Self-hosted Real-time AI-augmented Multi-channel Multi-project

OnlineChatCRM

Run Meta & Google ads straight to your project URL — same page captures the lead, starts live chat, and fires conversion events back to your ad platform so campaigns auto-optimise. WhatsApp-grade messaging, AI fallback agent, automated routing, complete CRM, and a back-office for admins, managers, agents, and QA — all in one self-hostable .NET application.

80+

Features

User Roles

1M+

Chats Scale

<100ms

Real-time Delivery

What is it?

One platform. Every customer conversation.

OnlineChatCRM is a complete live-chat and CRM platform that combines real-time messaging, AI fallback agents, automatic lead routing, a fully-featured back-office, multi-project branding, and an embeddable widget for any third-party website — all in a single self-hostable application backed by SQL Server.

Multi-channel live chat

Visitors open chat from a standalone page /p/{slug} or via an embeddable widget on any external site — both share the same conversation backend.

Multi-project tenanting

Run unlimited brands from one install. Each project has its own slug, branding, accent colour, welcome message, allowed embed domains and AI instructions.

Four-tier role model

Admin, Manager, Agent, and Quality Analyst — each with a tailored console and permission scope. Agents have independent Live Chat & CRM toggles.

AI agent with smart handoff

Optional virtual agent answers when no human is online and explicitly escalates to humans on demand. Per-project system prompt; respects 24/7 availability rules.

Round-robin auto-assignment

Queued chats route to available agents using a sticky round-robin index. Each agent has independent active & offline chat limits. Manual reassign on demand.

Built-in reporting

Chats, Chat data, User data, User analysis, CRM reports — all paginated, date-/agent-/stage-filterable, exportable as CSV. Single-click PDF transcript per chat.

PWA + system notifications

Installable on desktop & mobile. Agents receive system notifications, sound pings, and a numeric app-icon badge even when the dashboard tab is backgrounded.

Lead automation

Every chat becomes a lead with full UTM + device + geo attribution. Auto qualification rules, lead-stage workflow, follow-up scheduling, per-lead note trail.

Block abusive visitors

Admin or manager bans a visitor by IP in one click. Future chat-start attempts from that IP return 403 silently. Captured at first chat start, audit-trailed.

What sets it apart

Engineered for the long tail

Most platforms get the demo right and the long-tail wrong. These are the details that matter once you cross 10,000 conversations.

WhatsApp-style read receipts

Single ✓ (sent), double ✓✓ (delivered), double ✓✓ in green (read). Updates live without anyone refreshing.

Bidirectional typing indicators

Visitor sees the support pill the second the agent types; agent sees the same pill the second the visitor starts typing. Both directions, real-time.

15-second visitor offline grace

Brief tab switches and network blips don't flicker the agent UI — the chat stays in the Active bucket for 15s before the system marks it offline.

30-second agent refresh grace

An agent refreshing their browser doesn't lose their chats — reassignment fires only after 30s of disconnect, so a refresh or network blip is invisible.

Lazy message history

Long chats load the last 30 messages instantly; older ones stream in as the agent scrolls up. Chats with 50,000 messages still open in < 1 second.

Infinite-scroll inbox

Initial load shows 20 chats per bucket; the rest stream in as the agent scrolls. Active / Offline / Closed / Unassigned / Others — same pattern for all.

AI human-disguise

1-second delay before the AI shows the typing pill. Auto-marks the visitor's previous messages as read the moment AI starts composing — just like a human.

App-icon badge

Installed PWA shows a numeric badge on the home-screen icon, dock, or taskbar when there are unread chats. Cleared automatically when unread hits zero.

One-click IP ban

Block abusive visitors from the chat modal in one click. The IP is captured at first chat start; future chat attempts from that IP return 403 silently.

Auto cookie / cache hygiene

Staff browsers self-prune oversized localStorage and orphan entries every 30 min. Visitor side is intentionally never touched (session-critical data).

One-click PDF transcripts

Download any conversation as a multi-page PDF from the Chat data table. Colour-coded by sender, with timestamps and attachment links.

Real activity sort

The Updated column shows the actual last-message time, not when someone clicked the row. Opening a chat or changing a lead stage no longer disturbs sort order.

Agent permission split

Live Chat and CRM Leads are independent toggles per agent. Build a chat-only team, a CRM-only team, or both — without giving the wrong access.

Per-project AI personality

Each project has its own AI system prompt. E-commerce, healthcare, and finance brands get totally different bot behaviour from one install.

Server-paginated everything

Every report — chats, leads, user data, analysis — paginates server-side. Browser never receives more rows than the visible page. Scales to lakhs of rows.

For your customers

Visitor Experience

Two front-ends: a standalone branded page at /p/{slug} and an embeddable widget. Both share the same conversation backend and per-project branding.

Visitor page — standalone branded chat page at /p/{slug} with per-project header colour, welcome heading, and contact form.

Per-project branding

Each project picks its own title, subtitle, accent colour, welcome message, contact-form heading, contact-form description, and favicon/icon — all editable from the admin Projects tab.

Custom coloursCustom copyProject iconPer-project URL

Smart contact form

Name, country-selector phone (E.164 validation, per-country length rules), and email. Returning visitors with a stored token skip the form completely.

E.164 phone validationCountry pickerEmail validationToken resume

Image attachments

Visitor uploads an image with live preview before sending; agents receive it as a thumbnail bubble with full-resolution click-through. Lazy-loaded to keep history fast.

Preview-before-sendServer-side validationLazy load

Reply-to quoting

Click reply on any message and the composer shows a quoted preview. Click a quote bubble to scroll back to the original message; the original briefly highlights so the visitor finds it instantly.

Click-to-scrollLive-updates

Installable PWA

"Add to home screen" prompt for mobile / Chrome; iOS Safari instructions are auto-detected. Theme colour, manifest, service worker — full PWA stack so the chat behaves like a native app.

iOSAndroidDesktop Chrome

Source & intent capture

Every chat automatically captures: source domain, referrer, landing URL, UTM source/medium/campaign, device type, OS & version, browser & version, language, connection type, geo city/region/country, time on page, and returning-visitor flag. Available in the User analysis report.

