Solutions

What agents and skills can take off your plate

These are examples of agents and skills we can build. Each one has a person in the loop. We start with one, score every run, and only then talk about more.

Pilot terms: fixed-fee, typically 4 weeks, with a success metric agreed upfront. See how it works for the details.

CA and accounting firms

Illustrative phone screen for this use case, with sample data

Document chasing from clients

Reminds clients about missing documents and tracks who has sent what.

See details
The problem
Staff spend hours each month messaging clients for GST and ITR documents, and tracking who has sent what.
What the agent does
Sends polite WhatsApp or email reminders for missing documents, collects replies and files into a tracker, and flags clients who have gone quiet.
What a person still approves
The reminder wording and schedule, and any message to a client who is overdue by more than an agreed number of days.
What we score and measure
Share of documents received by the internal deadline, and staff hours spent chasing, compared with a baseline taken before the pilot.
Illustrative phone screen for this use case, with sample data

Invoice and bill data entry (Tally-ready)

Reads invoices and bills (PDF, photo), extracts fields and prepares a Tally-ready import file or a clean sheet.

See details
The problem
Purchase and sales invoices are retyped by hand into accounting software.
What the agent does
Reads invoices and bills (PDF, photo), extracts fields and prepares a Tally-ready import file or a clean sheet.
What a person still approves
Every batch before it is imported. Low-confidence fields are highlighted for a person to check.
What we score and measure
Field-level accuracy against your own checks, and minutes per invoice before and after.
Illustrative phone screen for this use case, with sample data

Due-date and compliance reminders

Builds a per-client calendar from the details you give us and sends reminders to the right people ahead of each date.

See details
The problem
Deadlines differ by client and are tracked in a mix of calendars and memory.
What the agent does
Builds a per-client calendar from the details you give us and sends reminders to the right people ahead of each date.
What a person still approves
The list of deadlines and the reminder content. Agents do not file anything.
What we score and measure
Reminders sent on time and any missed internal checkpoints, logged week by week.
Illustrative phone screen for this use case, with sample data

Collections follow-up for the firm's own fees

Sends courteous payment reminders with a UPI link, stops once payment is recorded, and keeps a simple ageing view.

See details
The problem
Your own professional fees are paid late because follow-up is awkward and easily forgotten.
What the agent does
Sends courteous payment reminders with a UPI link, stops once payment is recorded, and keeps a simple ageing view.
What a person still approves
Tone, the clients included, and any escalation. You approve before the first message goes out.
What we score and measure
Days to payment and the share of fees outstanding past an agreed date.
Illustrative phone screen for this use case, with sample data

Monthly client summary

Drafts a short plain-language summary per client from your tracker and books export.

See details
The problem
Clients ask the same questions: what is pending, what is due, what do I owe.
What the agent does
Drafts a short plain-language summary per client from your tracker and books export.
What a person still approves
The draft summary before it is sent to each client.
What we score and measure
Client queries received after the summary goes out, and staff time to prepare it.

Clinics and healthcare operations

Logistics only. Our agents give no medical advice, and patient data is used only with explicit consent. Read the full notice

Please read. Our agents do not give medical advice, diagnose, or recommend treatment. Messages are about logistics: appointments, reminders, forms and general clinic information approved by the clinic. Health data is sensitive: we collect and use it only with explicit patient consent and take extra care in how it is stored and who can see it.

Illustrative phone screen for this use case, with sample data

Appointment reminders and rebooking

Sends reminders on WhatsApp and offers available slots when someone asks to reschedule, using the slots your clinic supplies.

See details
The problem
Patients forget appointments, and rescheduling by phone ties up the front desk.
What the agent does
Sends reminders on WhatsApp and offers available slots when someone asks to reschedule, using the slots your clinic supplies.
What a person still approves
The reminder templates and the final booking for anything outside normal rules.
What we score and measure
Reminders delivered, rebookings completed without a phone call, and front-desk minutes per day.
Illustrative phone screen for this use case, with sample data

Follow-up after visits

Sends the clinic-approved follow-up message (for example, a reminder to book a review or collect reports) a set time after a visit.

See details
The problem
Follow-up messages are forgotten when the clinic is busy.
What the agent does
Sends the clinic-approved follow-up message (for example, a reminder to book a review or collect reports) a set time after a visit.
What a person still approves
The wording, written by the clinic. Anything a patient replies that sounds clinical goes straight to staff.
What we score and measure
Follow-ups sent on schedule and review visits booked after a reminder.
Illustrative phone screen for this use case, with sample data

No-show reduction

Confirms appointments ahead of time and fills a freed slot from a waiting list the clinic maintains.

