Get started

Act · QuoxChat · grounded chat

AI chat widgets for any website. $199 a year flat. BYOK. Unlimited.

A branded assistant that answers from your docs, your Q&A and your data, not the open web. No per-message fees, no conversation caps, no markup.

Standard $199/yrSecurity layers 7Adversarial scenarios 49CSS control levels 5
avatar state idle

Your widget's avatar · it reacts as the conversation moves

In plain words

What it is, where it lives, when to reach for it

What is it
A chat widget for your website that answers visitors from your own docs and data.
Where do I use it
On your website's pages; you manage it from the QuoxChat page in the dashboard.
When would I use it
When visitors ask the same questions your docs already answer and nobody has time to reply.
How do I use it
Create a widget on the dashboard's QuoxChat page, then paste its embed script into your site.

What it is

Grounded in your docs. Not the open web.

Deploy it on your site in 15 minutes: trained on your knowledge, styled to your brand, connected to your systems.

QuoxChat is an embeddable RAG chat widget. You give it your PDFs, your site, your Q&A pairs. It retrieves from that knowledge and only that knowledge, cites where each answer came from, and hands over to a human when it should.

The demonstration

Watch it answer, act and hand over.

A replay of the widget at work. The visitor asks, the widget retrieves from the knowledge base and cites its sources, and when the visitor asks for a person, a tool call pauses the thread and routes it to a human with the full context attached. The avatar reacts as the states change, exactly as it does on a live site.

  • Grounded answers. Every reply is retrieved from your own knowledge base, with the sources shown.
  • Conversation memory. The thread survives page changes, and a handoff carries the full context.
  • Embeddable anywhere. One script tag on any site, or go headless with the SDK.
  • BYOK. LLM calls go straight from your QuoxCORE instance to your provider, on your key.
AI disclosure built in: the widget identifies itself as an AI assistant to every visitor, in line with EU AI Act transparency requirements.
Ask us anything online
AI assistant
Type a question…
Talk to a human

A scripted recreation, not a live model. The source chips are the knowledge files a live widget cites with every answer.

The widget itself

Not a mockup. This is the real interface.

The replay above is a scripted recreation. Here is an actual QuoxChat widget: grounded answers with cited sources, styled to match the site it sits on.

A live QuoxChat widget answering a visitor's question with cited sources

Core capabilities

Everything you need on every tier.

Audit trail, white-label, human handoff and the Widget Room builder come standard. The Professional tier adds the agentic features: tools, smart routing, simulation.

Knowledge pipeline

Upload PDFs, crawl URLs, add Q&A pairs. Every widget gets its own vector collection, so one customer's knowledge never leaks into another's. Pinned answers take priority over retrieved chunks.

Five levels of CSS control

Theme presets, CSS variables, ::part() selectors, full stylesheet override, or go headless with the SDK. From drop-in to pixel-perfect brand match with no hacks.

White-label by default

No "Powered by Quox" badge. No attribution. Your widget, your brand, on every tier. Your customers never see our name.

BYOK: bring your own key

Use Claude, GPT, Gemini, or your own self-hosted models. Your API key, your costs, your data. No per-message fees, no conversation caps, no markup on LLM usage.

Full audit trail

Every message traced with AEE correlation IDs. Drill into any conversation to see the system prompt, retrieved sources, tool calls and confidence score. Ready for SOC 2 and GDPR audits.

Human handoff

When the AI can't help, a tool call pauses the conversation and routes it to a human with full context. No losing the thread, no re-explaining.

Headless SDK build your own UI

NPM package. We handle RAG, LLM routing and audit trails. You own the UI. Ship a widget that doesn't look like every other chatbot.

7-layer security architecture injection defence

Safety preamble, injection pattern detection, output filtering, domain allowlist, rate limiting, tool approval gates, and visitor rate limits. Works by default.

Professional tier adds

Tool and connector framework chatbots that act

Your widget can query your database, call your REST API, hit webhooks, or trigger QuoxFlow workflows. Turn a support chatbot into a self-serve portal.

Verified account actions server-signed identity

Your site signs each logged-in visitor with an HMAC, and the server injects that verified identity into every REST tool call. The model never chooses who it acts for, so a prompt injection cannot act on someone else’s account. Signed-out visitors get a sign-in invitation you write yourself.

Metered AI actions allowance per visitor

Give each signed-in visitor a monthly allowance of AI-performed actions, counted durably on the server. Your site can raise a customer’s allowance in the signed identity payload, so limits grow with what they pay for. Read-only help stays free.