UTMDeviceBrowserGeoReturning visitorTime on page

Cross-session resume

A project-scoped cookie + localStorage token persists across browser sessions. Visitor returns and the conversation continues exactly where it ended — regardless of how many months later. Tokens are project-isolated so the same browser visiting two of your projects keeps each conversation separate.

Another project, same backend — this project (slug rahul) uses a different accent colour, welcome heading, and contact-form copy. Backend code is identical.

Drop-in for any site

Embeddable Chat Widget

A 60 KB script that opens a Shadow-DOM-isolated chat bubble on any third-party website. Same backend, same per-project branding, same features as the standalone page.

Embedded widget on a third-party page — bubble bottom-right, full chat panel on click.

Standard embed

One-line script tag

Drop a single async script before </body> and the widget is live.

<script async
  src="https://yourdomain/widget/widget.js"
  data-project="default"></script>

Trigger link

Any element opens the widget

Add data-cp-open to a link / button anywhere on the page.

<a href="#" data-cp-open>
  Chat with us
</a>

Popup modal

Auto-open on page load

Centred iframe popup throttled to once per 7 hours per browser via localStorage. The throttle key is project-scoped so different projects can coexist on the same site.

CORS controlled

Allowed embed domains

Each project has its own domain whitelist. Wildcard subdomains supported (https://*.client.com). Changes apply within ~30 seconds, no restart needed.

Widget features at a glance

Shadow DOM isolation

Customer-site CSS can't bleed into the widget. Widget CSS can't bleed out. Always pixel-perfect.

Position config

Bubble can sit bottom-right or bottom-left per project setting. Customised via the embed tag's data-position attribute.

Unread badge

Red counter on the bubble when chats arrive while the widget is closed. Resets the moment the visitor opens the panel.

Lazy chat history

Returning visitor sees last 30 messages instantly; older messages load on scroll-up — just like WhatsApp Web.

Get the right agent on the right chat

Routing & Auto-Assignment

Smart routing rules decide whether a new chat goes to a human, the AI, or sits in the queue — based on availability, agent load, and a fair round-robin rotation.

Round-robin auto-assignment

When a visitor starts a chat, the server picks the next eligible agent using a persistent counter (so the rotation survives restarts). "Eligible" means: online + live-chat-enabled + not on break + below their active-chat limit + not assigned a duplicate by phone number.

Persistent counterSkip-if-overloadedPhone-dedup

Drain queue on agent online

The moment an agent comes online, the server drains the pending queue and assigns waiting chats to them — honouring all the same eligibility rules. No chat sits in queue when capacity exists.

Active and offline chat limits

Each agent has two independent limits: how many active chats they can hold at once (e.g. 3), and how many offline chats they can keep parked in their inbox for follow-up later (e.g. 5). Set per-agent in the Agents tab.

Per-agent limitsOverflow falls back to queue

Visitor offline sweeper

If a visitor disconnects and doesn't return within 15 seconds, the chat moves from Active to Offline. The original agent keeps it (so they can resume), unless they're offline themselves or over their offline limit — in which case it returns to the queue.

15s thresholdBackground sweeperAuto-reassign on overflow

Manual reassign by agent

An agent can hand their own chat off to another currently-online live-chat-capable agent in one click. The server validates the target is online, not on break, and not at their cap before accepting.

One-click handoffTarget validation

Manual assign by admin / manager

From the Chats modal an admin or manager picks any agent and assigns the chat. The selected agent is added to the chat instantly and notified via SignalR.

From Chats modalLive SignalR push

Agent inactivity sweeper

Background service logs out agents whose session hasn't been touched in 5+ minutes while holding open chats. Prevents zombie agents holding chats hostage. Bumped by a heartbeat ping so a reading agent isn't kicked.

On-break toggle

An agent on break gets no new chats — but keeps the ones they already have. Toggle on / off in one click. Useful for short pauses without losing context.

A chat's full journey

Chat Lifecycle & Status Buckets

Every chat moves through a clean state machine. Status changes happen automatically based on visitor and agent activity, so agents always see the right chats in the right bucket. No manual bookkeeping needed.

1. Queued (Unassigned)

When a visitor starts a new chat and no agent has been auto-assigned yet, the chat sits in Unassigned. Shows up in every agent's "Unassigned" tab so anyone can pick it up. Also auto-assigned via round-robin the moment an eligible agent becomes available.

Visible to all agentsAuto round-robinPickable manually

2. Active (Open + Visitor Online)

Once a chat is assigned to an agent and the visitor is currently on the page, it sits in the agent's Active bucket. This is where live conversations happen. Live typing, instant message delivery, read receipts — all working in real-time.

Live typingRead receiptsReal-time

3. Offline (Visitor Disconnected)

If the visitor closes their tab, switches to a different app, or loses internet, the chat goes into the agent's Offline bucket after a 15-second grace window. The grace stops minor tab-switches from flickering the status. The chat stays parked with the original agent so they can reply when the visitor returns; the visitor's next message wakes it back up to Active automatically.

15-second graceChat parked, not closedAuto-wake on return

4. Closed

An agent or admin explicitly clicks Close on a chat when the conversation is genuinely finished. The chat moves to the Closed bucket. Status flips to "closed" and the visitor sees a small "this chat has ended" banner. If the same visitor sends a new message later, a fresh chat is started instead of resuming the closed one.

Explicit closeVisitor bannerFresh chat on return

5. Reassigned to Queue

If the assigned agent goes offline for more than 30 seconds and their offline-chat limit is full, the system automatically releases the chat back to the queue. The next available agent picks it up via round-robin. Visitor doesn't wait alone.

30-second agent graceAuto-release on overflow

6. Others (View-Only)

The Others bucket shows chats assigned to other agents. Read-only for the viewer. Useful when an agent needs to peek at a teammate's chat for handoff context, but cannot reply. Lead-stage and notes are still visible.

Bucket count badges

Each bucket in the agent's sidebar shows a live count + a red unread badge. The total unread across all buckets also appears in the browser tab title (so "Agent Console" becomes "(3) Agent Console") — agents notice incoming chats even when they're on a different tab.

Where the work happens

Agent Console

An inbox-style dashboard purpose-built for agents handling hundreds of conversations per day.

