razorsharpplatform
Meet Razor AI
HUMAN FIRST. INTELLIGENCE ALONGSIDE.RAZORSHARP / 001

Yesthisisme,Vedant

x100x100x100x100
ALWAYS THINKING HOW TO HARNESS AI
x100x100x100x100
SEES EVERY WORKFLOW AS AN AGENT LOOP
x100x100x100x100
SPEAKS IN PROMPTS AND GITHUB LINKS

Your taste. Your ambition. Your way.

AND THIS IS WHAT’S NEXT
RAZORSHARP PLATFORM / POWERED BY RAZOR AI

Inside RazorSharp.
What is actually built.

A voice-first shopping experience. A merchant action workspace. A transaction architecture built to make consequential decisions inspectable.

IMPLEMENTATION WALKTHROUGH
AREA 01 / 20
FOR YOUR CUSTOMERS

A shopping conversation. With somewhere to go.

Say the occasion, budget and what matters. The Shopping Specialist helps turn intent into product choices; the voice gateway brings speech into the journey.

WHY IT MATTERSLess searching. A selection shaped by the customer’s intent.
Meet your shopping copilotShopping model-backed when configured · voice availability depends on connection
Implementation referencepackages/commerce-api/src/commerce_api/services/agent_bridge.py

Source reference, not a test result. Inspect in the local repository; frontend files may still be uncommitted.

ConversationSCHEMATIC
Earphones + snacks. Under ₹800.
Razor AI · finding a selection
EarphonesStore price
Movie snacksPaid share
Ask about this selection
CONVERSATION TO ACTION
“Earphones and snacks. Keep it under ₹800.”
  1. 1Understand intent
  2. 2Find available products
  3. Review the exact purchase
Code-based explanation · not a screenshot or live result
Motion paused
01 / 20
Voice shopping
RAZORPAY BUILDATHON / TRACK 1 THEME

AI growth meets
agentic commerce.

RazorSharp connects the merchant’s next action to the customer’s next purchase—with explicit controls at the point where money moves.

01

Help merchants grow

Merchant Command brings catalogue context, business questions and campaign proposals into one reviewable workspace.

Connected catalogue · analytics and campaigns include previews
02

Turn intent into commerce

Razor AI combines voice interaction and a Shopping Specialist with product discovery and exact-checkout review.

Shopping model bridge and voice gateway implemented; configuration required
03

Make agentic spending accountable

Reserve Pay explores selected-product, bounded, revocable authority with Kernel admission and durable execution.

UAP-style server simulation · no live NPCI or bank connection
04

Connect through commerce protocols

ACP/UCP interfaces and MCP/AP2 capabilities preserve authority boundaries across different entry points.

Protocol implementations · not marketplace availability or certification
05

Build beyond the happy path

Content-bound approval, idempotency, single-use grants and reconciliation address changing terms, retries and uncertain outcomes.

Backend controls · validate runtime behaviour with the test suites
06

Make the engineering inspectable

The evidence model connects intent, approval, execution and provider outcome. Missing evidence remains explicit.

Verifier implemented · homepage visuals are illustrations

Alignment with the AI Growth & Agentic Commerce theme, not a claim of endorsement or full scoring-rubric compliance.

BUILT TO EARN A CLOSER LOOK

Product depth.
Engineering substance.

Razor AI connects voice interaction and specialist responsibilities with two workspaces. RazorSharp surrounds that experience with exact-order approval, revocable authority, durable execution and inspectable evidence.
Connected paths, backend controls and frontend previews are different things. The source references show what each claim is based on.

See how it fits together
COMMERCE, IN CONVERSATION

Good things happen
when intelligence acts.

Your next great find. Your business’s next big move.
Powered by Razor AI. Built for a more thoughtful way to commerce.

Quick-commerce assortment: electronics, personal care, home essentials, stationery, toys, snacks and groceries
FOR YOUR EVERYDAY

A little less searching.
A lot more finding.

Everyday tech, beauty, home and groceries. One Razor AI shopping copilot.

Meet your shopping copilot
“Earphones, skincare, and snacks for tonight.”
FOR YOUR BUSINESS

Your ambition.
An unfair advantage.

Turn business questions into your next move.

Open Merchant Command
Illustrative business data
Intelligence proposes. You stay in control. Speak naturally. Shop thoughtfully.Powered by possibility
THE TRUST ARCHITECTURE

Every request has a route.
Every authority has a boundary.

