EasyIT AI Chat — Free AI Chatbot for WooCommerce & Business Sites

[eaic_chat] — basic chat
[eaic_chat provider="gemini" title="Support Bot" height="500"] — choose a provider
[eaic_chat profile="support-bot"] — load a saved bot profile

Changelog

2.12.1

• Improvement: The readme’s Installation steps now start with the two-minute setup wizard and list the optional next steps (spending limit, forms and hours, quick-answer buttons). No code changes.

2.12.0

• New: Enquiry and booking forms (EasyIT AI Chat → Forms & Hours). Up to five forms; choose from name, email, phone, preferred date and time, number of people and message (each off, optional or required) and add up to three questions of your own (free text or a list of choices). Templates for an enquiry and a booking are included.
• New: Forms appear as buttons above the chat input, can be the target of the «Talk to a person» button (Settings → General → Contact method → «One of my forms»), and can be placed on any page with the [eaic_form id="…"] shortcode.
• New: Requests screen (EasyIT AI Chat → Requests) with a new-request counter in the menu, status tabs (new, confirmed, done, cancelled), search, form and date filters, a detail view with private notes and a reply-by-email link, link to the related conversation, delete, and CSV export.
• New: Email notification for each request (to a global address or one per form, at most 30 an hour), optional confirmation email to the visitor, optional forwarding to your webhook (same URL and signature as chat events), and an eaic_request_created action for developers. The weekly summary mentions requests waiting for you.
• New: Business hours & offline mode — weekly opening hours (one interval per day) and closed dates in your site time zone, DST-aware. Outside hours the chat shows an offline notice with the next opening time and a button for your first form; the AI can optionally be switched off while closed (quick-question buttons and forms keep working); bookings can be limited to open hours. The notice is computed in the browser, so it is correct on cached pages and disappears at opening time.
• New: The AI is told your opening hours and whether you are open right now, so «are you open?» gets a true answer.
• New: Spam protection on forms — hidden trap field, speed check, 20 attempts and 5 accepted requests per address per 10 minutes, link-spam and banned-word filters, IP blocklist.
• New: Requests are included in WordPress’s Export / Erase Personal Data tools, covered by a retention setting (default 365 days) and removed on uninstall.
• Improvement: Form labels and chat buttons follow the Chat Language setting (Bengali included).

2.11.0

• New: Conversations inbox (EasyIT AI Chat → Conversations) — search, filter by rating or visitor type, read full transcripts with sources and ratings, export the list as CSV, download or delete a conversation. Administrators only.
• New: Quick-question buttons (Settings → General) — up to 8 of your own questions and answers shown at the start of a chat; answered instantly with no AI call.
• New: Proactive greeting (Settings → General) — optional bubble next to the floating chat button after a delay, shown once per visit on every page, the home page, or pages you pick.
• New: Page awareness (Settings → General, off by default) — the chat is told which published page or WooCommerce product the visitor is viewing.
• Improvement: Faster, lighter knowledge search. Vectors are stored in a compact format (about 4× smaller), and the index is scanned in small batches, so memory use stays flat and a question over 3,000 passages takes about 0.1 s. The Knowledge Base page shows how many passages are indexed and how long the last search took. Existing indexes keep working.
• Improvement: OpenAI embeddings now use 512 dimensions (cheaper to store and compare). If you use OpenAI for semantic search, re-index once; until then keyword search is used.
• Improvement: Keyword search now splits on punctuation, so «policy?» matches «policy» (also helps non-English text).
• Improvement: «How many…» counting now looks only at uploaded documents; it no longer counts words across your site’s posts.
• Improvement: If the Ollama embedding server is down, chat skips it for five minutes instead of waiting on it for every message.
• Improvement: The Knowledge Base page no longer loads every stored vector just to count them.
• Improvement: Deleting a conversation also removes its ratings.
• Fix: Clicking a suggestion, quick question or lead-card button in the floating chat panel could close the panel.
• Fix: The readme performance FAQ now states the real asset size and when the chat’s CSS and JavaScript load.
• Listing: new title, short description and tags describing what the plugin is for; shorter, benefit-first «Why EasyIT AI Chat?» section; new FAQ answers.

2.10.1

• Fix: The readme Description was over WordPress.org’s 2,500-word limit and was being truncated. It is now condensed; no features were removed.