Agent inbox with chat detail open

Show the bucket toggles, a chat list with unread badges, and a chat detail open on the right with the composer.

screenshots/agent-inbox.png

Five-bucket inbox

Active (live chats with the visitor present), Offline (visitor disconnected, chat parked with this agent), Closed (chat explicitly ended), Unassigned (queued, waiting for any human), Others (assigned to other agents — read-only). Each bucket has its own infinite-scroll backfill, search, and unread badge count.

ActiveOfflineClosedUnassignedOthers

Per-bucket search

Search by visitor name or phone number, debounced. Autofill is actively suppressed on staff search inputs so saved-login data never accidentally appears.

Starred chats

Each agent can star important chats. Stars are per-agent — the same chat can be starred by one agent and not another. Persists server-side across devices and reloads.

Quick replies

Each agent saves up to 5 canned replies. Click the lightning icon in the composer to pop the picker; clicking a reply pastes it into the textarea for review-and-edit before send. Saved server-side, sync across browsers.

One-click reassign to another agent

Hand a chat off to a teammate with a single click. Server validates the target is online, eligible (Live Chat permission), not on break, and below their active-chat limit. Conversation history is preserved.

On-break toggle

Visible online but not eligible for new chats. Existing chats stay assigned. One click on / off.

Lead-stage tagging inline

Mark a lead as New, Qualified, Not Qualified, or Converted right from the chat detail. The tag immediately appears in CRM reports and the admin Chats list.

NewQualifiedNot QualifiedConverted

System notifications + sounds

Triggered on: new assignment, new visitor message in an assigned chat, and new message arriving while the agent's tab is backgrounded. Uses the service worker so backgrounded PWAs still ring.

New assignmentNew messagePWA badgeSound

Session heartbeat

Browser pings /api/agent/heartbeat in the background so reading a long visitor message doesn't trip the inactivity sweeper. Resolves the "got logged out mid-chat" complaint other CRMs have.

CRM tab (if permitted)

Agents with the CRM Leads permission see All leads, Web leads, Uploaded leads, Google Sheet leads — with lead-stage selectors, follow-up date inputs, and final-remark fields. Permission is independent of the Live Chat permission.

Run the operation

Admin Dashboard

Full visibility and control over chats, agents, staff, projects, CRM, AI, and reports.

Admin Overview — live totals (chats, leads, lead stages) and an agents-online grid with current load per agent.

Add new agent. Name, email, password, active chat limit, offline chat limit, plus Live Chat / CRM permission toggles. Manage existing agents below.

Chat data report with the one-click PDF download column on the far right. Date / status / agent / lead-stage filters above; pagination below.

Marketing — User Data report. Visitor name, phone, email, project, source domain, referrer, UTM source / medium / campaign, first-seen time. Paginated, CSV exportable.

AI Integration page. Provider toggle (Claude / OpenAI), model picker, max tokens per reply, system prompt template, fallback rules. Master enable / disable at the top.

CRM — All Leads. Every chat-, web-, CSV-, or sheet-derived lead in one paginated table with inline lead-stage selector, follow-up date input, and final-remark field.

Overview snapshot

Total chats by status, lead-stage breakdown, live agent presence with load. Date-filtered.

Chats

Every conversation in one paginated table. Filter by All / Assigned / Unassigned / Offline / Closed. Click any row to open the chat modal with Assign / Lead Stage / Block Visitor toolbar plus reply composer.

Assign / reassign to any agent
Set lead stage
Block visitor IP
Reply directly from the modal

Agents

Add agents, set active & offline chat limits, toggle Live Chat / CRM permissions, force-logout, edit display name, reset password.

Staff

Manage admins, managers, and quality analysts. Role-aware editing (managers can't promote anyone to admin). Last-admin protection — can't demote the only remaining admin.

Chat data

Detailed per-chat report with Token, Name, Phone, Email, Project, Status, Agent, Lead stage, Updated (= last message time), and a one-click PDF Download column.

PDF transcript per chat
CSV export
Server-side pagination
Date + status + agent filters

Reports

Per-agent productivity, lead-conversion funnel, response-time stats. Date-range filtered. Exportable as CSV.

Projects

Create & clone projects, edit branding (title, subtitle, accent, welcome, icon), manage allowed embed domains (wildcards supported), per-project AI instructions. Pre-rendered embed code generator for widget / trigger-link / popup-modal embeds.

Marketing

User data + User analysis reports (UTM, device, geo) with date filters and CSV export. Marketing Tags surface for GTM / GA4 / Meta Pixel snippets injected site-wide.

AI Integration

Provider (Claude / OpenAI), model, max tokens per reply, system prompt template, fallback rules, escalation policy. Master enable / disable toggle. Per-project AI instructions live inside the Projects tab.

Server-paginated everywhere. Every table in the admin — Chats, Chat data, User data, User analysis, All Leads, Web Leads, Uploaded Leads, Google Sheet Leads — paginates server-side with a per-table rows selector (20 / 50 / 100), Prev / Next, and an "X-Y of Z" indicator. No infinite DOM lists eating browser memory.

A trimmed-down admin

Manager Console

Managers see almost everything admins see — minus Projects and AI Integration. A team-lead view with full operational visibility but no infrastructure access.

Capability	Admin	Manager	Agent	QA
View all chats & reports	Yes	Yes	Own only	Read-only
Assign & reassign agents	Yes	Yes	Own to others	No
Mark lead stage	Yes	Yes	Own chats	No
Block visitor by IP	Yes	Yes	No	No
Download chat PDF	Yes	Yes	No	No
Manage agents (add / edit / force-logout)	Yes	Yes	No	No
Manage staff (admins / managers / QA)	Yes	No admin promote	No	No
Edit Projects (branding, slug, embeds)	Yes	No	No	No
Edit AI configuration	Yes	No	No	No
Edit DB / global config	Yes	No	No	No

Server-side enforced. Hidden UI buttons aren't enough — every endpoint independently checks the actor's role server-side. A manipulated request can't bypass restrictions.

Quality oversight

Quality Analyst Console

A dedicated read-only console for QA reviewers — see every chat in real-time, but never send a message or change state.

