before a sales trip, a report on the ten largest buyers in a new state, with their plants, owners and recent news, every line linked.
- NGOs. A sourced brief on which foundations funded water projects in Rajasthan in the last three years, and how much each gave.
- Finance. Before lending to a new distributor, a lender gets a sourced brief on its directors, related companies, court cases and news.
What we don't promise. A report is only as complete and current as its sources. Every claim carries a link and a date, and the gaps are named.
-
Your software, opened safely to AI agents
We connect AI agents to the software you already pay for, with access you can see and switch off.
What we build
- Safe ways in to your accounts, CRM, mail, storage and messaging, built to one open standard.
- Read and write permissions set per agent and per system.
- A sign-in vault, so no password ever sits inside an agent.
- A record of every call an agent makes into your systems, with a receipt.
For example, in your trade
- Store chains. One shared way in lets every AI agent read stock, orders and returns from the store system, but only the finance agent can issue a refund.
- Larger firms. A shared services team opens its ticketing, HR and finance systems to AI agents through one door, with a log the auditors can read.
What we don't promise. Some software offers no way in. Where there's no door we say so, and look at an agent that works the browser or the desktop instead.
-
An AI agent for every person on the teamBuilt before
Each person gets an AI agent for their own tasks that knows their job and follows company rules.
What we build
- A setup that gives each person their own agent, with their role and duties.
- Shared how-to steps the company writes once and every agent follows.
- Company guardrails: what data each role sees and what needs a yes.
- A usage view that shows which agents save time and which sit idle.
For example, in your trade
- Agencies. Every account manager at a 60 person agency has an AI agent that drafts the weekly client report from the project tracker.
- Associations. At a leadership workshop, each delegate scans a card and gets an AI agent that knows their role.
What we don't promise. Not everyone will use it. We measure use per person, and fix or remove the agents nobody needs. Each person's general assistant lives in your own app, the browser, Slack or Teams, since WhatsApp only allows answers from your own records.
Built before
- An agent for every delegate, for an industry body in Asia Pacific, at a leadership workshop.
-
A team of AI agents for big jobs
AI agents split up a tender, audit or month-end pack and check every part before a person signs off.
What we build
- A lead agent that splits the job into parts and hands each one to a specialist.
- Specialist agents for reading, pricing, compliance and writing, each with its own tools.
- A checker agent that tests every part against your rules before it reaches you.
- One board showing each part, which agent has it and what's waiting for you.
For example, in your trade
- Construction. A 300 page tender is read, priced against your rate book and drafted into a bid, with every exception marked for your estimator.
- Law firms. An audit firm's year-end file is split by section and each part is checked against last year's working papers, with gaps listed for the partner.
- Power contractors. A power company's environmental clearance file is split by chapter, each part is checked against the rules, and the gaps are listed for the project head.
What we don't promise. A bid or an audit file never goes out unread. The agents draft and check, and a qualified person signs.
-
Research across hundreds of items at once
Hundreds of AI agents check suppliers, prospects or listings in parallel and hand you one clean table.
What we build
- A list builder that turns your question into a row per item and a column per fact.
- Agents working in parallel that visit each website, catalogue or registry and fill the row with its source.
- A confidence mark on every cell, so you check the doubtful ones first.
- Exports to your sheet or CRM, refreshed on a schedule.
For example, in your trade
- Agriculture. Every certified organic turmeric supplier in three states, with capacity, certificates and contact, in one sheet.
- Brokerages. Every commercial listing above 5,000 square feet in two micro markets, with rent, floor and owner type, refreshed weekly.
- Store chains. Prices of your 300 best sellers on five marketplaces, checked every morning, with the gaps marked.
What we don't promise. Not every row will be right, because public sources go stale. Each cell shows its source and date, and doubtful ones are flagged. We read sites only within their terms and never get around a login or a captcha.
-
AI agents that work portals and websites for you
AI agents sign in to supplier, insurer and government portals to download statements, check status and fill forms.
What we build
- Browser agents that follow your steps on each portal and adapt when a page changes.
- A secure vault for logins, with a person entering one-time passwords when a site asks.
- A record of every page visited and every field filled, with screenshots.
- A daily run that drops files into the right folder and flags what failed.
For example, in your trade
- Law firms. For an accounting firm, AI agents download each client's statements from twelve bank portals on the 2nd of every month and file them by client.
- Diagnostic labs. A clinic's AI agent checks claim status on four insurer portals each evening and lists the claims stuck for more than a week.
- Moving. An AI agent pulls proof of delivery from three carrier portals and attaches it to each invoice before the invoice goes out.