Smart LLM routing 70% cost reduction

Simple questions go to cheap models, complex reasoning goes to flagship. Same quality at a fraction of the token cost.

Simulation Arena red-team before you ship

49 adversarial scenarios run against your widget before it goes live. The security score gates production deployment.

Professional tier

Tools and connectors, configured per widget.

On the Professional tier a widget can query your database, call your REST API, hit webhooks, or trigger a QuoxFlow workflow, turning a support chatbot into a self-serve portal.

The QuoxChat tool and connector framework, configuring an action a widget can take

Now real

Five actions, not just answers.

QuoxChat does not just answer questions anymore. It captures leads, emails transcripts on request, hands off to a human without losing the thread, stays scoped to your product by default, and logs every one of those actions the same way it logs every message.

01

Captures the lead, no form

When a visitor shows intent, their name, email and what they are after gets saved straight into your leads, no contact form in sight. The conversation is the form. Your team is notified in the governed inbox the moment it lands.

Lead capture
02

Emails the transcript, on request

A visitor can ask to have the chat sent to their own address. A captcha gates the request so bots cannot use it to spam mailboxes, and the send itself is a witnessed action, logged like everything else. Available per site: add a reCAPTCHA key to the widget to turn it on. Not on by default.

Email transcript
03

Hands off with the full thread

When the agent hands off to a person, it lands in your governed human-in-the-loop inbox with the whole conversation attached, not a dead end. Every handoff is logged the moment it happens.

Human handoff
04

Scope-locked to your product

The widget only answers from what you have given it and refuses to wander into off-topic requests, so it stays on brand and does not burn your API budget on questions that have nothing to do with your product.

Guardrails
05

Every action, in the ledger

Lead captures, transcript sends and handoffs are all recorded as AEE events, the same audit trail as every message, so you can prove what the agent did. See a real trace in the audit trail section below.

Witnessed actions

Your knowledge, your key, your servers.

Knowledge uploads and conversation logs live on your QuoxCORE instance, Quox-hosted or self-hosted. BYOK means LLM calls go directly from QuoxCORE to your provider, with no third-party middleman. And every message is traced with an AEE correlation ID you can read, export and audit.

Why not just use Intercom?

Because they charge per seat, per resolution, and lock features behind enterprise tiers.

We don't. Flat annual pricing, every core feature on every tier, and the agentic features on Professional. Intercom charges $39 per seat per month plus $0.99 per AI resolution; Drift charges $2,500 a month. QuoxChat Standard costs what a single Intercom seat costs per year.

CapabilityIntercom FinDrift AIHelpScout AIQuoxChat
Entry price$39/seat/mo + $0.99/resolution$2,500/mo$65/user/mo + $20/AI seat$199/yr flat
BYOKNoNoNoYes
Unlimited conversationsPer-resolution feesThrottledThrottledYes
White-labelEnterprise onlyAdd-onNoAll tiers
Audit trail per messageLimitedLimitedLimitedAEE per message
Self-host optionNoNoNoYes
Tool-calling frameworkLimitedLimitedNoYes (Pro)

The audit trail

Every conversation, traced and reviewable.

Drill into any conversation to see the system prompt, retrieved sources, tool calls and confidence score. Knowledge uploads and conversation logs live on your QuoxCORE instance, Quox-hosted or self-hosted.

quoxcore · quoxchat logs
The QuoxChat Logs tab: a filterable list of six conversations with CSV and JSON export, the selected knowledge-tab-test thread showing a visitor question and a six-point assistant reply, a Medium 72 percent confidence badge, a Context chip, and an Export CSV button.
  1. 1Six conversations, filterableSearch by message or visitor, filter by status or rating, or export the lot as CSV or JSON.
  2. 2Know exactly which thread you are inThe open conversation is highlighted in the list, alongside two other knowledge-tab-test runs, a probe and a live visitor session.
  3. 3The full exchange, timestampedThe visitor asks "what can I do with a quox?" and the assistant replies at 20:28:19, 759 plus 388 tokens.
  4. 4A structured, sourced answerSix numbered points, cryptographic proof, persistent memory, governance, self-hosting, trust protocols and getting started, not a one-line reply.
  5. 5A confidence score on every replyThis one reads Medium at 72 percent, with a Context chip to see what it drew on.
  6. 6Export the transcriptExport CSV sits right above the reading pane, ready whenever a reviewer needs the record offline.

Captured from the real QuoxChat Logs tab, not a mockup.

Use cases

Built for every front-line team.