Live chat monitoring

Every conversation update streams to QA in real-time via SignalR — new messages, typing pills, receipts, agent presence.

Filter by agent & status

Drill into a single agent's work or focus on offline / closed reviews. The AI shows as a virtual "AI" filter so QA can isolate bot-handled chats for grading.

Full history with lazy load

Long chats open in 30 messages and stream the rest on scroll-up — same UX as agents.

Total chat control

Chat Management

From the Chats modal, an admin or manager has every lever they need on a single chat.

Chat history modal. Visitor info on top · Assign agent / Lead stage / Block Visitor toolbar · full chat thread below · reply composer at the bottom.

Lead-stage tagging

Mark New / Qualified / Not Qualified / Converted from the modal dropdown. Change is immediate and flows into CRM reports. Admin / manager / agent on assigned chats / QA — all four can mark, with role-aware authorisation.

NewQualifiedNot QualifiedConverted

Manual agent assignment

Pick any agent from the dropdown and click Assign. The new agent is added to the conversation group, joins the chat's SignalR room, and receives the conversation history — no refresh needed on their side.

One-clickLive SignalR push

Block visitor (IP ban)

One click on Block Visitor: the source IP (captured at first chat start) is added to BlockedVisitors, the current chat is closed, and future chat-start attempts from that IP return a generic 403. Reason and operator are recorded for audit.

IP banAudit trailAuto-close current chat

One-click PDF transcript

From the Chat data table, click the PDF button on any row to download the full conversation as a multi-page PDF. Header shows visitor info + project + status + agent + lead stage; body shows every message colour-coded by sender with timestamps and attachment links.

QuestPDFPer-conversationMulti-page

Close chat

Explicitly close a chat from the modal — status flips to Closed, conversation moves out of the agent's Active bucket, and the visitor sees a "this chat has ended" banner. Re-engagement still creates a new chat (or resumes the closed one if the visitor returns and types).

Reply from the modal

Admins and managers can reply to a visitor directly from the chat-history modal — their reply appears as a "from-admin" bubble (distinguishable from agent bubbles in the visitor's chat).

From chat to closed deal

CRM & Lead Management

Every conversation automatically becomes a lead. Multiple ingestion sources flow into a unified inbox, with lead-stage workflow, follow-up scheduling, and notes.

Four lead sources

All conversations (every chat is a lead), Web leads (public POST endpoint for landing-page forms), Uploaded leads (CSV upload from admin), and Google Sheet leads (background poller pulls from configured sheets every 60s). Each source is tracked separately and rolled up in the unified view.

ChatsWeb formsCSV uploadsGoogle Sheets

Lead-stage workflow

Four stages: New, Qualified, Not Qualified, Converted. Every lead starts as New; any operator with the right permission can move it forward. Stage filters across every CRM report.

NewQualifiedNot QualifiedConverted

Follow-up scheduling

Each lead carries a follow-up date and a remark. Agents see "due today" leads surfaced on their dashboard so nothing slips. Date pickers are inline in the CRM tables.

Per-lead notes

Append-only note trail per lead — every agent interaction is attributable. Final remarks are separately editable. Notes are timestamped and operator-tagged.

Google Sheets sync

Background poller pulls leads from configured Google Sheet sources every 60s (interval is configurable per source). Uses a service account so there's no per-user OAuth dance.

Permission-scoped CRM

Agents toggled off "CRM Leads" never see the tab; agents toggled off "Live Chat" never see the inbox. Run a chat-only team or a CRM-only team if that fits your structure.

Lead attribution

Each chat-derived lead carries the full marketing capture: UTM source / medium / campaign, referrer, landing URL, source domain, geo, device, browser, returning-visitor flag. Available in the User analysis report and in the lead's own row.

Always-on first responder

AI Agent

An optional virtual agent that picks up chats when humans aren't available — and quietly gets out of the way the moment a human is.

Choose your AI provider

Pick from Anthropic Claude (recommended) or OpenAI. Choose a model from a clean dropdown list or type a custom model ID. Set how many tokens the AI can use per reply. Everything is done from a friendly admin screen — no JSON editing, no config files.

Anthropic ClaudeOpenAICustom model id

Different AI personality per project

Every project has its own system prompt. So your e-commerce store can have a friendly shopping assistant, your medical practice can have a careful health helper, and your finance brand can have a formal advisor — all from a single install, with one prompt box per project.

Feels like a real human typing

The AI doesn't reply instantly — that would give it away. Instead, the moment a visitor sends a message, the AI waits 1 second, then shows the "Support is typing…" pulsing pill bubble in the visitor's chat. While the pill is showing, the AI is actually composing its reply. The reply lands and the typing pill disappears — just like a real agent who read the message, thought for a moment, and started typing back.

1-second pause before typing pillPulsing typing indicatorNatural reply timing

Auto green ticks — just like a human

Real agents open the chat, read the message, then reply. Visitor sees two green ticks (read receipt) right before the reply lands. The AI copies this exactly — the moment it starts composing, the visitor's previous unread messages flip to green double-tick "read" status. So when the AI reply finally arrives, the visitor sees the same natural sequence: delivered → read → reply. No "single tick still unread" tell that screams "bot".

Single tick → double green tickBefore the AI reply, not after

AI knows when to call a human

The AI can decide that a question is too complex, too sensitive, or out of scope — and explicitly say "I need to bring in a teammate". When that happens, the chat immediately moves to the human queue and is permanently flagged so the bot never tries to take it back, even if all humans go offline. The visitor gets the same smooth handover they'd get from one human to another.

Safety net — falls back to a human if AI fails

If the AI provider has a problem — rate limit, auth error, network blip — the chat transparently moves to a human agent per your configured rules. Visitor doesn't see an error; they just notice that "support" is replying instead of the bot. Zero customer pain even when the AI provider is having a bad day.

One master switch

Single toggle in the AI Integration page turns the entire AI agent on or off. When off, every chat routes to live agents exactly as if the AI never existed. Useful for testing, or for putting the AI on pause without losing your configuration.

AI in the QA agent filter

The Quality Analyst console treats the AI like any other agent — you can filter chats by "AI Agent" in the agent dropdown. QA can then audit AI-handled chats separately and grade them, helping you continually improve the system prompt.