See details
The problem
Empty slots from no-shows cost the clinic time and income.
What the agent does
Confirms appointments ahead of time and fills a freed slot from a waiting list the clinic maintains.
What a person still approves
The confirmation rules and who is offered a freed slot.
What we score and measure
No-show rate before and after, using the clinic's own appointment records.
Illustrative phone screen for this use case, with sample data

Intake form collection

Collects intake details through a short chat or form, with consent captured first, and prepares a tidy record for staff.

See details
The problem
Patients fill paper forms at the desk, and data is retyped.
What the agent does
Collects intake details through a short chat or form, with consent captured first, and prepares a tidy record for staff.
What a person still approves
Staff review the record before it enters the clinic's system. Consent is checked at the start.
What we score and measure
Minutes saved at check-in and share of forms complete before arrival.
Illustrative phone screen for this use case, with sample data

Routine FAQ answers

Answers from a fixed list the clinic approves.

See details
The problem
Staff repeatedly answer timings, location, fees and preparation instructions.
What the agent does
Answers from a fixed list the clinic approves. For anything else, it hands the chat to a staff member.
What a person still approves
The FAQ content and the hand-over rules. No medical questions are answered.
What we score and measure
Share of messages resolved from the approved list and hand-over accuracy on a sample.

Distributors and manufacturers

Illustrative phone screen for this use case, with sample data

WhatsApp order intake

Reads the message, transcribes voice notes, matches items to your product list and prepares a draft order for confirmation.

See details
The problem
Orders arrive as messages, voice notes and PDFs, and someone retypes them into the order system.
What the agent does
Reads the message, transcribes voice notes, matches items to your product list and prepares a draft order for confirmation.
What a person still approves
Every draft order before it is confirmed. Unclear items are queried rather than guessed.
What we score and measure
Draft-order accuracy on a sample you check, and time from message to confirmed order.
Illustrative phone screen for this use case, with sample data

Payment collections with UPI links

Sends reminders with UPI payment links based on a ledger export, and records responses and promised dates.

See details
The problem
Chasing outstanding payments from dealers takes daily calls.
What the agent does
Sends reminders with UPI payment links based on a ledger export, and records responses and promised dates.
What a person still approves
The list of accounts to remind and the tone. Disputed or sensitive accounts are left to your team.
What we score and measure
Average days outstanding and the share of reminders followed by payment.
Illustrative phone screen for this use case, with sample data

Dispatch and status updates

Sends dispatch and status updates from the sheet or system export your team keeps, and answers simple status questions.

See details
The problem
Dealers keep asking where their order is.
What the agent does
Sends dispatch and status updates from the sheet or system export your team keeps, and answers simple status questions.
What a person still approves
The message templates, and any exception such as delay or short supply.
What we score and measure
Status queries handled without staff and time spent answering them.

Startups and SaaS teams

Illustrative phone screen for this use case, with sample data

Support triage

Reads incoming tickets, tags and prioritises them, drafts replies from your documentation and routes to the right person.

See details
The problem
Support inboxes mix bugs, billing questions and sales enquiries.
What the agent does
Reads incoming tickets, tags and prioritises them, drafts replies from your documentation and routes to the right person.
What a person still approves
Replies before they go to customers, until your team is comfortable with the quality.
What we score and measure
Time to first response and the tags corrected by your team on a sample.
Illustrative phone screen for this use case, with sample data

Lead follow-up

Enriches the lead from information you provide, drafts a first reply and schedules follow-ups.

See details
The problem
Inbound leads go cold because follow-up is inconsistent.
What the agent does
Enriches the lead from information you provide, drafts a first reply and schedules follow-ups.
What a person still approves
Each outbound message, or a template you pre-approve.
What we score and measure
Response time to new leads and meetings booked, compared with your baseline.
Illustrative phone screen for this use case, with sample data

Internal operations agents

Automates a defined internal routine across your sheets, email and tools.

See details
The problem
Founders and ops people repeat the same internal tasks: reports, reconciliations, onboarding checklists.
What the agent does
Automates a defined internal routine across your sheets, email and tools.
What a person still approves
Anything that changes data in a system of record, until trust is established.
What we score and measure
Hours of recurring work removed and error rate versus manual.
Illustrative phone screen for this use case, with sample data

Pilot-to-production help for AI prototypes

Reviews your prototype, adds evaluation sets, guardrails, logging and a human-review step, and helps plan the move to production.

See details
The problem
A working demo exists, but it breaks on real inputs and has no monitoring.
What the agent does
Reviews your prototype, adds evaluation sets, guardrails, logging and a human-review step, and helps plan the move to production.
What a person still approves
Release decisions stay with your team.
What we score and measure
Pass rate on an evaluation set you agree on, and the failure types we find and fix.

