RobnuAI sees your data — and only your data.
RobnuAI is the seller-scoped AI surface inside Robnu. Built on the same multi-tenant foundation as the rest of the platform: every read, every write, every tool call locked to your seller_id by both the application layer and Postgres RLS. Free during early access — no card.
- RobnuAI A1 ships tenant-safe chat + document extraction today, with persistence in ai_conversations / ai_messages / ai_tool_calls / ai_predictions / ai_document_extractions.
- Listing optimization, ads automation, reviews intelligence, forecasting, and agents are scheduled — same gateway, same safety contract.
- PII redaction sits in the gateway. The model never sees raw names, phones, addresses, or GSTINs.
What RobnuAI ships in A1, and what comes next.
Eight surfaces sharing one gateway, one persistence schema, one safety contract. Live today on the bottom row; scheduled on the top.
Copilot
Ask any question grounded in your seller data. Tenant-scoped from token to tool call.
Document extraction
Vendor invoices, settlement files, packing slips → typed fields with confidence scores.
Listing optimization
Title, description, and image checks tuned to Ajio + Meesho ranking signals.
Ads automation
Bid + budget moves on marketplace ad consoles, with explainable proposals.
Reviews intelligence
Cluster review themes and surface the SKU-level fixes that move stars.
Forecasting
Demand + return-rate forecasts grounded in your real settlement history.
Agents
Long-running agents that close manifests, file claims, and chase missing payouts on schedule.
Safety + tenancy
withSellerScope on every read and write. PII redaction at the gateway. Full audit trail.
One gateway. One schema. One audit trail.
Every AI surface — copilot today, ads-automation tomorrow — goes through the same services/ai/gateway.ts abstraction. Persistence is uniform: ai_conversations, ai_messages, ai_tool_calls, ai_predictions, ai_document_extractions, ai_usage_ledger. Every row carries seller_id; every read sets the RLS GUC.
- · No bypass paths — adapters cannot call providers directly.
- · PII redaction runs before the provider; re-hydration runs after.
- · Tool calls log both arguments and results into ai_tool_calls.
- · Usage is metered per seller into ai_usage_ledger for fair limits.
Honest answers about how RobnuAI works.
No. The A1 baseline ships seller-scoped persistence (ai_conversations, ai_messages, ai_tool_calls, ai_usage_ledger, ai_predictions, ai_document_extractions), a single gateway abstraction, and tenant-safe tool calls into your own data. Chat is the first surface; ads, listings, reviews, and forecasting all reuse the same plumbing.
Every read and write runs through withSellerScope, which enforces seller_id at both the application layer and Postgres RLS. PII fields are encrypted at rest with a per-seller DEK wrapped under your AWS KMS alias. PII redaction sits in the gateway path — names, phones, addresses, GSTINs are redacted before any external provider call and re-hydrated only on the way back.
RobnuAI uses a single gateway abstraction so the underlying provider can change without touching product code. The current default is a hosted frontier model with a deterministic local fallback for development and offline tests.
Yes. Sellers under 25 orders/day get the full RobnuAI surface free, forever. Heavy users will see usage caps when paid pricing launches, but never on the surfaces that move money — copilot, document extraction, claim drafts.
Only inside guarded actions you approve in advance. Mutations go through the same mutation guard the rest of Robnu uses (ROBNU_AJIO_MUTATIONS_ENABLED, dry-run by default). Until you flip that flag, RobnuAI proposes but never pushes.
Talk to RobnuAI inside your own dashboard.
Sign up free, connect Ajio, ask the copilot why a payout is short. No card.