Under the hood

Real-time Engine

Built on SignalR. Every event — new message, typing, receipts, presence, inbox change, AI status — flows through the same hub with strict per-side authorisation.

SignalR hub

A single ChatHub serves visitor, agent, admin, and QA connections. Each role joins its own group with strict authorisation on every method.

Authorized JoinConversation

Visitors can only join their own conversation group. Agents only chats assigned to them. Admins / QA — any conversation. Silent denial if unauthorised.

Auto-reconnect

Network blips reconnect automatically. After reconnect the client re-joins the currently-open conversation group so live ticks & typing resume without refresh.

Multi-window aware

An agent can open multiple tabs — presence aggregates by connection set. Closing one tab doesn't mark them offline.

Targeted broadcasts

Receipts, typing, and message events broadcast to only the relevant groups — not every connected client. Keeps the wire small.

Backplane ready

Single-instance today, designed to extend with Redis backplane when horizontally scaling.

WhatsApp-grade UX

Receipts & Typing Indicators

Both directions. Real-time. No refresh.

Sent

Message persisted server-side. Single tick on the sender's bubble.

Delivered

Recipient's connection received the SignalR broadcast. Double tick, grey.

Read

Recipient's tab is focused and they opened the chat. Double tick, green.

Typing indicators

Visitor → agent

The agent sees a pulsing "Visitor is typing…" pill above their composer the second the visitor starts typing. Auto-clears 5s after the last keystroke as a safety net.

Agent → visitor (incl. AI)

The visitor sees "Support is typing…" while the agent (or AI) composes. AI adds a deliberate 1-second pre-roll so it feels like a human reading first.

Never miss a chat

Notifications & PWA

System-level notifications, sound, app badges, and service-worker delivery — even when the dashboard tab is backgrounded or the window is minimized.

Browser

System notifications

Permission prompt is gated on a user gesture (iOS / Safari requirement — auto-requested calls are silently denied). Clicking a notification focuses the agent window and opens the right chat.

Service Worker

Background delivery

Dedicated /sw.js handles showNotification + click-handling even when the tab is fully backgrounded — critical for installed PWAs on iOS 16.4+.

PWA

App-icon badge

Installed PWA shows a numeric badge on the home-screen icon / dock / taskbar when there are unread chats. Cleared automatically when unread hits zero.

Sound

Audio ping

Web Audio API sine sweep on new message / new assignment. Plays only when the page is focused (browsers throttle audio in background tabs).

Notification triggers: new visitor message in an assigned chat · new chat assignment from the queue · new visitor message in the currently-open chat IF the agent's tab is backgrounded (otherwise the visible bubble is enough).

Know your numbers

Reports & Analytics

Server-paginated, server-filtered, CSV-exportable. Works the same whether you have 100 conversations or 100,000.

Chats

Top-level conversation list with All / Assigned / Unassigned / Closed / Offline pill filters, lead-stage filter, agent filter, date range, and full-text name/number search.

Chat data

Per-conversation detail report with Token, Name, Phone, Email, Project, Status, Agent, Lead stage, Updated (= last message time), and a one-click PDF transcript download.

PDF exportCSV export

User data (Marketing)

Visitor identification — name, phone, email, project, source domain, referrer, UTM source / medium / campaign, first-seen time.

User analysis (Marketing)

Deep visitor analytics — device type, OS & version, brand, browser, language, connection type, geo city / region / country, time on page, returning-visitor flag, full UTM trail.

Reports

Per-agent productivity, lead-conversion funnel, response-time stats. Date-range filtered.

CRM reports

All leads, Web leads, Uploaded leads, Google Sheet leads — each paginated, agent-assignable, stage-filterable, CSV-exportable.

One install, many brands

Multi-Project Architecture

A single install hosts unlimited projects. Each project is a fully isolated, branded chat experience.

Per-project branding

Title, subtitle, accent colour, welcome message, contact-form heading + description, project icon — all editable from the admin Projects tab.

Slug-based URLs

Each project lives at /p/{slug} with its own canonical URL. Customer cookies are scoped per-project — no token leakage.

CORS allowed origins

Each project carries its own whitelist of customer domains that can embed the widget. Wildcard subdomain support: https://*.client.com.

Per-project AI instructions

Each project carries its own AI system prompt — e-commerce, healthcare, finance projects get completely different bot behaviour.

Embed code generators

Standard widget embed, trigger-link embed, and auto-popup modal embed — pre-rendered in the admin Projects tab, copy-paste ready.

Default project fallback

One project is marked default — used when a widget loads without a slug, when a visitor lands on /, or when an embed leaves data-project off the script tag.

Plays well with others

Integrations & Ready-Made Embeds

Copy-paste ready scripts for capturing leads from any website — without touching that website's code. Plus first-class connectors for the tools you already use.

Web Lead — Single-Line Script

Capture leads from any website’s own form

You have a landing page with its own contact form. You want every form submission to flow straight into this CRM as a fresh lead — with the page URL, UTM parameters, and the visitor's input automatically attached. Add a single line of JavaScript to your landing page (no plugin, no build step, no tag manager needed):

<script src="https://yourdomain/widget/lead-form.js"
        data-project="default"></script>

That's it. The script automatically finds <form> elements on the page, hooks their submit, posts name + number + email + the form's source URL + any UTM params to our public /api/web-leads endpoint and the lead lands in the admin Web leads table within seconds. Original form behaviour (success redirect, thank-you page) stays untouched.

One line of codeAuto UTM captureSource URL capturedNo CRM logic on page

Google Sheets — One-Click Connect

Pull leads from any Google Sheet, no OAuth pain

Your team enters leads in a Google Sheet. Marketing has a lead capture form going into a Google Sheet. A partner shares leads via a Google Sheet. All of them flow into the same CRM, automatically:

Share the sheet with our service-account email (one-time, takes 10 seconds).
Paste the sheet's URL or ID in the admin's Google Sheets tab.
Pick the columns — which one is name, which is phone, which is email.
Done. The background poller checks the sheet every 60 seconds and pulls new rows in.

No per-user OAuth dance. No "log in with Google" buttons. No expired tokens to re-authorise. You can connect 1 sheet or 50 sheets — each one is configured independently with its own poll interval and column mapping.