Engineering and QA teams

Agents help your engineers. Nothing is merged, released or run in production by an agent. Read the full notice

How we approach this. These agents help your engineers; they do not replace review. Nothing is merged, released or pushed to production by an agent, secrets are never placed in prompts or logs, and release decisions stay with your team. We work inside your repositories and test environments under your access rules and an NDA where you need one.

Illustrative phone screen for this use case, with sample data

Test-case generation from specs and screens

Reads your specs, designs and existing tests, finds what is not covered and drafts new test cases in your framework's format.

See details
The problem
Test suites lag behind features, and gaps are found late, usually by users.
What the agent does
Reads your specs, designs and existing tests, finds what is not covered and drafts new test cases in your framework's format.
What a person still approves
Every proposed case before it is added to the suite. The agent flags where it is unsure instead of guessing.
What we score and measure
Coverage gaps found, share of drafted cases your team accepts, and engineer hours spent writing routine cases.
Illustrative phone screen for this use case, with sample data

Overnight test runs and build judging

Sorts overnight test failures into app bug, test issue or environment, with evidence.

See details
The problem
Failures from a long test run need triage each morning: real bug, flaky test, or environment problem.
What the agent does
Runs or reads the results of the night's tests, groups failures, sorts them into likely app bug, test issue or environment issue, and prepares a morning report with evidence. It does not report a pass without evidence.
What a person still approves
The verdicts on anything labelled a bug, and which failures get tickets.
What we score and measure
Time to triage after a run, share of verdicts your engineers agree with on a sample, and flaky tests identified.
Illustrative phone screen for this use case, with sample data

Spec-to-feature drafting

Drafts a feature from your spec and tests, and checks it against a requirement list.

See details
The problem
Building a feature from a spec, test cases and designs involves a lot of routine, rule-following code.
What the agent does
Builds a draft implementation from your spec and tests, follows your architecture and lint rules, and works through a checklist of every requirement. Missing inputs become listed blockers instead of silent gaps.
What a person still approves
All code goes through your normal review and CI. Merges and releases stay with your team.
What we score and measure
Requirements covered against the checklist, review comments per change and engineer hours per feature versus your baseline.
Illustrative phone screen for this use case, with sample data

Device and IoT test automation

Drives real phones and test rigs, for example mobile app automation together with relay-controlled devices, and records results for each run.

See details
The problem
Testing apps that control physical devices usually needs people to press buttons and watch results.
What the agent does
Drives real phones and test rigs, for example mobile app automation together with relay-controlled devices, and records results for each run.
What a person still approves
The test plan and which scenarios are safe to run unattended. Hardware safety rules are set by your team.
What we score and measure
Scenarios automated, hours of manual testing removed per release and defects caught before release.

Capital markets: broking and trading firms

Decision support for registered firms. No investment advice, no trading signals, no orders placed by our agents. Read the full notice

Important. This is decision-support and operations tooling for registered brokers, research desks, proprietary desks and advisory teams. GM Capital does not give investment advice or research recommendations, does not create or sell trading signals, and our agents never place or execute orders. Outputs are reviewed by the firm's qualified or registered staff. The firm remains responsible for its own SEBI compliance, and any use involving client communication or algorithmic trading should be checked with the firm's compliance officer first. Past performance and backtests do not predict future results. We do not claim to be SEBI registered. Pilot scope and fee for this area are agreed per firm.

Illustrative phone screen for this use case, with sample data

Market and news monitor

Sends a daily digest of news and filings on your watchlist, with sources.

See details
The problem
Analysts spend hours scanning news, filings and announcements across the stocks and sectors they cover.
What the agent does
Watches the stocks, sectors, filings and news sources you choose and sends a daily and intraday digest with sentiment summaries and links to the sources.
What a person still approves
Which sources and watchlists to use, and what the team does with the digest. The summaries support reading; they are not recommendations.
What we score and measure
Analyst time spent scanning, and relevant items caught compared with your own manual check.
Illustrative phone screen for this use case, with sample data

Strategy rule-checker

You give us your own written rules.

See details
The problem
Checking whether current conditions match the firm's documented strategies is manual and easy to miss.
What the agent does
You give us your own written rules. The agent flags when market conditions match them and drafts a note for an analyst. It does not place orders.
What a person still approves
Every flag and note. A human analyst decides what, if anything, to do.
What we score and measure
Flags that the analyst agrees were valid, missed matches against a manual review, and time from trigger to review.
Illustrative phone screen for this use case, with sample data

Idea assistant

Proposes variations of the firm's own strategies as research hypotheses, labelled as ideas, not advice.

