Ask your ERP in plain English, Hindi, or Hinglish — and get sourced answers back
The copilot resolves questions against your actual ledgers, party masters, and Custom DocTypes — then cites every voucher it used. Team members skip the CSV export, and finance retains the audit trail.
Natural-language ERP query lets Indian businesses type or speak a question like "show me overdue invoices in Q3 above ₹1 lakh, grouped by state" and receive a grounded answer with per-voucher citations from TallyPrime, Zoho, ERPNext, or a custom ERP. Questions can be posed in English, Hindi, or Hinglish. Every answer names the specific vouchers, DocTypes, or exports it drew from, so the finance team can drill into the source of every number without opening the ERP client. Queries the copilot cannot ground in the customer's own data are refused rather than answered — no invented numbers.
Last updated 2026-07-02 · Part of the AI Copilot for ERPs early access
What this capability does
Every ERP question that starts with "how many" or "show me" belongs in a natural-language interface. Right now those questions live in Excel pivots, month-end reports, or a WhatsApp message to the accountant asking for "the number by EOD." Natural-language ERP query collapses that flow. A user types the question, the copilot introspects the target ERP's schema, resolves the query against the live ledgers or DocTypes, and returns a grounded answer with links back to the source vouchers. On TallyPrime this runs through the local Perimattic bridge with read-only ODBC. On Zoho it uses the Marketplace OAuth token scoped to the modules you nominated. On ERPNext it walks the DocType schema (including Custom DocTypes) and pulls through the standard REST API. On custom ERPs the copilot indexes whichever fields you mapped during onboarding — REST, CSV, or middleware. Question input is auto-detected across English, Hindi, and Hinglish, and voice input over WhatsApp is transcribed the same way. Because every answer names the source vouchers, statutory auditors and internal finance leads can verify claims without leaving the tool.
How it works
End to end, from question to grounded action.
Question intake
A user types or speaks the question through the ERPPlugs dashboard, an embedded chat surface in the ERP, or WhatsApp. Language auto-detects between English, Hindi, and Hinglish. The copilot rewrites the question into a structured query plan before touching the ERP.
Schema-aware resolution
The copilot resolves the plan against the ERP's live schema — Tally companies and ledger groups, Zoho modules and organisation scope, ERPNext DocTypes and Custom Fields, or your mapped custom-ERP fields. Access is bounded by the user's role and the ERP's own permission model.
Grounded answer generation
The retrieved rows are handed to the language layer, which composes a plain-language answer that names the source vouchers, invoices, or DocTypes. Questions the copilot cannot resolve from your data are refused with an explanation, not answered from generic training data.
Drill-down and audit log
Every answer is clickable — the user drills into the exact vouchers used. Every query, resolved plan, and returned voucher list is written to your audit storage with user identity and timestamp for statutory review.
How this shows up on each ERP
Same capability, tuned to each ERP's data model.
Read-only ODBC through the local Perimattic bridge. Multi-company Tally hierarchy respected — queries scope to the active company by default. Financial-year filters and party-master aliases are honoured automatically.
Marketplace OAuth token, region-aware routing across IN / US / EU / AU data centres. Cross-module queries — a question spanning a Sales Invoice, an Inventory PO, and a linked CRM Deal — resolve in one pass with per-module citations.
DocType-aware from the introspection pass onward. Custom DocTypes and Custom Fields are indexable from day one. Queries respect the Company boundary and any User Permission constraints already set in your ERPNext instance.
Queries resolve against the fields you mapped during onboarding — REST + webhook, CSV / Google Sheets, or middleware ingestion. Field-mapping console lets you rename or add fields without breaking the copilot.
What this looks like in practice
Real question-and-answer patterns from pilot deployments.
“Show me overdue invoices above ₹5 lakh, aged 60+ days, grouped by state.”
A grouped table by state code with 14 invoices totalling ₹1.34 Cr overdue. Each row links to the source Sales Invoice voucher and the buyer's party master.
“इस महीने कौन से वेंडर को दो बार पेमेंट गया है?”
A duplicate-payment report with fuzzy match on PAN and GSTIN, flagging four vendors with paired vouchers listed. Answer text mirrors the input language.
“Compare Shopify order revenue booked in Tally to Shopify's own report for July.”
A side-by-side reconciliation showing ₹42.6L in Tally against ₹43.1L on Shopify, with the four missing orders identified by Shopify order ID and the reason each did not book (three payment reversals, one manual entry pending).
“What was our average collection period per debtor segment last quarter?”
A segment breakdown — Enterprise 47 days, Mid-Market 34 days, SMB 22 days — with the underlying voucher counts and the definition the copilot used for "collection period."
Why this matters
No CSV pivots
Answers arrive with citations. No manual Excel work.
Multilingual by default
English, Hindi, and Hinglish — voice or text.
Grounded, not generic
Every answer names the source vouchers used.
Refuses invented answers
The copilot rejects questions it cannot ground.
Frequently asked questions
How does the copilot know it can answer a question from my ERP data?+
During introspection the copilot maps the ERP's schema — Tally ledger groups, Zoho modules, ERPNext DocTypes, or your custom-ERP field mapping. When a question arrives, the copilot plans the query against that schema. If required fields are missing from your scope, the copilot explains what is missing rather than fabricating an answer. This is unusual for generic AI chatbots and is the discipline that keeps the copilot usable in a finance workflow.
What does "grounded" mean in practice?+
Every answer names the vouchers, invoices, or DocTypes it drew from. A user can click through to open each source in the underlying ERP. During pilot onboarding your finance lead can require citations to be visible on every answer before enabling autonomous actions. Grounding is what makes the copilot defensible to a statutory auditor asking how a number was derived.
Do I need to teach the copilot my ledger structure?+
No. Introspection reads your ledger group tree, party masters, and voucher headers automatically. On ERPNext, Custom DocTypes and Custom Fields are picked up on the initial connect and on every subsequent redeploy. On custom ERPs, the field mapping is captured during the onboarding workshop and version-locked so schema changes surface as alerts rather than silent breakage.
How is the multilingual support different from Google Translate?+
The copilot understands accounting and tax terminology in Indian languages — CGST/SGST/IGST, HSN, ITC, party ledger, ageing bracket — not just literal translation. A Hindi question about ITC on capital goods resolves against the correct ledger set. Translation-first tools miss these mappings and return generic answers.
Can I embed the natural-language query into my existing ERP UI?+
Yes for TallyPrime (through the Perimattic bridge dashboard) and Zoho (through the Marketplace app that surfaces a chat panel inside the Zoho UI). ERPNext customers embed the copilot as a Frappe Portal page. Custom ERPs typically expose it either as a Slack app, a WhatsApp bot, or an iframe in your internal admin — whichever surface your team already uses.
What happens with sensitive queries — payroll, director salaries?+
Sensitive DocTypes or ledger groups can be excluded from the copilot's scope during onboarding — payroll, director loans, related-party transactions. If a user asks a question against an excluded scope, the copilot explains that the scope is out of bounds instead of resolving the query. This mirrors ERPNext User Permissions or Tally's security-level model.
Related AI Copilot capabilities
Every capability composes with the others.
Available today on these ERPs
Each edition uses the same connectivity ERPPlugs already ships.
Want this capability running on your ERP?
Join the AI Copilot early-access pilot. We calibrate against your own data during the first 2–4 weeks and lock in early-adopter terms before general availability.
Join Early Access