Service-account authNo expired tokens60-second syncColumn-mapping UI

Standard widget embed

One async script tag adds the chat bubble to any website — with the project's branding, allowed embed domains, and full feature set.

Trigger link embed

Any link or button with data-cp-open opens the chat panel when clicked. Useful for "Chat with us" buttons in headers/footers.

Popup modal embed

Opens chat in a centered iframe popup on page load. Throttled once per 7 hours per browser. Perfect for re-engagement campaigns.

Marketing Tags

Paste GTM, GA4, Meta Pixel, or Hotjar code into the admin Marketing Tags textareas — injects into every page automatically. Zero code change.

Anthropic Claude

First-party AI provider. Recommended default for best quality replies. Custom model ID support.

OpenAI

Alternative AI provider. Same admin UI — just flip the provider toggle. Useful if you already have an OpenAI subscription.

CSV upload

Bulk-import leads from spreadsheets or external systems. Admin uploads a CSV — rows land in the Uploaded leads CRM table.

Installable agent app

Agents install the dashboard as a PWA on their phone or laptop. Looks and feels like a native app — with system notifications, sounds, and app-icon unread badges.

Installable visitor app

Visitors can install the chat page as a PWA on their phone. Useful for healthcare, education, or finance projects where customers come back regularly.

No developer needed

Marketing Implementation Without Coding

Add Google Tag Manager, Meta Pixel, GA4, Hotjar, or any tracking script — without touching a single line of code. Just paste, save, and you're done.

Admin → Marketing → Marketing tags page

Two big textareas labelled "Head snippet" and "Body snippet", a Save tags button, and clear help text explaining what each snippet does.

screenshots/marketing-tags.png

Head snippet box

Paste the script that's supposed to go inside the page <head> tag. This is where Google Tag Manager's container code goes, GA4's base gtag script goes, Meta Pixel's fbq base code goes, and Hotjar's tracker goes. The platform automatically injects whatever you paste into every page served — visitor chat, agent console, admin, manager, QA, login — all of them.

Google Tag ManagerGA4Meta PixelHotjar

Body snippet box

Paste the script that's supposed to sit immediately after the opening <body> tag. This is where GTM's <noscript> iframe goes, and where any chat scripts that must be in body go. Same auto-inject behaviour — every page on the platform gets it.

Saved instantly — live on next page load

Click "Save tags" and the snippets go live straight away. Anyone who visits the platform after you save sees the new tracking. No re-deploy, no restart, no developer involved.

Customer site widget stays clean

The embeddable widget that lives on third-party customer websites uses Shadow DOM, so the marketing tags pasted here only affect pages served by this platform. Your customer's site is never polluted by your trackers.

Fire events to Meta & GA4 when leads qualify

Meta Pixel

Lead-stage events flow to Facebook

When an agent or admin marks a chat as Qualified, Not Qualified, or Converted, the platform fires a DOM event the page can pick up. In Google Tag Manager, set a trigger on that event and fire a Meta Pixel Lead / CompleteRegistration / Purchase event to Facebook — with the visitor's hashed email and phone for advanced matching, and the project / agent / campaign / UTM as custom parameters.

// Example GTM trigger payload
fbq('track', 'Lead', {
  content_name: 'qualified',
  value: 0,
  currency: 'INR',
  fbc: '...',
  em: hashedEmail,
  ph: hashedPhone
});

Google Analytics 4

GA4 conversion events

Same pattern for GA4: the lead-stage change fires a custom event you can map in GTM to a GA4 generate_lead or conversion event. UTM parameters from the original chat-start are sent along, so your Acquisition reports actually attribute the qualified lead back to the correct campaign / source / medium.

gtag('event', 'generate_lead', {
  currency: 'INR', value: 0,
  source: utmSource, medium: utmMedium,
  campaign: utmCampaign
});

The full marketing loop, closed

Visitor lands from a Google Ad → UTM auto-captured at chat start → visitor becomes a lead → agent qualifies them → Meta Pixel / GA4 receive a Qualified Lead event with the original UTM. Your ad-platform optimisation gets the real signal (qualified lead, not just form fill), so ad budget flows to the campaigns that actually convert. All without a single line of code on your end.

The killer use case

Run Meta & Google Ads → Project URL → Lead + Live Chat + Campaign Optimisation

The biggest reason businesses choose this platform. Point your ads directly to your project URL — https://yourdomain.com/p/{slug} — and the same page that captures the lead also starts the live conversation, qualifies the lead, and fires events back to Meta and Google so your campaigns auto-optimise. No separate landing page. No lost time. No leaky funnel.

Step 1 · Ad click

Customer sees your Meta or Google ad and clicks. Landing-page URL on the ad = your project URL /p/your-slug. No separate landing page needed — the chat page itself is the landing page.

Step 2 · Lead saved

Customer lands on your branded chat page. UTM source / medium / campaign auto-captured. They fill name + phone + email and tap "Start chat". Lead saved instantly into your CRM with full marketing attribution.

Step 3 · Live chat starts

Same second, an agent (or AI) picks up the chat. Customer is talking to support before they even close the tab. No "wait for someone to call you back". Conversion rate jumps dramatically because customer never goes cold.

Step 4 · Campaigns optimise

Agent marks the lead Qualified or Converted. Event fires to Meta Pixel and GA4 with the original UTM attached. Meta and Google use this signal to find more users like the one who converted. Your ad budget automatically flows to the campaigns that actually close deals.

Why this single feature changes everything

Other CRMs make you build a landing page in WordPress, hook a contact form to a webhook, email the lead to a sales rep, who calls back 4 hours later when the customer has already bought from a competitor. This platform collapses that entire chain into one URL. Click → lead → live chat → close — in the same browser session, in under 60 seconds, with marketing attribution preserved end-to-end.

Cost saving

One URL replaces the whole stack

No landing-page builder subscription. No third-party form service. No CRM webhook integrator. No email-to-sales automation tool. The ad URL goes straight to your project page and every piece runs in-house from a single platform.

Higher conversion

Customer doesn’t go cold

Industry data shows lead conversion drops 10x if first contact takes longer than 5 minutes. With ad → chat in the same session, you reply before the user closes the tab. Conversion rate jumps because nobody has time to forget about you, get distracted, or buy elsewhere.