What we don't promise. We never get around captchas or a site's terms, and a person signs in wherever a site asks for a one-time password.
-
AI agents that use your old softwareNew
An AI agent uses your older desktop software screen by screen, the way a trained clerk would.
What we build
- An AI agent on a locked-down computer of its own, taught your screens and steps.
- A check after every entry that reads the screen back and compares it with the source.
- A live view where your staff can watch and take over at any step.
- A queue of tasks it couldn't finish, with a screenshot of where it stopped.
For example, in your trade
- Manufacturing. An AI agent keys approved vendor bills into a desktop accounts package and reads each one back before saving.
- Diagnostic labs. A diagnostic lab's AI agent copies results from an old instrument's software into the patient system, with a technician approving each batch.
What we don't promise. On screens that change, it starts slower than a person and makes mistakes, so every entry it makes is read back.
-
AI agents that report to you every morning
Every morning the numbers, the overdue jobs and three things to chase arrive on your phone.
What we build
- Agents that read your systems overnight and check them against your rules.
- A morning brief on WhatsApp or mail with only what needs you, each line linked to the record.
- Alerts during the day when something crosses a line you set.
- A weekly summary for the leadership meeting, written the same way every week.
For example, in your trade
- Construction. Every morning, each site head gets yesterday's progress against plan, the materials short for tomorrow and the bills waiting for approval.
- Hospitality. The owner of three hotels starts each morning with last night's occupancy, rate and reviews, and the one review that needs a reply.
- Store chains. A brand gets a morning note of stock running out in the next ten days at each store, with the reorder drafted.
What we don't promise. It can't see what your systems don't record. The brief is only as complete as the data behind it.
-
An AI agent that chases until it's done
An AI agent follows up politely on payments, documents and approvals by WhatsApp and mail until each one closes.
What we build
- A follow-up agent that knows who owes what, by when, and how they like to be reached.
- A reminder ladder you set: a friendly note, a firmer one, then a call from a person.
- Replies read and understood, so 'paid yesterday' or 'sent the file' closes the item.
- A live list of everything outstanding, oldest first.
For example, in your trade
- Law firms. A CA firm's AI agent collects bank statements and invoices from 80 clients before the filing date and lists who still hasn't sent them.
- Education. A school's AI agent reminds parents about pending fees in Hindi or English and stops the moment the payment shows up.
- Recruitment. A staffing firm's AI agent chases selected candidates for their documents and tells the recruiter which offers are stuck.
What we don't promise. It doesn't collect money or documents by itself. It reminds within the hours and the tone you set, and a person handles disputes.
-
AI agents that learn how your company worksNew
AI agents remember each customer and correction, and write your procedures down as steps they follow every time.
What we build
- A library of step by step procedures written from your manuals, checklists and your best people's habits.
- Memory of each customer, supplier and past decision, kept in your own systems.
- A learning loop: when a person corrects the agent, the fix is written down and reused.
- A review page where you see, edit or delete what the agent has learned.
For example, in your trade
- Relocation. After a consultant fixes a lease clause once, the AI agent applies the same fix to every lease for that city.
- Hospitality. After the revenue manager corrects its group booking prices twice, the AI agent prices every group booking that way and shows the rule it learned.
What we don't promise. It doesn't learn without review. Everything it learns is shown to you, and you can edit or remove it.
-
AI agents that clear your software backlog
AI coding agents take small fixes and features, test them and send each change to your engineers to approve.
What we build
- Coding agents connected to your code, tickets and test suite.
- Rules and limits for each coding agent: what it may change, run and spend.
- Every change opened for review with the tests it ran and a plain summary.
- A nightly run on the backlog, with what's done and what's stuck reported each morning.
For example, in your trade
- Software companies. Each night AI agents take the bug tickets tagged small, fix them with tests and leave pull requests for the morning review.
- Agencies. An agency's AI agents update forty client websites for a new privacy rule and send each change for a developer's yes.
What we don't promise. Agents don't replace your engineers. Every change is reviewed and merged by a person who owns the code.
-
A new business line run by AI agentsNew
Launch a service where AI agents do the daily work while a small team supervises and answers for it.
What we build
- A service design: which steps agents run, which a person approves and which a person does.
- The agents, the customer channel and the back office built as one system.
- A supervisor console with the queue, the exceptions and the numbers.
- Weekly measures of quality, cost per job and what customers said.
For example, in your trade
- Relocation. A visa document check service where AI agents review every file overnight and two specialists approve and answer clients each morning.
- Education. A coaching institute runs doubt clearing for 5,000 students, with AI agents answering from its own notes and teachers taking the hard ones.
