Tukify — AI Shopping Assistant for WooCommerce

Tukify is an AI chatbot and shopping assistant for WooCommerce that works like a knowledgeable sales assistant who never sleeps. A shopper types a question in plain language - or uploads a photo - and Tukify finds the right products, answers honestly from your store's real data, recommends complementary items, and guides them to checkout, all inside the chat without leaving the page. It is a real AI shopping assistant, not a scripted FAQ bot. It understands intent with conversational search, replies with product cards (image, price, stock, add to cart), and only ever recommends real, in-stock products from your own catalog - so it never invents items or prices. It also steps in on its own: when a shopper stalls on the cart or checkout page, Tukify offers help with the things that usually hold a sale up. Tukify is bring-your-own-key: connect your own AI provider - Google Gemini, OpenAI (ChatGPT), Anthropic (Claude), xAI (Grok), or Groq (fast inference) - and Tukify never pays for inference. Every AI request runs server-side in PHP, so your API keys are never exposed to the browser, and your product embeddings stay in your own site database, keeping your catalog data on your server. What Tukify does for shoppers

• AI chatbot for WooCommerce - a natural, conversational way to shop, available around the clock.
• Conversational search - matches by meaning, not just keywords ("warm clothes for winter" surfaces jackets and sweaters).
• Product recommendations - cart-aware upsell suggests genuinely complementary items based on what is already in the cart.
• Visual / image search - a shopper uploads a photo and Tukify finds similar products in your catalog, great for "find me something like this". A photo of a room or an outfit is broken into its individual items and each one is searched separately.
• In-chat product detail popup - tap a product to see full details, an image gallery, variations (size, colour) and add to cart, all without leaving the chat.
• Add to cart in chat - a quantity stepper and add-to-cart on every product card and in the detail popup, plus an optional guided in-chat checkout.
• Order status - a secure order-status lookup, verified by order number and billing email, right in the conversation.
• Clarifying questions - asks one short follow-up (with tappable quick replies) when a request is too vague, instead of dumping mediocre results.
• Natural-language filters - understands price, colour, size, brand, stock, category and sorting ("cheapest wireless headphones under $200") and applies them as real WooCommerce queries.
• Comparison mode - a side-by-side comparison of 2-3 products with a one-line trade-off summary.
• Policy and FAQ answers - answers shipping, returns and warranty questions from your own pages and a custom Q&A knowledge base.
• Abandonment help - if a shopper stalls on the cart or checkout page, or looks about to leave, Tukify offers a hand with the things that usually hold people up: shipping cost, sizing, and payment. Dismissible, and never more than once per visit.

What Tukify gives store owners

• Bring your own API key - use Google Gemini, OpenAI / ChatGPT, Anthropic Claude, xAI Grok, or Groq. You pay your provider directly; Tukify never bills for inference.
• Choice of models - pick a chat provider and an embedding provider independently, and choose the model for each.
• Background indexing - embeds your catalog in the background via Action Scheduler, and only re-embeds products that changed.
• Presentation surfaces - a global floating chat widget plus three Elementor widgets (Chat, Search, Recommendations), each rendered in a Shadow DOM so your theme's CSS cannot break them.
• Conversion analytics - cart recovery rate, resolution rate, chat-assisted average order value against your store-wide average, and attributed revenue, over any date range. Plus top queries, zero-result queries, and unmet-demand insights.
• Knows a problem from a purchase - when a shopper sends a photo of a damaged or wrong item, Tukify stops selling and offers the order lookup and your returns policy instead.
• Privacy-first - all AI calls are server-side, keys are masked in the admin, and your catalog data stays in your own database. Analytics are aggregate counts only.

Coming soon: a WhatsApp channel, so the same AI shopping assistant can answer shoppers on WhatsApp through Meta's WhatsApp Business Cloud API. It is previewed in Settings but is not active in this release - nothing is sent or received yet. Groq (fast inference) Tukify now supports Groq as a chat provider - the fast-inference platform at groq.com (this is not xAI's Grok). Groq returns replies very quickly and has a generous free tier, which makes the chat feel close to instant. Groq is chat only, so embeddings keep using Gemini or OpenAI, and image search falls back to a vision-capable provider. Which provider does what

• Google Gemini - chat, embeddings and image (vision) search. Recommended default; has a free tier.
• OpenAI (ChatGPT) - chat, embeddings and image (vision) search.
• Anthropic (Claude) - chat and image (vision) search. No embeddings - use Gemini or OpenAI as the embedding provider.
• xAI (Grok) - chat and image (vision) search. No embeddings - use Gemini or OpenAI as the embedding provider.
• Groq (fast inference) - chat only, very fast, with a free tier. No embeddings or vision - use Gemini or OpenAI as the embedding provider, and image search falls back to a vision-capable provider.