Smarter ad spend

Meta & Google learn from the right signal

Form-fill events tell Meta "this user filled a form" — useful but weak. "This user became a qualified lead" is the real signal. When your agent marks a chat Qualified or Converted, that event flows back to Meta Pixel and GA4 with the original UTM. The ad platform uses this to find more high-quality lookalike users and stop wasting budget on cold clicks.

Marketing attribution

Closed-loop attribution — finally

UTM source, medium, campaign captured at chat start. Stays attached to the lead all the way through Qualified, Converted, and revenue stages. You finally know which Meta ad campaign, which Google keyword, which Instagram reel actually generated paying customers — not just clicks.

24/7 readiness

AI catches the ad clicks while humans sleep

Ads run 24/7 but your humans don’t. The AI agent picks up chats during nights and weekends, qualifies the visitor, and parks the conversation for a human follow-up the next morning. Your ad spend works around the clock without paying for a night shift.

Multi-project ads

Different ads → different projects

Running ads for two brands? Healthcare and finance need totally different branding, different AI personality, different agent team. Easy: point Brand A’s ads to /p/brand-a and Brand B’s ads to /p/brand-b. Same install, same admin, two completely separated conversation streams.

Real-world example — ad to conversion in 3 minutes

0:00 — Meta serves a video ad targeting "small business owners in Mumbai".
0:08 — Customer taps the ad. Lands on https://yourbrand.com/p/loans?utm_source=meta&utm_campaign=mumbai-q1.
0:15 — Customer reads the welcome message, fills name / phone / email, taps "Start chat".
0:16 — Lead created. Round-robin assigns the chat to Rahul (online + capacity available).
0:18 — "Customer is typing..." pill shows in Rahul’s console. Rahul opens the chat.
0:20 — Customer’s first message arrives. Rahul replies instantly. Two-way live chat starts.
2:45 — Customer asks for loan amount and tenure. Rahul shares the rate slab.
2:55 — Customer says "OK book a callback for tomorrow 11am". Rahul marks lead Qualified.
2:55 — Meta Pixel receives a Lead event with UTM meta / mumbai-q1. Meta’s algorithm learns this campaign produces qualified leads, automatically increases delivery to similar users.
Next day — Rahul calls at 11am, closes the deal, marks lead Converted. Purchase event fires to Meta with the revenue value, attributing the sale back to the original Mumbai campaign.

Built for hostile internet

Security, DDoS Protection & Hacking Defence

Multiple layers of automatic protection. Brute-force login attempts get blocked. Scanners get blocked. Rate-limit abusers get blocked. Common web attacks are blocked by hardening headers. All of this runs automatically, with zero configuration.

Automatic IP-block system (DDoS & brute-force defence)

Failed-login auto-block

If anyone fails to log in 5 times within 10 minutes from the same IP, that IP is automatically blocked for the next 30 minutes. Stops brute-force / credential-stuffing attacks dead before they can guess the right password. Counters auto-reset after a quiet period so legitimate forgotten-password retries don't lock out real users.

5 strikes → 30 min blockRolling 10-min window

Scanner / enumeration auto-block

Bots that hammer the API looking for vulnerabilities — making unauth'd requests without a normal browser User-Agent — are detected and tracked. 50 suspicious requests in 10 minutes triggers a 60-minute IP block. Most vulnerability scanners give up and move on before they hit anything interesting.

50 strikes → 60 min blockUA-aware

Bad-upload auto-block

Visitors attempting to upload malicious files — wrong MIME type, forbidden extension, oversized payload — get tracked. 8 bad uploads in 10 minutes triggers a 2-hour block. Stops file-upload abuse cold without requiring a separate WAF.

8 strikes → 2 hour blockMIME + ext + size checks

Rate-limit auto-block

5 rate-limit-exceeded events from the same IP within 10 minutes triggers a 20-minute block. So an abuser who keeps hitting our request-per-minute limit eventually gets pushed away instead of constantly being throttled at the door.

5 strikes → 20 min block

First-in-pipeline IP check

The IP block check is the very first thing that runs on every request — before routing, before static files, before any controller. A blocked IP gets a flat 403 response with no body, no headers, no information leakage. Sub-microsecond lookup, so even under heavy traffic the check itself can't be a bottleneck.

Manual block (admin / manager)

Admin or manager can manually block a visitor IP from the chat modal in one click. The block is permanent (until manually unblocked). Used when you know a particular visitor is abusive and you don't want to wait for the auto-strike system to catch them. Audit-trailed with operator name and reason.

Reverse-proxy aware

Honours the X-Forwarded-For header from your reverse proxy (NGINX / Plesk / Cloudflare). The leftmost IP in the chain is the real client — so when Plesk forwards a request, we correctly track the original visitor's IP for blocking, not Plesk's local address.

Browser-level attack protection (hardening headers)

Clickjacking protection (X-Frame-Options)

Staff pages (/admin, /agent, /qa, /login) are blocked from being iframed by any other site. So an attacker can't trick a logged-in admin into clicking buttons on a hidden overlay. Visitor pages stay framable because the widget needs to iframe them.

Content-Security-Policy (XSS protection)

Staff pages run with a tight CSP — only scripts from our own origin and one trusted CDN (SignalR) can run. So even if someone manages to inject HTML, no malicious JavaScript can load and steal session cookies.

script-src self + signalr CDNframe-ancestors none

MIME-sniff protection

X-Content-Type-Options: nosniff stops browsers from guessing a file's type. So an attacker can't upload "evil.png" that the browser then "discovers" is actually executable HTML/JS.

Referrer privacy

Referrer-Policy: strict-origin-when-cross-origin stops your staff URLs leaking to external sites. So if an admin clicks a link in the staff console that goes to an external site, the external site doesn't learn the full admin URL the click came from.

Permissions-Policy lockdown

Camera, microphone, geolocation, accelerometer, magnetometer, payment, USB — all browser permissions are blocked by default on staff pages. A malicious third-party script can't silently request access to any of these. The widget runs in its own Shadow DOM so it's not affected.