Razor AI helps decide what to ask for. The Transaction Trust Kernel decides whether a money action is permitted. A protocol never bypasses that decision.

01

Razor AI proposes

A purchase request is intent. Model output cannot grant itself financial authority.

02

Authority is checked

Trusted buyer approval or valid delegated authority must match the exact purchase.

03

Kernel admits or refuses

Locked state, bound content and financial invariants determine the admission decision.

04

Executor carries it out

A bound Execution Grant and durable command lead to provider execution and evidence.

FINANCIAL AUTHORITY

Transaction Trust Kernel

Enforces sale-bound merchant terms and non-overridable platform financial invariants.

  • Approval binds to checkout version and canonical content hash.
  • Admission revalidates merchant state, reservations and buyer or delegated authority.
  • Payment, refund and Reserve Pay controls preserve refundable balance, grant use and duplicate-execution protection.
  • Financial decisions and their audit evidence commit together.
Merchant approval cannot weaken these protections.
MERCHANT OPERATIONS

Merchant Action Controller Command

Governs who may propose and approve merchant changes, then invokes the business module that owns the operation.

  • Merchant identity, permissions, typed proposals and action-content hash.
  • Human approval and the merchant-action lifecycle.
  • Catalogue, inventory, pricing and Merchant Policy remain business-domain concerns.
  • Financial remedies require a separately authorised financial API and Kernel path.
Cannot rewrite buyer approvals, checkout versions, Policy-at-Sale Receipts or Execution Grants.
INTEROPERABILITY ≠ AUTHORITY

Protocols carry requests and evidence.
They do not mint permission to spend.

These are the repository’s declared pins and claim boundaries, not a live conformance report or an external certification.

Protocol boundaries declared in commerce_protocols/core/pins.py
Protocol / pinRole / declared boundaryWhere the claim stops
ACP2026-04-17Checkout interoperabilityCompatible interfaceDoes not mean live distribution in ChatGPT Instant Checkout.
UCP2026-08-25Commerce profile and lifecycleLocal conformance pinSeparate onboarding is required for Gemini or Google AI Mode.
MCP2025-06-18Scoped tool accessCompatible interfaceTool access is not buyer consent. Claude is outside the tested runtime.
AP2v0.2.0Cryptographic evidence verificationLocal conformance pinThis project narrows the cryptographic profile to ES256. Evidence still passes Kernel checks.
RESERVE PAY / UAP

Governed spending. Simulated rail.

Selected-product scope, capacity, authority epoch and revocation constrain Reserve Pay. The provider path is a labelled simulation, not a live NPCI or bank connection. UAP is not an additional pin in the four-protocol matrix above.

EVIDENCE, NOT A GREEN TICK

The proof is recomputed.

The Money Action Proof Chain reads committed rows, recomputes the checkout hash and verifies audit chains. Missing evidence stays missing. A homepage diagram is not a successful payment or a verifier result.

Inspect the implementation grounding 6 source references

This explanation is grounded in repository code. Runtime results depend on the deployed configuration and the individual transaction.

Atomic admission and reapproval
packages/transaction-kernel/src/transaction_kernel/admission.py
Merchant ownership and import boundary
packages/merchant-controller/src/merchant_controller/__init__.py
Protocol pins and disclaimers
packages/commerce-protocols/src/commerce_protocols/core/pins.py
Published protocol status
packages/commerce-api/src/commerce_api/routers/protocols.py
Reserve Pay API and simulation boundary
packages/commerce-api/src/commerce_api/routers/reserve.py
Committed evidence and hash recomputation
packages/commerce-api/src/commerce_api/services/proof_chain.py
THE CHALLENGES WE TAKE SERIOUSLY

Commerce is easy.
Until something changes.

A customer changes their mind. A merchant changes an offer. A network drops after payment. RazorSharp is built around the decisions hiding between the happy-path screens.

01

Too many filters. Too little context.

Start with an occasion, a budget or a spoken request. Let the shopping copilot help turn it into a selection.

DISCOVERY → INTENT
02

An insight is not an action.

Give merchants a place to investigate stock, prepare changes and review the next move—not just read another chart.

INSIGHT → PROPOSAL
03

The price moved. The yes didn’t.

Bind approval to the exact checkout. Relevant changes require a new review, even when the new total is lower.

CHANGED TERMS → FRESH APPROVAL
04

A retry must not be a second purchase.