2.10.0

• New: «AI Chat» Gutenberg block (provider, bot profile, title, placeholder, height; WooCommerce bots available as bot types when WooCommerce is active).
• New: Elementor widget with the same options.
• Fix: Plugin Check findings in 2.9.0 — ordered translation placeholders, escaping/prepare annotations, and the leads CSV export no longer uses fwrite().
• Fix: «Tested up to» updated to WordPress 7.1.

2.9.0

• New: Setup wizard (EasyIT AI Chat → Getting Started → Start the setup wizard, and linked from the setup notice). Connects a provider after testing the key with a real request, saves your business details, and indexes your site — in about two minutes.
• New: Business Facts (Settings → General) — hours, phone, email, address and free-text policies are sent to the AI as authoritative facts.
• New: Human handoff (Settings → General) — a «Talk to a person» button (WhatsApp, phone, email or link) shown when the knowledge base had no match, after a «not helpful» rating, or under every answer if you prefer.
• New: Chat health alerts — provider failures are recorded; after two in a row an admin notice explains the problem in plain language (wrong key, out of credit, wrong model, unreachable Ollama…) with links to fix it, and you can receive one email a day until it is resolved. Administrators also see this advice in the chat itself; visitors still see only a generic message.
• New: Answer cache (opt-in, Settings → General) — identical first questions are answered from a saved copy with no AI call. Follow-ups and uploads are never cached.
• New: Budget caps apply to the WooCommerce bots and their usage is now tracked. Administrator-only tools such as «Ask Your Store» are exempt.
• New: Privacy tools — chat conversations (for logged-in users) and leads are included in WordPress’s Export and Erase Personal Data tools, and a suggested paragraph is added to the Privacy Policy guide.
• New: Bengali chat labels — Settings → General → Chat Language.
• Improvement: If a streamed reply fails before any text arrives, the chat retries once the ordinary way instead of showing an error.
• Improvement: With Claude or DeepSeek as the chat provider, the Knowledge Base can use an OpenAI or Gemini key (if one is entered) for semantic search.

2.8.0

• New: Source citations — when an answer comes from your site content or Knowledge Base, the chat shows clickable «Sources» under it. Only published, public pages are ever linked.
• New: Usage & cost tracking — token usage and estimated cost per provider/model on the Analytics page, using the token counts the AI provider reports (estimated when a provider doesn’t report them).
• New: Budget caps — daily token, monthly token and monthly cost limits. When one is reached the chat pauses with a message you choose and you receive a single email. Custom model prices can be entered for Together AI and custom providers.
• New: Real streaming for OpenAI, Anthropic (Claude), Google Gemini, DeepSeek and custom OpenAI-compatible providers — replies appear as they are generated instead of all at once.
• New: Cloud embeddings — Ask This Site and the Knowledge Base can now use OpenAI or Gemini embeddings for semantic search, so no Ollama server is needed. Choose the service under Knowledge Base → Embedding Service (default: follows your chat provider). Changing the embedding service means re-indexing; the page tells you when that’s needed, and keyword search is used in the meantime.
• New: Knowledge gaps report on the Analytics page — questions with no match in your content, plus answers rated «not helpful».
• New: Lead capture in the main chat (opt-in), new-lead email notifications, and CSV export of leads.
• New: Weekly summary email (opt-in).
• Improvement: A short, one-time review request appears on this plugin’s own admin pages after the chat has been used for a while; it can be dismissed permanently.
• Improvement: If the Ollama embedding server is unreachable, chat falls back to keyword search immediately instead of waiting on a timeout for every message.
• Fix: Saving the Knowledge Base page no longer blanks your AI provider API keys or unticks settings that live on the Settings page.
• Fix: The non-streaming chat endpoint now uses the Knowledge Base like the streaming one.
• Removed: The bundled Freemius SDK (about 23 MB). It was no longer used. The plugin makes no calls except to the AI provider you choose.

2.7.0

• New: File & image uploads in chat — visitors can attach images and documents (.txt/.pdf) to a message. Images are analyzed by vision-capable providers (OpenAI, Anthropic, Gemini); documents are read as text and given to the AI as context, working with every provider. Enable globally in Settings → File & Image Uploads, or per bot profile.
• Fix: Plugin Check compliance — replaced discouraged unlink()/rename() calls with wp_delete_file() and a documented phpcs:ignore, and annotated already-validated $_POST/$_FILES reads on the new upload endpoint.