Account & data security

BCrypt password hashing

Admin / manager / agent / QA passwords stored as BCrypt hashes — intentionally slow, salted per-user. Even if the database is stolen, cracking a single password takes years.

HttpOnly session cookies

Session tokens are HttpOnly, Secure on HTTPS, SameSite=Lax — no JS access, no CSRF replay across origins.

Role-aware endpoints

Every endpoint independently verifies role server-side. Hidden UI buttons are advisory only — manipulated requests fail server-side.

Login rate limiting

SecurityService records strikes against the source IP on every failed login; repeat offenders get blocked.

Visitor IP block

Abusive visitors get IP-banned in one click; future POST /api/conversations from that IP returns 403 silently. Audit-trailed with reason and operator.

Strict CORS

Per-project allowed-origins whitelist. The widget's API calls fail server-side if the origin isn't in the project's whitelist.

XSS-safe rendering

All chat text rendered via createTextNode + linkifier with strict URL whitelist (http / https only). javascript: / data: / vbscript: never enter the link path.

Production startup guards

In Production env, the app refuses to start if the connection string is empty or the default admin password equals the literal admin123. Crashes loudly — better than shipping with default creds.

Visitor data isolation

Project-scoped cookies + localStorage tokens. A visitor on project A can't see project B's conversation even from the same browser.

Built for the long-tail

Performance & Scale

Designed for organisations with lakhs (100,000+) of conversations and lakhs of messages.

Composite SQL indexes

Critical hot paths are covered by composite indexes — e.g. Messages(ConversationId, CreatedAt) makes "last 30 messages of a chat" an O(log n) seek even at lakhs of messages per conversation.

7x speedup measured

Server pagination everywhere

Chats, Chat data, User data, User analysis, CRM lead lists — all paginate server-side. Frontend never receives more rows than the page size.

Initial-page + infinite-scroll

Agent inbox initial load is 20 chats per bucket; older chats stream in on scroll. Chat history initial load is 30 messages; older messages stream on scroll-up. Same UX as WhatsApp / Telegram.

ExecuteUpdate for batch ops

"Release agent's chats on disconnect" runs as a single SQL UPDATE statement — no load-track-loop-save. Critical during mass-disconnect storms (server restart / shift change).

Stable pagination tiebreakers

Every message sort uses ORDER BY CreatedAt DESC, Id DESC so same-millisecond messages don't cause duplicate / skipped rows during pagination.

Light vs. heavy entity loaders

Read-only loaders skip Include(Messages) by default; detail views load only the last N messages. Single conversation lookups never accidentally hydrate the entire history.

Background sweepers

VisitorOfflineSweeper (15s threshold), AgentInactivitySweeper (5 min idle with open chats), AI fallback timer — all run as HostedServices with try/catch isolation.

Client-side cache hygiene

Staff browsers self-prune oversized localStorage IDs lists (cap at 5,000), drop orphan entries from previous users, and clear Cache Storage /api/* entries every 30 min.

The bottom line

Business Benefits

Why this changes how you run customer conversations.

Direct ads → lead → live chat → campaign auto-optimisation

Point Meta and Google ads straight to your project URL. The same page captures the lead and starts live chat — no separate landing page needed. Lead-stage events fire back to Meta Pixel and GA4 so your ad platform learns which campaigns produce qualified leads, not just clicks. Read the full flow →

Faster lead-to-conversion cycle

AI handles initial qualification 24/7; humans only pick up qualified or complex chats. Less time wasted, more deals closed.

Lower agent headcount per chat volume

AI fallback + smart routing + offline parking = the same team handles 3-5x the chat volume of a traditional setup.

Multi-brand on one infrastructure

Run unlimited projects from one install. Marketing agencies and multi-brand businesses save on licensing and ops.

Marketing attribution out-of-the-box

Every chat captures UTM, referrer, landing URL, device, geo — so you know exactly which campaigns drive which leads.

No SaaS lock-in

Self-hostable .NET application. Your data, your servers, your control. No per-seat pricing escalation as your team grows.

Mobile-first agents

Installable PWA + system notifications + app badges = agents handle chats from their phone without a separate app.

Enterprise-ready compliance

BCrypt hashing, IP blocking, audit-trail notes, server-side role enforcement, CORS whitelisting — meets the questions enterprise security teams ask.

Scales to enterprise volume

Engineered for lakhs of conversations and lakhs of messages — SQL Server composite indexes, infinite-scroll UX, server pagination throughout.

Customer-perceived quality

WhatsApp-grade receipts, typing pills, instant chat resumption — the bar your customers already know. They don't realise the chat lives on your server.

One-day deployment

Self-contained .NET publish, Plesk / IIS-ready, web.config pre-generated, idempotent schema patches. From zero to live in a single afternoon.

Under the hood

Technology Stack

Modern, proven, supported. No exotic dependencies.

ASP.NET Core 10

Web framework

SignalR

Real-time

EF Core 9

ORM

SQL Server

Database

Serilog

Logging

BCrypt.Net

Password hashing

QuestPDF

PDF generation

SkiaSharp

Image / rendering

Google Sheets API

Integration

Anthropic SDK

OpenAI API

Vanilla JS

Frontend

Shadow DOM

Widget isolation

Service Worker

PWA

Web Push API

Notifications

App Badging API

PWA badges

See it in action

Visual Tour

Captured directly from the running app. More screenshots can be added as screenshots/<name>.png files referenced from this section.

Login. Staff sign-in supports email or username + password. Failed logins are rate-limited per IP server-side.

Visitor page. Standalone branded chat at /p/default. Header colour, welcome heading, contact-form copy, and Start-chat button colour all come from per-project settings.

Another project, same backend. The same code path renders a different brand. Slug /p/rahul uses its own branding entirely.

Embeddable widget. The same chat behaviour delivered as a bubble + popup panel that any third-party site can embed with a one-line script tag.

Admin overview. Total chats by status, lead-stage breakdown, and a live agents-online grid with each agent's current load.

Admin chats. All conversations in one paginated table with status / lead-stage / agent / date filters and full-text search.

Ready to deploy?

OnlineChatCRM ships as a self-contained .NET application — drop on Plesk / IIS, fill in your DB connection string and admin credentials, and you're live in under an hour.

Deployment guide Back to top