Use idempotency and single-winner admission to protect the financial path from repeated requests.

RETRY → SAME OPERATION
05

No response is not a failed payment.

Keep uncertain outcomes visible. Reconcile provider evidence instead of encouraging another payment.

UNCERTAINTY → RECOVERY
06

“Done” needs something behind it.

Connect the approved content, financial decision, execution and provider outcome through inspectable evidence.

CLAIM → CHECKABLE RECORD
FOR THE BUSINESS BEHIND THE STORE

Less dashboard watching.
More considered action.

Merchant Command brings investigation, proposals and approvals into one workspace. The goal: help merchants move from “what happened?” to “what should I do next?”

Catalogue connected · analytics preview

Find the signal behind the store.

Bring catalogue health, inventory and business context into the same workspace. Separate measured records from illustrative analysis.

  • Catalogue and stock context
  • Orders and outcome context
  • Evidence before recommendations
NEXT OUTCOMEA question worth acting on
Interactive product-flow illustration · not live analytics
THE TRANSACTION TRUST KERNEL

Intelligence can propose.
Authority must be proven.

Merchant terms govern the sale. Independent platform invariants govern the money. Neither a model response nor a merchant approval can bypass the financial checks.

ILLUSTRATIVE SCENARIO · NO TRANSACTION EXECUTED

Your yes belongs to these exact details.

An approval names the checkout version, content hash, amount and currency. The Kernel checks the bound purchase before admitting financial work.

CONTENT + AUTHORITY + CURRENT STATE
BEFORE₹420
AFTER₹420
Ready for admission
  1. 1Review exact content
  2. 2Record buyer authority
  3. 3Revalidate and admit
Approval bindingExact content, amount and currency
Single-use grantsExecution authority cannot be replayed
Refund limitsBounded by refundable paid amounts
Safe recoveryUnknown outcomes remain unresolved
TWO COPILOTS. CLEAR RESPONSIBILITIES.

Meet Razor AI.
Two copilots. Defined roles.

Specialists help with different jobs. Trusted services own consequential actions. The boundary is part of the product, not a sentence hidden in a prompt.

VOICE + TEXT

Say what you need.
Keep the context.

Speech recognition, conversation and spoken responses share a gateway path. Connection and provider configuration determine availability.

Voice is an interface—not financial authority.
01

Shopping Specialist

Understands shopping intent and proposes grounded product selections.

Model-backed when configured
02

Checkout Specialist

Coordinates deterministic checkout information and guarded purchase steps.

Deterministic runner
03

Support Specialist

Uses order and sale-term context to support issue handling.

Deterministic runner
Buyer approval → Transaction Trust Kernel → Action Executor
ENGINEERING YOU CAN INSPECT

Follow the purchase.
All the way to the evidence.

Ten links connect intent to the final financial state. The verifier recomputes content hashes and checks audit chains. Missing evidence stays missing.

EVIDENCE MODEL · NOT A LIVE VERDICT
LINK 03 / 10

Content hash

Canonical checkout content is hashed again, not trusted because a stored label says verified.

Explaining a check is different from verifying a purchase.
Inspect the verifier
BUILT FOR AGENTIC COMMERCE

Open interfaces.
Explicit boundaries.

Protocol support and controlled execution belong together. An adapter is not a marketplace listing, and a simulated debit is not a live bank payment.

ACPCommerce transport

ACP-compatible checkout interfaces. Not a live ChatGPT Instant Checkout integration.

UCPProfiles & signed interactions

Version-pinned UCP support with explicit claim boundaries. Not availability inside Gemini.

MCP / AP2Tools & authorization evidence

Capability-scoped tools and protocol evidence. Protocol access does not create buyer consent.

Reserve / UAPBounded delegated spending

Selected-product authority, limits, epoch and revocation. Provider execution is simulated; NPCI/banks are not connected.

AI GROWTH & AGENTIC COMMERCE

Merchant insight-to-action, conversational discovery, checkout controls and inspectable payment outcomes—the project’s contribution to the Track 1 theme. This is a build in progress, not a claim of certification or complete rubric coverage.

EXPLORE THE PRODUCT. INSPECT THE SYSTEM.

Thoughtful on the surface.
Accountable underneath.

Two ways into commerce. One commitment to keeping consequential actions within their authority.

Try the shopping copilot Open Merchant Command
Explore the public source