We build practical AI employees that understand orders, customers, policies, and approvals. The first one is Inbox, an AI support employee for Shopify brands.
Operating layer
Commerce AI employee network
Messages
Orders
Policies
Reasoning
Approval
Audit
0
unsupported sends
6
validation gates
1
flagship product
Commerce teams need AI that understands orders, customers, policies, and approval rules.
Useful AI employees should show their work before they act.
Autonomy should increase only after the system earns trust through evidence and audit history.
Flagship product
It is not a generic chatbot. Inbox reads the customer thread, Shopify order context, approved knowledge, case history, and safety checks before it prepares the next reply.
Operating model
The product can feel simple because the system underneath is strict: gather context, prepare the action, validate, ask for approval when needed, then leave a record.
Connect to the systems where commerce work happens: inboxes, stores, knowledge, shipping, returns, and billing.
Resolve the customer, order, policy, prior cases, brand voice, and missing information before drafting or recommending.
Draft the reply, classify the case, recommend a next step, or gather evidence for a human decision.
Run grounding, citation, recipient, policy, PII, and prompt-injection checks before anything reaches a customer.
Keep sensitive actions like refunds, replacements, discounts, and policy exceptions under human control.
Record the context, draft, actor, approval, send, and repair path so trust can be reviewed later.
Product system
Inbox is the first product because support exposes the whole commerce truth: customers, orders, policies, fulfillment, refunds, and brand voice.
Inbox
liveInbox investigates customer conversations with Shopify context, approved knowledge, preflight checks, and approval routing before a reply reaches a customer.
Voice AI
plannedA future commerce voice employee for order status, delivery questions, returns, and handoff-ready cases.
Brand Brain
designA shared knowledge layer for future AI employees so answers stay consistent across channels.
Ops Agent
plannedA future employee for fulfillment exceptions, returns operations, customer follow-ups, and internal task routing.
Proof without fake trust
The homepage is ready for customer logos, testimonials, recognitions, and case studies. Nothing named is shown until the proof and permission are real.
Customer logo slot
placeholder
Render only after written customer approval, or show as an explicit placeholder in draft-only sections.
Customer logo slot
placeholder
Do not imply the brand is a customer until approved proof exists.
Customer logo slot
placeholder
Use approved logo assets, alt text, and permission record before publishing.
Founder testimonial slot
placeholder
No named quote should render until the quote, attribution, and logo usage are approved.
Operator testimonial slot
placeholder
Use anonymized modeled outcomes only when clearly labeled as modeled.
Recognition or certification slot
placeholder
Do not claim Startup India, DPIIT, SOC 2, investors, or accelerators without evidence.
Trust architecture
A commerce AI employee needs more than a prompt. It needs boundaries, approvals, evidence, observability, and a way to prove what happened.
Start with drafting and recommendations. Increase autonomy only for low-risk work that has proven accuracy and audit coverage.
Every answer should be grounded in customer context, commerce data, and approved knowledge instead of unsupported guesses.
Design every product around separate tenant data, role-based permissions, encrypted secrets, and strict access boundaries.
Track model calls, validation outcomes, approval activity, blocked sends, repair paths, and trust metrics.
RBAC
Roles and approval permissions
Audit
Every decision reviewable
RAG
Approved knowledge and citations
Evals
Scenario checks before autonomy
Who this is for
A good company homepage should help every viewer understand where they fit in a few seconds.
Get support leverage without handing your customer voice to an unchecked bot.
Turn messy customer messages into structured work with evidence, status, and ownership.
Move faster through queues while sensitive customer decisions stay reviewable.
Review the architecture: connectors, tenant boundaries, audits, AI validation, and future model routing.
Working product
The current demo shows the V2 Inbox workspace: dashboard triage, Customer 360, knowledge citations, AI health, audit proof, and connector setup.
Build trust first
See Inbox first. Then we expand the same architecture across the commerce work that needs speed, context, approval, and accountability.