2.6.0

• New: Getting Started page (EasyIT AI Chat → Getting Started) — a live setup checklist (provider connected? floating widget on? content indexed?), a WooCommerce Bots section that explains what’s unlocked and links straight to installing WooCommerce if it’s missing, and a Security Snapshot summarizing your current rate limits, guest-chat, GDPR gate, and prompt-injection settings at a glance.
• Improvement: The WooCommerce Bots settings (Settings → UI) and the Shortcode Builder page now clearly flag when WooCommerce isn’t active instead of silently showing store-only options with no context — existing settings are never hidden or cleared, just annotated.
• Improvement: Settings, Analytics, Test Chat, and Shortcode Builder pages are now fluid-width and reflow responsively instead of being pinned to a fixed 900px column, so they make better use of wide screens; the sidebar is sticky on desktop.
• Fix: The admin stylesheet now cache-busts on its file modification time instead of the plugin version string, so CSS-only changes take effect on a normal reload instead of requiring a hard refresh.

2.5.0

• New: Ask This Site — enable auto-indexing in Knowledge Base settings and the AI automatically learns your published posts/pages (or any public content type you choose), using the same document search that powers the Knowledge Base (RAG) feature. No manual upload needed. Re-indexes on every update, de-indexes on unpublish/trash/delete. Includes a one-click «Reindex All Site Content Now» batch button for existing content.

2.4.1

• Fix: The default AI provider was «Ollama» pointing at localhost:11434 — on a fresh install with no self-hosted Ollama server (the vast majority of WordPress hosts), the chat silently failed with no indication of what to fix. Default provider is now OpenAI, which fails instead with an actionable «add your API key» message, and the Settings page now opens on whichever provider tab is actually configured as default.
• New: A dismissible «add your API key to get started» notice now appears in wp-admin until a provider is configured, linking straight to Settings.
• Fix: Errors from the AI provider (e.g. «No API key configured») were shown verbatim to every site visitor in the public chat widget, not just admins — this leaked internal configuration detail and looked broken to end users. Admins still see the real error; visitors now see a generic «temporarily unavailable» message.

2.4.0

• New: WooCommerce Bots — Order Status Bot ([eaic_order_chat]), Product Q&A Bot ([eaic_product_chat]), Return/Refund Bot ([eaic_return_chat]), and Smart Product Finder ([eaic_product_finder]) are now built into the free plugin. Activate automatically when WooCommerce is present.
• New: «Ask Your Store» — admin-only AI chat that answers questions about revenue, orders, top products, low stock, and new customers.
• New: Smart Floating Widget mode — the floating chat button can now auto-switch between the Order Bot, Product Bot, and general chat based on the page being viewed (Settings → UI → Floating Widget → Mode).
• New: Lead Capture — optional pre-chat name + email form on the WooCommerce bots; leads appear on the Analytics page alongside top-asked products and unanswered queries.
• Security: Return Bot now requires guests to verify their billing email before it will discuss an order (previously any visitor could look up any order number).
• Fix: «No-Storage Mode» (Settings → Security) was silently ignored on streaming chat — it now correctly stops conversations from being saved when enabled.
• Removed the «Upgrade to Pro» link — every feature that used to require EasyIT AI Chat Pro is now part of the free plugin.

2.3.4

• Fix: Plugin Check compliance — replaced move_uploaded_file() with wp_handle_upload(), escaped exception output, added phpcs:ignore for set_time_limit() and table-name DB queries, prefixed view variables with eaic_ namespace.

2.3.3

• Fix: Ollama now streams tokens in real time — first word appears within seconds instead of waiting for the full response to complete.

2.3.2

• Fix: RAG count queries («how many X») now use keyword-matched chunks as context, preventing wrong items from appearing when the embedding model returns poor results.
• Fix: Embedding timeout reduced from 60 s to 8 s so keyword fallback kicks in quickly when the embedding model is slow or misconfigured.
• New: Copy and Download buttons added to every AI response message — always visible, no hover required.
• Fix: Plugin version bump ensures browsers load updated JS/CSS (cache-bust).

2.3.1

• Fix: RAG document processing — per-chunk embedding wrapped in try/catch so one bad chunk no longer aborts the whole document.
• Fix: PHP set_time_limit extended during processing to prevent timeout on large PDFs.
• Fix: Stuck documents in «processing» state are reset to «error» on next admin page load so they can be re-processed.
• Fix: DB Setup button now reliably creates tables when RAG is first enabled.