What we don't promise. A business run by agents still has people: a named person supervises it and answers for every customer.
-
An AI chief of staff for the owner
Runs the owner's diary, mail and travel, writes a brief before each meeting and chases every promise made.
What we build
- A morning brief on the owner's phone: today's meetings, the mails only the owner can answer and what's overdue.
- Meetings booked, moved and declined by a message, inside the rules you set for your week.
- Every mail sorted, with replies drafted in the owner's own words and waiting for a tap.
- A one page brief before each meeting: who they are, what was said last time and what's still open.
- Every promise made in a meeting or a mail tracked, with a nudge to whoever owes it.
For example, in your trade
- Manufacturing. Every morning a plant owner gets a short brief with the day's buyer meetings, the three mails only he can answer and a supplier who missed the date he promised.
- Law firms. A law firm's managing partner sends a voice note asking for lunch with a client next week, and her AI chief of staff offers three slots and books the one she taps.
- Brokerages. Before every site meeting a developer's MD reads one page on the contractor: last month's promises, the bills waiting and the delay on block C.
What we don't promise. Nothing is sent, booked or paid without the owner's tap, it reads only the accounts the owner connects, and on WhatsApp it never runs as a general assistant.
-
Candidates found for the roles nobody applies to
Finds people who fit a hard role in public profiles, past applicants and referrals, and drafts your recruiter's note.
What we build
- The role written with the hiring manager as the experience and skills that really matter.
- A search of public professional profiles, your past applicants and staff referrals for people who fit.
- A shortlist with the evidence for each match and what still needs checking.
- A personal first note to each person, drafted for your recruiter to edit and send.
- Opt outs and do not contact lists kept, and where each name came from recorded.
For example, in your trade
- Manufacturing. A cement company finds plant heads with kiln experience in three states, including two who applied for another role last year.
- Diagnostic labs. A hospital group finds ICU nurses with the right registration who live near its new branch.
- IT firms. An IT services firm finds engineers with payments experience in Pune for a bank project, with a note to each about the work.
What we don't promise. We use only profiles people made public and your own applicant records, honour every opt out, never score age, gender, caste or religion, and a recruiter sends every message.
-
Every code change reviewed by AI before it's merged
Every code change is checked for bugs, security holes and broken house rules, with one-click fixes for your reviewer.
What we build
- A review of every change as it's opened: bugs, security holes, missing tests and slow code.
- Your house rules written down once, so every review checks them the same way.
- Comments on the exact lines, each with a suggested fix your reviewer can accept or reject.
- A weekly read of what the reviews caught and what slipped through.
For example, in your trade
- Software companies. A product company's AI review flags a missing permission check on a new screen before the senior developer opens the change.
- Finance. A bank's in-house team gets every change checked against its own security rules, on a server inside the bank.
What we don't promise. AI review catches a lot but never everything, so a person still approves every merge.
-
A control room for every AI agent
See what every AI agent is doing, approve its work in one queue, cap its spend and stop it.
What we build
- One screen of every agent, its current job and its last ten actions.
- An approval queue on your phone for sends, payments and record changes.
- A spend limit for each agent and each month, and a stop button that works.
- Rules each agent loads before it starts: what it may touch and what it never does.
For example, in your trade
- Finance. An NBFC's operations head sees every AI agent working on loan files, approves the five letters waiting and pauses one that is stuck.
- Software companies. A product company caps each AI coding agent at a daily budget and sees which ones are waiting for a review.
What we don't promise. An agent can make a mistake. You'll see it, you can stop it, and you can trace what it did.
Most owners don't know
What AI agents can now do.
A sample morning at an old billing screen: an AI agent finds the order, types its lines into a bill and leaves it for a yes.
An AI agent that uses a computer like a clerk
It opens your old desktop software, reads the screen, then clicks and types.
A hundred researchers at once
A hundred agents each check one supplier, then hand you one table.
AI agents that write their own playbook
After each job it writes down how, and uses that the next time.
It works today. We keep it working tomorrow.
How it works- Work it out, most consultants stop here
- Build it, most software firms stop here
- Run it
- Answer for it
- We stay
Name the job. We'll show you the agent that would do it.
Bring it to 45 minutes with Eashan. You leave knowing which agent we'd build first, and why. Nothing to sign.
What stays yours
- Your accountsin your company's name
- Your codeevery line of it
- Personal dataabout your customers and staff, on your own machine unless you approve otherwise in writing
- Plain files and a how-to guideif you stop
Eashan Gupta, founder. You meet him on the first call.