Three agents that work like a real finance team — a Financial Advisor who talks to the owner, a Financial Analyst who researches the outside world, and an Accountant who reads the books in TallyPrime. They coordinate to do the work and to answer the hard questions. Built on Hermes.
This is the effective rubric. Every design choice below serves one of these — nothing else.
A complete job, from the owner's request all the way to a real outcome — not a menu of features.
Multiple agents visibly handing off and combining their work toward one outcome.
The partner tools we identified doing real work inside the flow, not bolted on.
The Advisor is the face and the orchestrator. It never answers alone — it spawns the Analyst (the outside world) and the Accountant (the books), then synthesizes.
Talks to the owner, frames the question, delegates to the Analyst and Accountant, fuses their inputs into one recommendation, and delivers it.
Researches the outside world — demand, prices, interest rates, regulatory conditions, competitors — and reports what it means for the decision.
Speaks with TallyPrime. Pulls the current reality — sales history, cash, receivables, payables — and posts new entries. Keeps a check on the situation.
The #1 ask across a year of waitlist (40% of signups). Proves the plumbing — and it's a coordinated chain, not one agent.
Receives the invoice photo/PDF on WhatsApp or Telegram, opens a job, delegates.
Reads the image → structured fields: vendor, GSTIN, date, line items, taxable value, CGST/SGST/IGST, total.
Checks the tax math and whether the vendor exists as a ledger. Unknown vendor or missing GSTIN → verifies it live via Linkup. Flags anything unresolved.
Maps to the correct ledgers, builds the voucher, posts to Tally on confirm, returns the voucher number.
"Booked INV-2231 from Bajaj Steel — ₹1,50,000 (₹1,27,119 + ₹22,881 GST) as Purchase voucher V-104. Vendor GSTIN verified via Linkup."
The Advisor can't answer this alone. It needs the books (Accountant) and the market (Analyst) at once — which is exactly what makes the coordination real.
Frames the decision, states the plan, spawns two requests in parallel.
Pulls 6 months of prawn sales by destination + customer: "₹4.2 Cr total — 62% to the US, and US is down 30% over the last two months. Top US buyer Gulf Coast Seafood owes ₹38 L, 55 days overdue. China 16%, Vietnam 12%."
Researches the outside world: "China vannamei demand is strong; Vietnam is a re-processing hub for Indian raw material; the US carries anti-dumping duties on Indian shrimp and freight/insurance is spiking. Selling to China needs GACC + MPEDA registration (~weeks)."
Fuses internal + external into one recommendation, spoken back to the owner.
"Your books already confirm the US is softening and ₹38 L is stuck there. The market backs the pivot: prioritize China — but start the GACC/MPEDA registration now, it's the long pole. Keep Vietnam as a re-processing channel, hold a small US allocation for paying buyers only, and collect that ₹38 L before extending more US credit. Loop your CHA in on the China paperwork."
Indian shrimp leaning on the US — and the live debate about diversifying to China and Southeast Asia — is a real industry story. Point Linkup at the real destination-market and regulatory data; let the "US–Iran war" be the narrative shock that starts the question. The Analyst returns facts a judge can verify.
Dodo Payments doesn't fit an advisory flow cleanly — include it only as an "activate your plan" checkout, or leave it out. A natural tool story beats a forced one.
Bound the Analyst's research to 2–3 concrete Linkup queries. Rehearse one happy path per flow. Make every handoff visible — each agent posts its own line so the judge sees the coordination. Flow 2 (advisory) is the demo showpiece; note it reaches beyond the bookkeeping the waitlist asked for.