WhatsApp-native · 18 regulators · Cited answers

Ask your compliance copilot anything —
get cited answers on WhatsApp in seconds.

GST rules, RBI circulars, SEBI deadlines, MCA forms, FSSAI categories — every answer comes back with the source regulation cited. English or Hindi. Audit-logged. No app to install. No hallucinated answers.

Trajectory-traced with Langfuse · Prompt-injection defence · DPDP-compliant conversation logging.

S
Sahayak Compliance Co-pilot
online
For an export of HSN 8542 to UAE, what's the IGST refund mechanism?
11:42
Under Rule 96 of the CGST Rules, you have two options:
1. Refund of IGST paid — claim via shipping bill (auto-route).
2. LUT export without IGST — pre-file Form RFD-11.

For HSN 8542 (electronic ICs), no special restriction applies.
Sources:
· CGST Rules 2017, Rule 96
· Notification 16/2017-CT, dated 7-Jul-2017
11:42 ✓✓

The problem

Your junior staff are Googling compliance answers. Your CA is answering the same question for the third time this month.

!

Google answers are stale

First-page results cite circulars that were superseded 18 months ago. Junior staff copy them anyway because the layout looks authoritative. Then the demand notice arrives.

AI

Generic AI hallucinates rules

ChatGPT will cheerfully invent a Section 197A that doesn't exist, or quote a 2019 RBI master direction as if it's current. No citation, no confidence, no auditability.

CA

Your CA is the bottleneck

"What's the TDS rate on a non-resident payment for technical services under DTAA with Singapore?" — your senior partner answers this twice a quarter from memory. The fourth time, it's a Saturday.

The solution

A copilot that lives in the tool your team already uses 11 hours a day.

1

WhatsApp Business API native

Add the Sahayak number to your team's WhatsApp. No app install, no SSO friction. Works on the cheapest Android phone your articled assistant carries.

2

18 regulators indexed

CBIC, RBI, SEBI, MCA, FSSAI, DGFT, CBDT, EPFO, ESIC, IRDAI, PFRDA, BIS, IBBI, CERT-In, NPCI, CCI, TRAI, IGR. Re-indexed weekly. Every answer cites the regulation it draws from.

3

Auditable + observable

Every conversation is logged with Langfuse trajectory traces. Admins can review trees of reasoning, flag bad answers, and re-train. Prompt-injection defence is on by default.

Used by CA firms and CFO offices across India

Customer logo 1
Customer logo 2
Customer logo 3

Logos pending customer permission to publish.

Pricing

Per-seat, predictable, no per-question gouging.

Starter

₹2,999/month

For a solo CA / single compliance lead.

  • 1 WhatsApp seat
  • 500 queries / month
  • 18 regulators, English + Hindi
Start with Starter
Most popular
Growth

₹9,999/month

For a CA firm or finance team with 5 active queriers.

  • 5 WhatsApp seats
  • Unlimited queries
  • Conversation audit log + Langfuse
  • Admin review of flagged answers
Book a Demo
Enterprise

Custom

White-label, SSO, dedicated tenant.

  • White-label number + branding
  • SSO (SAML / OIDC)
  • Dedicated tenant, isolated data
  • Custom regulation packs
Talk to Sales

See full pricing & feature matrix →

FAQ

Sharp questions. Honest answers.

How accurate is the copilot? Is it really better than ChatGPT?+

It's a different tool. ChatGPT is a generalist; we're a retrieval-augmented system over an India-specific regulation corpus that we re-index weekly. Every answer is grounded in a cited paragraph. If we can't find a grounded paragraph, we say "I don't know — escalate to your CA" rather than guessing. We don't claim 100% accuracy; we claim auditable accuracy.

Which language does it understand?+

English + Hindi. Hinglish ("kya GST rate hai HSN 8542 pe?") works fine — we don't require a clean register. Replies default to the language of the question. Tamil, Telugu, Marathi, Gujarati, Bengali are on the Enterprise roadmap.

What happens if someone asks something off-topic?+

The copilot will politely decline — it is scope-bounded to Indian regulatory and compliance topics. Off-topic prompts are logged for the admin (rate-limited, no full content captured beyond classification) so you can spot anomalies.

How does prompt-injection defence work?+

Two layers. Layer 1: input classifier flags prompts that attempt role-confusion, data-exfiltration, or out-of-scope behaviour, and routes them to a stricter prompt. Layer 2: output validator refuses any response that doesn't carry at least one valid citation. Caught attempts are logged and you can review them in the Langfuse trace UI.

Where is conversation data stored?+

AWS Mumbai. TLS 1.3 in transit, AES-256 at rest. Conversation transcripts retained 30 days on Starter / 90 days on Growth / configurable on Enterprise. DPA available on request — read the Trust statement.

Can I see exactly what the copilot used to answer a question?+

Yes. Growth + Enterprise admins get a Langfuse-powered trajectory view: the retrieved chunks, the system prompt, the model trace, the tool calls. You can flag a trace as "bad" and it goes into our weekly evaluation set.

Bring three real compliance questions to the demo.

We'll answer them live on WhatsApp, with citations. If the answer is wrong, we tell you why.