See details
The problem
Teams lack time to explore variations of their existing strategies.
What the agent does
Proposes variations of the firm's own strategies as research hypotheses, labelled as ideas, not advice.
What a person still approves
Qualified people backtest and approve any idea before it is used for anything.
What we score and measure
Number of hypotheses reviewed and the share the team judged worth backtesting.
Illustrative phone screen for this use case, with sample data

Backtesting and journaling helper

Runs your strategies on your data and writes a weekly summary and journal.

See details
The problem
Backtests are run ad hoc and decisions are not recorded consistently.
What the agent does
Runs the firm's strategies on historical data the firm supplies, keeps a trade and decision journal and produces weekly performance and drawdown summaries.
What a person still approves
The strategy definitions, the data used, and the conclusions drawn. Results describe the past only.
What we score and measure
Reproducibility of backtests, journal completeness and hours to produce the weekly summary.
Illustrative phone screen for this use case, with sample data

Back-office and client operations

Triages client queries and drafts KYC and ledger follow-ups for approval.

See details
The problem
Client queries, KYC documents, contract-note and ledger reminders, and margin or pay-in follow-ups consume operations time.
What the agent does
Triages client queries, chases KYC and documents, drafts reminders for contract notes, ledgers and pay-ins, summarises calls and meetings, and keeps compliance-ready records.
What a person still approves
Every customer-facing message, especially margin and pay-in follow-ups, and the compliance records.
What we score and measure
Turnaround on follow-ups, documents received on time and operations hours per week.

Reusable skills for any team

A skill teaches an AI assistant how your team does a task, so anyone gets the same quality every time. Each run is scored, and the skill is improved from the misses.

Illustrative phone screen for this use case, with sample data

Business review skill

Reads your numbers, finds what changed, suggests fixes and learns which ones worked.

See details
The problem
Teams look at the numbers, try a fix, and rarely check whether it worked, so the same problems come back.
What the skill does
Reads the figures you share each week, points out what changed and where it comes from, suggests fixes, and records which ones were tried and whether the numbers moved. Later suggestions use that history.
What a person still approves
Every decision. The skill suggests and your team chooses what to try. It gives decision support, not financial, legal or tax advice.
What we would score
Suggestions your team accepted, and whether the numbers moved afterwards. We do not promise improvement.
Illustrative phone screen for this use case, with sample data

Agent testing skill

Tests your AI agent against real examples after every change and scores each run.

See details
The problem
A prompt or model change quietly breaks something that worked, and customers find it before you do.
What the skill does
Runs your agent against a set of real examples you agree on, scores every run, compares it with the last one and flags what got worse.
What a person still approves
Whether a change goes live. A flagged regression holds the change until a person reviews it.
What we would score
Regressions caught before release, and how closely the scores agree with your reviewers.
Illustrative phone screen for this use case, with sample data

Quote-writing skill

Teaches an AI assistant your template, wording and pricing rules.

See details
The problem
Quotes come out differently depending on who writes them, and pricing rules get missed.
What the skill does
Drafts quotes from your price list and template, checks discounts and scope clauses, and flags anything outside your rules.
What a person still approves
Every quote before it is sent, and any discount over your limit.
What we would score
Pricing rules followed and template fields correct, on every run.
Illustrative phone screen for this use case, with sample data

Code review skill

Applies your team's review checklist to every pull request.

See details
The problem
Reviews vary by reviewer and the same mistakes keep reaching the main branch.
What the skill does
Reads a change against your checklist, flags missing tests, naming and security issues, and drafts review comments.
What a person still approves
The review comments and the merge. Nothing is merged by the skill.
What we would score
Checklist items caught, and comments your reviewers accepted or rejected.
Illustrative phone screen for this use case, with sample data

Client onboarding skill

Runs the same onboarding steps every time and chases what is missing.

See details
The problem
New clients are set up differently each time and steps get forgotten.
What the skill does
Follows your onboarding checklist, tracks what is done, and drafts reminders for what is outstanding.
What a person still approves
Reminders and any message that goes to a client.
What we would score
Steps completed and nothing outstanding without a reminder.
Illustrative phone screen for this use case, with sample data

Monthly report skill

Drafts your monthly summary from your sheets, with checks.

See details
The problem
Monthly reports take hours and numbers get copied wrongly.
What the skill does
Pulls figures from your sheets, drafts the sections, checks totals against the source and flags unusual changes.
What a person still approves
The report before it is shared.
What we would score
Figures matching the source and unusual changes caught.
Not on the list?

We start from your workflow

If your team repeats a task that follows clear steps and touches messages, documents or spreadsheets, there may be an agent that can help. We will say plainly if there is not.

Describe your workflow