26 things you don't have to build
Everything below is shipped and running today — the widget your visitors see, the desk your agents work in, and the chat that lives inside your own admin panel.
The chat your visitors see
A bubble on every page that answers in under a second, opens conversations on its own, and looks like it was built by your designer.
AI trained on your own content
Upload documents or point us at your website — pages are crawled, chunked and embedded, and the assistant answers from that content only. Nothing in your knowledge base, nothing in the answer.
Real human takeover
One click mutes the AI and hands the conversation to a person, mid-chat, with no page reload. The visitor sees the switch: the bubble is relabelled with the agent's name.
Proactive triggers
Open the widget by yourself after N seconds on a page with a message you write — the classic "Need a hand with checkout?" nudge. Configured in Settings and enforced at the gateway.
A widget that looks like you
Solid colour or a two-stop gradient at any angle, ten launcher icons, an optional label and logo, left/right docking, corner radius, square or rounded bubbles, light or dark panel — with a live preview while you edit.
Conversations that close themselves
Idle chats park as Pending and auto-resolve after your own window (30 minutes by default) with a closing message. Any new message reopens the thread, so a returning visitor never loses their history.
One script tag
Paste a single tag before </body>. The widget renders inside a shadow root, so it can't collide with your CSS and your page can't leak into it.
A support desk, not a notification feed
The three-pane inbox your agents live in all day: conversations on the left, the thread in the middle, everything you know about the visitor on the right.
Shared team inbox
Open, Pending, Resolved and unassigned queues with live counts, full-text history, and states that update for everyone at once. No refresh button anywhere.
Auto-distribution
New chats go to whoever is actually available, balanced by current load, instead of sitting in a pile nobody owns.
Saved replies
Type a shortcut, get the full answer. Managed in Settings, shared across the team, so the fifteenth refund question reads exactly like the first.
Private notes & assignments
Notes the visitor never sees, attributed to the person who wrote them, plus assignment so two agents never answer the same chat twice.
Visitor profile
Device, browser, IP-derived location and the page they're chatting from — beside the thread, before the first reply goes out.
Desktop notifications & sound
Agents get a native notification and a chime the moment a chat needs them, even when the tab is buried.
Team with real permissions
Invite members and grant Inbox and Knowledge access separately. Admins manage settings and billing; agents just answer.
Activity tracking
Every takeover, reply, note and resolution is attributed to a person and lands in a filterable feed — accountability without a spreadsheet.
The part other chat tools don't have
Support chat is only half of it. The same platform also runs the internal chat inside your back-office — the one your staff, resellers or franchisees use to reach the person above them.
Admin-side inbox widget
A second script (panel-widget.js) for your own admin panel. It opens as an inbox of the logged-in person's direct upline and downlines, with presence, unread badges, read receipts and typing indicators.
A separate panel assistant
Staff can ask the Assistant how your back-office works. It answers from an admin-panel knowledge base kept apart from the public one — internal documents never reach a website visitor.
Admin Inbox oversight
The whole hierarchy drawn as a graph: thicker lines for busier threads, a red dot for unread. Touch a line to read that conversation. Read-only, for the top admin who needs to see everything.
Embed the portal itself
Put the entire support portal in an iframe inside your own admin panel, so your team never learns a second tool or logs into a second product.
Nothing is lost while you're away
Message someone who is offline and they get a clear "saved — they'll reply soon" notice, with the message waiting in their inbox when they sign back in.
Per-account switches
The internal chat and the oversight graph can be granted to a single account regardless of its tier — a plan is a default here, not a cage.
Built to be run in production
Multi-tenant from the first migration, metered honestly, and deployable without handing your customers' conversations to a third party.
Isolation at the database
Every table carries a tenant id and Postgres row-level security enforces it, so one customer's data cannot be read from another's session even if application code slips.
Usage billing you can explain
A fixed monthly base plus metered conversations and AI tokens. Each month closes into one invoice showing the full counts — base, usage, totals.
Hard limits, no surprise bills
Reach your plan's message ceiling and the AI pauses instead of quietly billing on. You'll have seen the 80% warning long before that.
Your model, your choice
The AI provider is one switch: a hosted Claude model, any OpenAI-compatible endpoint, or a local model on your own hardware for conversations that must never leave the building.
Safe by construction
Tenants store an icon key, never markup; logo URLs are restricted to http(s); nothing customer-supplied is ever interpreted as HTML in the widget.
Production guardrails
Demo data refuses to seed in production and services that sign tokens refuse to boot with a weak or default secret. The unsafe path isn't documented away — it's blocked.
A pill means a paid tier — nothing else is held back
Cards marked Mini+ or Essentials+ unlock on that tier and every tier above it. Everything unmarked — the assistant, human takeover, the shared inbox, theming, notifications, the whole conversation lifecycle — is on the free plan too. The admin-side inbox can also be switched on for a single account regardless of its tier; just ask.
The things people ask first
Will the AI make things up about my business?
It answers from your knowledge base — the pages you import and the documents you upload — and says it doesn't know rather than inventing a policy you never offered. When it can't answer, a human takes the chat.
What counts as usage?
Handled conversation messages and the AI tokens they consume. Your plan's monthly base covers an allowance; anything past it is metered at a per-unit rate and itemised on one invoice at the end of the month.
What happens when I hit my plan's limit?
The AI pauses until the next cycle or an upgrade — your team can still reply by hand, and you get a warning at 80% long before the ceiling.
Can I use this inside my own admin panel?
Yes, two ways. The whole support portal can live in an iframe inside your product, and a second widget turns your back-office into an internal inbox where staff chat with their upline and downlines and ask a panel-specific assistant.
Do my conversations have to go to a third-party AI?
No. The AI provider is a configuration switch — point it at a local model running on your own hardware and no message content leaves your infrastructure.
Can I change plans or cancel?
Any time. Plans are month to month; yearly billing simply pays for ten months and gives you twelve. Upgrades take effect immediately, and limits follow the new plan the moment it's applied.
See it running before you decide
Two demos, no signup: the widget on a website, and the team chat inside an admin panel.