You choose a chat provider and an embedding provider independently in Settings. Retrieval (RAG) always uses the embedding provider. If you change the embedding provider or model, Tukify prompts you to reindex, because embeddings from different models are not compatible and are never mixed in the same index. Requirements

• WooCommerce (active).
• An API key for at least one supported AI provider: Google Gemini, OpenAI, Anthropic (Claude), xAI (Grok), or Groq (fast inference). Gemini and Groq offer free tiers that are enough for development and small stores.
• If you pick Claude, Grok, or Groq for chat, you also need a Gemini or OpenAI key for embeddings (those providers have no embeddings endpoint).
• PHP 7.4+.
• Elementor is optional (only needed for the Elementor widgets).

1.6.0 Three new features in this release: a proactive abandonment intervention, conversion analytics, and smarter image search. Abandonment intervention

• The chat now opens itself with a helpful nudge when a shopper stalls on the cart or checkout page, or looks about to leave.
• Three triggers - checkout idle, cart idle, and exit intent - each switchable, with editable idle thresholds and message templates.
• The nudge offers quick-reply chips for the most common blockers: shipping cost, sizing, and payment. Each is answered from your live cart, shipping zones, and payment gateways plus your knowledge base, with no AI call.
• One frequency cap covers every trigger, so a shopper is nudged at most once per session and at most once per N hours.
• Never fires for logged-in administrators or shop managers, so it stays out of your way while you work on your own store.
• Replaces the separate "Exit intent" and "Proactive re-engagement" settings with a single Engagement tab. Your existing settings are migrated automatically.

Conversion analytics

• New KPIs on the dashboard: cart recovery rate, resolution rate, chat-assisted average order value against your store-wide average, and attributed revenue.
• New date-range filter - 7, 30, or 90 days, 1 year, or all time - and a trend line of attributed orders per day.
• Chat-to-sale is now range-aware and counts distinct sessions on both sides of the ratio. It previously compared every attributed order against chat sessions only, which could report over 100%.
• Order totals are recorded at attribution time, so revenue metrics need no per-order lookups when the dashboard loads.
• Aggregate counts only. No customer, order contents, or personal data is stored or displayed.

Smarter image search

• Uploaded images are now understood before they are searched. A quick classification decides whether the photo shows a single product, a scene, a problem with an item, or something unclear.
• A photo of a damaged, faulty, or wrong item no longer returns products to buy. Tukify offers the secure order lookup and your own returns policy instead.
• Scene photos - a room, an outfit, a flatlay - detect the main items and search for each one, now covering furniture, lighting, decor, textiles, appliances, kitchenware, and plants as well as clothing.
• An image Tukify cannot read now asks one short question instead of guessing.
• Simple product photos take exactly the same path as before, and any classification failure falls back to it, so image search keeps working even if the extra step is unavailable.

1.5.0

• New: in-chat product detail popup (gallery, variations, add to cart).
• New: Groq (fast inference) chat provider.
• Coming soon: WhatsApp channel (preview in settings).
• Improved plugin listing and documentation.

1.4.6

• Checkout button now always appears at the bottom of the chat whenever the cart has items, in both new and existing chats (was inconsistent before).
• Added a subtle, dismissible review request on the plugin's own screens.

1.4.5

• Improved the readme and plugin listing (search keywords, clearer descriptions, multi-provider wording).

1.4.4

• New Tukify logo (transparent SVG) shown in the chat launcher, avatar, and admin header - it renders cleanly inside the round chat bubble, with no square corners.
• Added an Appearance setting to customize the chat bubble background colour, with a live preview and a low-contrast hint.

1.4.3

• Fixed prices showing as raw HTML entities (e.g. "$76.00" instead of "$76.00") in the chat checkout and order-status cards. Currency output is now decoded to plain text everywhere prices appear in the chat.

1.4.2

• Corrected plugin header metadata for the WordPress.org review: author name and distinct plugin/author URIs.
• Removed plugin-wide suppression of other plugins' admin notices; notices are now scoped to Tukify's own screens.

1.4.1

• Performance: added an (event_type, created_at) index for analytics range queries and a daily purge so the events table stays bounded.
• Performance: the semantic-search scan now reads embeddings in batches to cap peak memory on larger catalogs.

1.4.0

• Added API usage tracking with a per-day tokens/requests chart and estimated cost in Logs / Analytics.
• Added response caching (query embeddings + knowledge-base answers) with a TTL, enable/disable, and a clear button; nothing user-personal is ever cached.

1.3.0

• Added back-in-stock notifications: shoppers can ask to be emailed when an out-of-stock product returns, with a consent step, an admin list, an email template, and one-click unsubscribe.

1.2.0

• Added a quantity stepper on chat product cards.
• Added site-wide RAG over posts, pages, and products, with source citations under answers.
• Added secure order-status lookup, a size and fit advisor, "shop the look" multi-item visual search, demand insights, proactive re-engagement, and an opt-in in-chat checkout.
• Redesigned the admin dashboard.

1.1.0

• Added clarifying questions, natural-language filters, comparison mode, policy/FAQ knowledge base, cart-aware upsell, and exit-intent re-engagement.
• Added image-based (visual) search.
• Fixed cart accumulation when adding to cart from the chat widget.
• WooCommerce HPOS compatibility declared.

1.0.0

• Initial release: semantic search, RAG chat, background indexing, cart integration, analytics dashboard, floating widget and Elementor widgets.