KEYXE is designed as a software workspace for two distinct categories of commerce information: seller operations and advertising performance. Our approach prioritizes accurate definitions, transparent data sources and tools that people can inspect before acting.
Illustration of the intended data flow. Live ingestion from Amazon is not active.
The loop
Connect → Organize → Explore → Act (reviewed)
A simple sequence, with a permission check at every arrow.
CONNECT
By permission
An account owner authorizes a specific Amazon service. Today, only the synthetic demo is connected.
ORGANIZE
With definitions
Raw reports become a normalized model with documented metrics, currencies and freshness.
EXPLORE
Interactively
People filter, compare and export — and see how each number is calculated.
ACT (REVIEWED)
With approval
Suggestions are reviewed by a person. Account changes are a future, permission-gated capability.
Principles
Five design commitments
Not yet approved by Amazon
Data Access by Permission
Authorization belongs to the account holder. A seller decides whether KEYXE may read seller data; an advertiser decides whether KEYXE may read advertising data. Account associations and access scopes are verified on the server, never trusted from the browser.
By design, each Amazon connection will be tied to one KEYXE organization and checked on every request.
Seller and advertising authorizations will be stored, refreshed and revoked independently.
KEYXE never asks for an Amazon password; authorization happens on Amazon’s own consent screens.
Interactive demo · synthetic data
Source-Aware Reporting
Every figure carries its context: data source, account and marketplace, currency, reporting period and freshness. Missing data is treated as unknown — never quietly turned into zero.
Ratios such as ACoS or unit session percentage are computed from period totals, with N/A for zero denominators.
Seller ordered sales and ad-attributed sales are labeled differently and never added together.
Each marketplace keeps its own currency; KEYXE does not invent exchange-rate conversions.
Interactive demo · synthetic data
Interactive Analysis
Dashboards, sortable tables, grouped summaries and trends let people explore questions at their own pace — then export what they see.
Filters for marketplace, period, product, campaign, ad format and status.
CSV and JSON export of exactly the rows on screen, labeled as synthetic in the demo.
Shareable links that reproduce a view, and plain-language metric explanations.
Planned
Reviewable Workflows
Advice and action are different things. KEYXE drafts suggestions with their evidence and assumptions, and keeps any future account-changing step behind explicit permission and human approval.
Today: bid-change proposals are a simulation in the demo and never reach Amazon.
Planned: dry-run previews, approval records and idempotent execution for permitted changes.
Never: unattended automatic changes to bids, budgets or listings.
Planned
MCP Connectivity
Discoverable, read-only tool schemas with structured example outputs. Live data access through MCP will be gated by authorization and data-use policy.
Browser-local MCP Tool Explorer available now with synthetic results.
Hosted, authenticated MCP gateway planned — no public MCP endpoint exists today.
Sending real Amazon data to external AI clients is disabled pending policy review.
Data pipeline
Five stages, each with a job to do
The intended architecture for live integrations. Expand a stage for details.
Seller reports and data from the Selling Partner API; campaign reports from the Amazon Ads API. Each source is enabled only after Amazon approves the application and an account owner authorizes it. Today, KEYXE’s synthetic fixture generator is the single source.
Consent is collected on Amazon’s screens with anti-forgery state checks. Refresh credentials stay encrypted on the server, scoped to one organization, and can be revoked independently for seller and advertising access.
Incoming rows are checked against expected schemas, currencies and marketplaces. Late or partial reports are flagged instead of silently filled. Duplicates are removed by stable keys.
Metrics are computed from totals over the selected period with explicit formulas. Seller and advertising results are presented side by side only where policy allows, and never added together.
MCP-style tools expose approved, read-only queries with structured outputs that carry source, freshness and warnings. Transfer of real Amazon data to external AI providers is off unless a separate review permits it.
This describes design intent, not live processing of Amazon data. See Data Foundation for definitions.
See the approach in action.
Every principle on this page is visible in the interactive demo, using fictional sample data.