01Customer support

  • Deflect repetitive questions with grounded, cited answers
  • Escalate complex cases with full context to a human
  • Auto-capture leads when a visitor shows high intent
  • Works 24/7, multi-language, scales to traffic spikes

02E-commerce

  • Answer product questions from the catalogue
  • Recommend items based on the visitor's stated need
  • Trigger checkout actions or apply discount codes
  • Handoff to a human for high-value orders

03Documentation sites

  • Turn docs into a conversational search
  • Cite the exact doc page the answer came from
  • Suggest next steps or related tutorials
  • Surface common confusion points to the docs team

04SaaS onboarding

  • Guide new users through their first setup
  • Answer feature questions while they explore the dashboard
  • Link them to the right docs and videos
  • Log questions to surface product gaps

The embed

Install in two lines.

Paste before the closing body tag. Works on any site, any stack, any CMS.

index.html
<script src="https://your-core.com/quoxchat/widget.js"
    data-widget-id="wid_abc123" async></script>
1 Create the widget in the dashboard's Widget Room.
2 Upload your docs or point it at a URL.
3 Copy the snippet, paste it into your site. Under 15 minutes end to end.
Maestro, the AI builder agent inside QuoxCORE, can generate this snippet, configure the widget, upload your knowledge and test it for you.

Pricing

Two tiers. One honest price each.

No seat fees. No per-message fees. No LLM markup. Your API key, your cost. Both tiers get unlimited widgets per org and unlimited messages per month.

Standard

Everything grounded

$199 /year

Unlimited widgets and messages, with 1 GB of knowledge storage.

  • Widget Room, the 7-tab builder
  • Knowledge pipeline: files, URLs, Q&A
  • 5-level CSS control
  • White-label
  • Budget management
  • Full audit trail (AEE)
  • Human handoff
  • Headless SDK

Professional

Everything agentic

$499 /year

Everything in Standard, with 5 GB of knowledge storage, plus the agentic features.

  • Tool and connector framework
  • Smart LLM routing
  • Simulation Arena, 49 adversarial scenarios
  • Proactive engagement
  • Voice I/O
Free tier: one widget, 1,000 messages a month, 50 MB of knowledge. No card to start.

Questions

Common questions.

Is there a per-message or per-resolution fee?

No. Flat annual price, unlimited conversations. The only variable cost is the LLM calls, and those go through your own API key (BYOK).

Can I use my existing LLM account?

Yes: Claude (Anthropic), GPT (OpenAI), Gemini (Google), or any OpenAI-compatible endpoint including self-hosted models via Ollama or vLLM. Configure once per widget.

How does QuoxChat compare to Intercom Fin?

Intercom Fin charges $39 per seat per month plus $0.99 per AI resolution. A mid-sized support team with 10,000 resolved conversations a month costs $9,900 a month. QuoxChat Standard is $199 a year flat with no resolution cap.

Can the widget take actions, not just answer?

On the Professional tier, yes. The tool and connector framework lets your widget query databases, call REST APIs, trigger webhooks, or launch QuoxFlow workflows. Account actions run under a server-signed visitor identity, and you can meter them with a per-visitor monthly allowance. Turn a chatbot into a self-serve portal.

What happens to my data?

Your knowledge uploads and conversation logs live on your QuoxCORE instance, either Quox-hosted or self-hosted. BYOK means LLM calls go directly from QuoxCORE to your provider, with no third-party middleman.

Is there a free tier?

Yes: create one widget, 1,000 messages per month, 50 MB of knowledge. Pro features (tools, smart routing, simulation) are Professional-only.

How long does it take to set up?

Under 15 minutes. Create the widget in the dashboard, upload your docs or point it at a URL, copy the embed snippet, paste it into your site. Maestro, our AI builder agent, can walk you through every step.

Where it stands

What's live, what's beta, what's parked.

Live and proven
  • Widget serving is stable and live on quox.ai itself, rate-limited and entitlement-gated
  • Conversation logs are stable: sorted, day-labelled, and read in plain language
Beta and partial
  • Human handoff into a governed room is proven end to end, but an operator must reply from the dashboard: there is no phone-side reply path yet
  • A fresh widget ships with no tools seeded, so lead capture, transcript email and handoff sit inert until an admin wires them up
  • The BYOK failover chain is visible in evidence, but the current fallback hop drops tool calls, so a degraded provider also loses handoff and actions
Parked, not shipped
  • Inbound email into a conversation is deliberately deferred
  • Twilio call-me is parked, not scheduled
  • Telegram operator relay is blocked pending a bridge relay mode

Ready when you are

Ship a widget this week.

Start free. Upgrade when you need tools, smart routing, or the Simulation Arena.

Standard $199 a year · Professional $499 a year · free tier to start