Everything Deskhand does. And how it stays safe doing it.
An AI support agent with access to real orders needs more than good answers — it needs verification, policy, and a paper trail. Here's the whole picture, feature by feature.
Read · Free tier
Answers from your store, not about it
Every answer is pulled live from WooCommerce at the moment of asking. No stale index, no “according to our FAQ…”.
Order status & history
Status, dates, line items, totals and fulfilment state for the customer's verified orders — answered in seconds, day or night.
Shipment tracking
Reads tracking details from the most common WooCommerce shipping plugins, with a manual mapping fallback for anything unusual.
Product info & live stock
Catalogue questions answered from real product data, respecting your catalogue visibility and stock display settings.
Store policies & FAQs
Returns windows, shipping times, delivery zones — authored by you once, quoted accurately by the agent forever.
Human handoff
Anything it can't resolve becomes a tidy escalation to your team with the full conversation attached. Always on, always free.
Conversation log
Every conversation recorded in your dashboard, so you can see exactly what customers ask and how the agent answered.
Verify · Free tier
Nobody sees an order they don't own
Logged-in customers
The order must belong to their account. No match, no data — the agent won't even confirm the order exists.
Guests
A single-use 6-digit code goes to the email address stored on the order — never one the visitor types in. Ten-minute expiry, five attempts, then a lockout. That's proof of ownership, not a formality.
Lookups return the same response whether an order exists or not, so the chat window can't be used to fish for valid order numbers or emails.
Act · Pro
The part nobody else ships: it acts
Three write actions, each policy-gated, each off by default, each logged. Enough to resolve the tickets that eat your day — and nothing reckless.
Cancel orders
Within the window you define — by status, age, order value, payment method, and whether it's shipped. Cancelled orders get an order note and an audit entry.
Change shipping addresses
Fix the typo before the parcel ships to the wrong house. Only before dispatch, only under your rules, optionally same-country only.
Request returns
Opens an RMA for eligible items within your returns window. The customer leaves with a resolution, not a form to find.
Deliberately absent: refunds. Cancelling an order is reversible housekeeping; moving money back out of your account is not. That stays human.
Govern · Pro
A policy engine that reads like a sentence
No rule language to learn, no JSON to edit. Your policy is plain English in the admin, and deterministic code at runtime — evaluated against live order state at the moment of action, never what the model remembers.
Allow customers to cancel an order when the order is pending or processing, placed less than 24 hours ago, under £250, and not yet shipped. Otherwise, escalate to a human.
Three presets ship in the box — Conservative, Balanced, Permissive — so you can be live in one click and tighten from there.
Oversee · Pro
Trust it because you can check it
Immutable audit trail
Every proposed action, the policy decision, and the outcome — written once, never edited. Plus a WooCommerce order note on every change, right where your team already looks.
Simulate mode
Run any customer message through the agent as a dry run before a single write tool is enabled. See what it would do, then decide whether to let it.
Rate & spend limits
Per-session and per-IP rate limits, a conversation-length cap with forced handoff, and a monthly spend ceiling with your choice of behaviour on breach.
The widget · Free tier
Invisible to your Core Web Vitals
It loads on every storefront page, so it's built like it matters: a self-contained web component under 25KB gzipped, styles included. No React, no framework, no fight with your theme.
Free answers. Paid resolves.
The free tier is a complete product — reading, verifying, answering and escalating, forever, with no caps on conversations, messages, or tokens. Paying unlocks the part where the agent acts: write actions, the policy engine that governs them, and the audit tools that let you trust them.
| Capability | Free | Pro | Agency |
|---|---|---|---|
| Chat widget, streaming replies, theming | ✓ | ✓ | ✓ |
| Guest identity verification (one-time code) | ✓ | ✓ | ✓ |
| Order status, tracking, product & stock answers | ✓ | ✓ | ✓ |
| Store policy & FAQ answers | ✓ | ✓ | ✓ |
| Human handoff / escalation | ✓ | ✓ | ✓ |
| Conversation log | ✓ | ✓ | ✓ |
| Bring your own model API key | ✓ | ✓ | ✓ |
| Policy engine with presets | — | ✓ | ✓ |
| Cancel orders | — | ✓ | ✓ |
| Change shipping addresses | — | ✓ | ✓ |
| Request returns / RMA | — | ✓ | ✓ |
| Audit trail view | — | ✓ | ✓ |
| Simulate — dry-run the agent safely | — | ✓ | ✓ |
| White-labelling (admin + widget) | — | — | ✓ |
| Central multi-site policy deployment | — | — | ✓ |
| Pooled model billing (coming) | — | — | ✓ |