2.3.0

• New: Knowledge Base (RAG) — upload PDF documents; the AI answers questions using only your document data (Retrieval-Augmented Generation).
• New: Semantic search via Ollama embeddings with automatic keyword fallback when embeddings are unavailable.
• New: Accurate «how many X» counting — scans all chunks with plural/singular stemming for exact results.
• New: Admin Test Query UI — inspect which document chunks are retrieved for any query.
• New: Document status auto-polling — admin page refreshes processing status without a manual reload.
• New: Configurable chunk size, chunk overlap, top-K results, similarity threshold, and embedding model.

2.1.1

• Fix: Custom provider settings now save reliably (JSON decode fallback prevents silent data loss on save).
• Fix: Settings page stays on the active tab after saving (sessionStorage-based tab restore).
• Fix: Custom provider slugs (custom_1, custom_2 …) are now accepted as the Default Provider.

2.1.0

• New: Custom Providers — add any OpenAI-compatible API endpoint (LM Studio, custom gateway, Ollama proxy, etc.) via Settings → Custom tab. Use with [eaic_chat provider="custom_1"]. Includes per-provider Test Connection button, Enable/Disable toggle, configurable URL, API key, model, and timeout.

2.0.1

• Bug fixes from code audit: fixed critical array key mismatch in feedback ownership check (feedback now correctly accepts/rejects based on actual session owner), fixed missing closing div in AI Avatar settings card (cosmetic HTML issue), improved no-storage mode (no longer creates orphan session records in DB), made frontend_i18n() a static method, updated Anthropic default model, fixed Tested up to header (6.8), improved copy button error handling in Shortcode Builder.

2.0.0

• Major release — Security Suite. Eight new security features in Settings → Security:
• Access Restriction — Allow everyone, logged-in users only, or specific user roles.
• IP Blocklist — Block specific IP addresses from using the chat.
• Word Filter — Block or warn when a message contains banned words/phrases (case-insensitive, one per line).
• Prompt Injection Detection — Detect and block common jailbreak/injection patterns automatically.
• No-Storage Mode — Opt out of saving conversations to the database entirely (GDPR-strict setups).
• Anti-Bot Math Captcha — Simple arithmetic challenge before the first message. No external API required.
• Abuse Alert Email — Get an email when the rate limit is exceeded. Configurable recipient.
• Message Length Limit — Configure max characters per user message (50–4000, default 4000).

1.0.31

• Added Webhook Support — configure a webhook URL in Settings → Webhook. After each AI response, the plugin sends a non-blocking POST to your URL with JSON payload: session_uuid, user_message, ai_response, provider, timestamp. Optional HMAC-SHA256 secret key adds an X-EAIC-Signature header for verification. Compatible with Zapier, Make (Integromat), n8n, and any HTTP endpoint.

1.0.30

• Added Multiple Bot Profiles — create named configurations (slug, name, provider, title, system prompt) in Settings → Profiles. Load any profile with [eaic_chat profile="your-slug"]. Profiles appear in the Shortcode Builder for one-click selection.

1.0.29

• Added Shortcode Builder — a new admin page (EasyIT AI Chat → Shortcode Builder) lets you configure the chatbot visually. Choose provider, title, placeholder, height, and system prompt with dropdowns and inputs. The shortcode updates live and a Copy button puts it on your clipboard instantly. No typing required.

1.0.28

• Added Context Length Control — a new «Context Window (messages)» setting (1–20, default 10) controls how many previous messages are sent to the AI with each request. Lower values reduce token cost; higher values give the AI more conversation memory. Find it in Settings → AI Behavior.

1.0.27

• Added Read Aloud (TTS) — a speaker button appears on hover over any AI message. Click to have the browser read the message aloud using the Web Speech API (SpeechSynthesis). Click again to stop. No external service required — uses the built-in browser engine.

1.0.26

• Added GDPR Consent Gate — optionally show a consent banner before the chat activates. Users must click an accept button before chatting. Consent is stored in a browser cookie (365 days). Configurable message text and button label. Enable in Settings → UI.

1.0.25

• Added Fullscreen Mode — an expand button in the chat topbar switches the widget to full-screen overlay. Press Escape or click again to exit. Works on any page without layout changes.

1.0.24

• Added Session Search — a search box in the chat sidebar lets users filter conversations by title in real time. No extra server requests — filtering is client-side.

1.0.23

• Added Message Feedback — 👍/👎 buttons appear below each AI response. Ratings are stored in your database and visible in the Analytics dashboard (Helpful / Not helpful counts). Feedback resets if you regenerate a response.

