- Product guarantee
- 2 human gates
- Product guarantee
- ₹5,00,000 budget guard
- Product guarantee
- 30-minute approval token
- Product guarantee
- 48-hour delivery follow-up
How it works
Two human gates bracket the AI work. Nothing external is sent until the second gate is passed.
Five specialist agents, one pipeline
Between the two gates, five Haiku 4.5 agents run in order — each one responsibility, one validated output, evidence attached.
Evidence, not vibes
Every number carries an evidence id
Prices, ratings and delivery dates are never asserted by a model. Each one is computed in code from a cited source and rendered with an evidence chip, such as ₹19,499. No evidence, no number — the dashboard shows a dash with a tooltip instead of a guess.
Security
The system is built so a spend decision can always be explained and never taken by accident.
- Approval token
- Single-use, 30-minute token minted only when an admin selects an option — it gates every purchase order, email and follow-up.
- Budget guard
- A purchase order over ₹5,00,000 is blocked automatically and routed to finance review. It is never auto-issued.
- Row-level security
- Every table is scoped by tenant with Postgres RLS — the API and the dashboard both sit behind it.
- Fixtures by default
- Live marketplace, pricing and vendor adapters are opt-in per platform; the default build runs entirely on fixtures.
See a procurement decision your finance team can trust.
Sign in to the demo tenant, or create an account to try the full request-to-PO flow.