1.0.22

• Added Copy Message — hover any AI response to reveal a copy button. Copies the plain text of the message to the clipboard. Shows a checkmark confirmation on success.

1.0.21

• Added Message Timestamps — each chat message now shows the time it was sent (HH:MM format) next to the sender label. Timestamps are preserved when loading conversation history.

1.0.20

• Added Stop & Regenerate — the Send button turns into a red Stop button while the AI is responding. Click Stop to cancel the stream at any time. After a response completes, a Regenerate button appears below the last AI message to re-run the same query.

1.0.19

• Improved Typing Indicator — replaced the hourglass thinking bubble with a smooth 3-dot animated typing indicator (classic messaging-app style). Dots use the configured accent color and animate with a staggered bounce. Timer appears after 5 seconds for slow responses.

1.0.18

• Added Analytics Dashboard — new admin page (EasyIT AI Chat → Analytics) showing total conversations, total messages, messages today, active chats this week, most used provider, and a 7-day message bar chart. All data from your own database, zero external tracking.

1.0.17

• Added Floating Chat Widget — a fixed launcher button that appears on every page. Click to open a slide-up chat panel. Configurable position (bottom-right / bottom-left), label, and uses your accent color. Enable in Settings → UI. No WooCommerce required.

1.0.16

• Added Conversation Export — a download button in the chat topbar lets users save the current conversation as a .txt file. Enable in Settings → UI.

1.0.15

• Added Voice Input — enable a microphone button in the input area. Uses the browser’s Web Speech API (Chrome, Edge, Safari). Speech is transcribed directly into the text field. Enable in Settings → UI. Requires HTTPS.

1.0.14

• Added Color Customization — set Accent, User message, and AI message bubble colors via color pickers in Settings → UI. Changes apply instantly with a reset-to-default button for each color.

1.0.13

• Added Custom AI Avatar — upload any image from the WordPress Media Library to replace the default 🤖 emoji in all AI message bubbles. Configurable in Settings → UI.

1.0.12

• Added Suggested Questions — display clickable question chips below the welcome area. Clicking a chip sends it instantly. Up to 6 chips, one per line. Configurable in Settings → UI.

1.0.11

• Added Welcome Message — enable a custom AI greeting bubble that appears when a new chat session starts. Configurable in Settings → UI. Supports markdown. Not stored or sent to the AI.

1.0.10

• Added «Upgrade to Pro →» link in the plugins list for easy access to the Pro version.

1.0.9

• Fixed fatal error on activation — Freemius SDK vendor files now included in the plugin package.

1.0.8

• Renamed Freemius helper function to eaic_fs() for consistency with plugin prefix.

1.0.7

• WooCommerce bots moved to EasyIT AI Chat Pro add-on.
• Updated default models: GPT-4o-mini (OpenAI), Claude 3.5 Haiku (Anthropic), Gemini 2.0 Flash (Gemini).
• Added newer model examples in settings: GPT-4.1, o3, o4-mini, Claude 3.7 Sonnet, Gemini 2.5 Pro/Flash.

1.0.6

• Minor bug fixes and stability improvements.

1.0.5

• Minor bug fixes and stability improvements.

1.0.4

• Added Google Gemini provider (Gemini 1.5 Flash, Gemini 1.5 Pro, Gemini 2.0 Flash).
• Added auto-title generation — first message generates a meaningful session title via the active AI provider.
• Added data retention cron — sessions older than the configured number of days are purged automatically.
• Added per-IP rate limiting as a secondary hard cap alongside the existing per-user/session limit.
• Added Lock System Prompt setting to prevent front-end prompt injection on public sites.
• Improved guest cookie security: SameSite=Lax attribute now set via PHP 8.0 array signature.
• Rate limit window and max values are now configurable in Settings → Security.

1.0.3

• Renamed shortcode from [easyai] to [eaic_chat] to use the plugin’s eaic prefix (WordPress.org review feedback).

1.0.2

• Renamed plugin and folder to comply with WordPress.org trademark guidelines.
• All exception messages now escaped before being thrown.
• All direct database queries paired with object-cache reads/writes.
• All AJAX handlers verify nonce before reading $_POST.
• Removed deprecated load_plugin_textdomain() call (handled automatically since WP 4.6+).
• All view-scoped variables prefixed to avoid global namespace collisions.
• Excluded development files from production zip.

1.0.1

• Initial public release.