Description
Sparxio adds a fully customizable AI-powered chat widget to your WordPress site. Visitors chat with an AI assistant that replies in their own language, answers from your knowledge base, and guides them toward your goals — making a purchase, booking a demo, or submitting an enquiry.
You connect your own AI provider with your own API key, so you stay in full control of usage and cost. Nothing is routed through a third-party middleman.
Why Sparxio
- Bring your own AI key — no per-message markup, no forced subscription to use the chat
- Answers grounded in your own content, not generic replies
- Genuinely multilingual — the interface ships in 30 languages
- Built for WordPress: Shadow DOM isolation, mobile-optimised, no theme conflicts
Free Features
- 10 AI providers — Anthropic (Claude), OpenAI (GPT), Google Gemini, xAI (Grok), DeepSeek, Mistral, Groq, OpenRouter (hundreds of models), a local Ollama model, or any OpenAI-compatible API (Together, Fireworks, Cerebras, Perplexity, LM Studio…) — with your own key. Use any model ID, including ones released after this version
- Never goes silent — pick a backup AI provider that answers automatically if the main one fails. Requests adapt to each model’s options, and a retired model is replaced by a current one (you are told in AI Settings)
- Accurate answers — the bot only states addresses, numbers, prices and policies it was actually given; otherwise it says so instead of guessing
- OpenRouter model browser — search hundreds of live models, free ones first, with prices, plus up to three backup models
- Full branding control — colour presets or custom colours, header icon, brand name, and CTA button. No attribution is shown unless you add your own footer text.
- 30-language widget UI — including German, Spanish, French, Italian, Portuguese, Dutch, Polish, Swedish, Danish, Finnish, Norwegian, Greek, Czech, Hungarian, Romanian, Ukrainian, Russian, Arabic, Turkish, Hindi, Chinese, Japanese, Korean, and more
- Unlimited knowledge base — teach the AI from FAQ entries, documents, web pages, and sitemaps so answers reflect your business
- Lead capture & CSV export — collect name, email, and phone inside the chat, then view, manage, and export leads to CSV
- Chat logs — full transcript of every conversation with the visitor’s device and page (and country, when your site is behind Cloudflare)
- Basic dashboard — sessions, messages, and activity at a glance
- Live typing indicator — shown the moment a visitor sends a message, until the answer arrives
- Cost controls — history limits, rolling summarisation, and session timeouts keep API usage low
- Built-in setup guide — a live checklist walks you through configuration
Sparxio Pro (separate plugin)
Everything above is free and fully functional. For stores and teams that want more, Sparxio Pro is a separate plugin, sold on our website, that adds:
- WooCommerce chat — product search, add-to-cart, and verified order-status lookups inside the chat
- Live agent takeover — jump into any conversation in real time, then hand back to the AI
- Advanced analytics — conversations, leads and top questions for a chosen period, plus WooCommerce orders placed after a chat (paid orders only, minus refunds, per currency)
- Proactive engagement — greet visitors based on time on page, exit intent, or the page they are viewing
External Services
This plugin contacts the third-party services below only when you configure or choose them. Nothing is sent to any third party unless you add an AI provider key, or opt in to Freemius when activating the plugin.
AI providers. When a visitor sends a chat message, the plugin sends that message, the recent conversation history, your system prompt, and any matching knowledge-base excerpts to the AI provider you selected in AI Settings, using your own API key. Nothing is sent until you add a key and select a provider. Only the selected provider is contacted — plus the backup provider you may choose in AI Settings, and only when the selected one fails.
- Anthropic API (api.anthropic.com) — used when Anthropic is the selected or backup provider. Terms / Privacy
- OpenAI API (api.openai.com) — used when OpenAI is the selected or backup provider. Terms / Privacy
- Google Gemini API (generativelanguage.googleapis.com) — used when Gemini is the selected or backup provider. Terms / Privacy
- DeepSeek API (api.deepseek.com) — used when DeepSeek is the selected or backup provider. Terms / Privacy
- OpenRouter API (openrouter.ai) — used when OpenRouter is the selected or backup provider. OpenRouter forwards the request to the model you choose (and to your backup models, if set). Your site URL and site name are also sent as attribution headers. When OpenRouter is selected or has a key and an administrator opens AI Settings, the public model list (openrouter.ai/api/v1/models) is fetched, without a key or any site data, and cached for six hours. Terms / Privacy
- xAI API (api.x.ai) — used when xAI (Grok) is the selected or backup provider. Terms / Privacy
- Mistral AI API (api.mistral.ai) — used when Mistral is the selected or backup provider. Terms / Privacy
- Groq API (api.groq.com) — used when Groq is the selected or backup provider. Terms / Privacy
- Custom OpenAI-compatible endpoint — used when “Custom” is the selected or backup provider. Requests go only to the address you enter (for example Together, Fireworks, Cerebras, Perplexity, or your own LM Studio / vLLM server); that service’s own terms and privacy policy apply.
- Ollama (your own server, default http://localhost:11434) — used when Ollama is the selected or backup provider. Requests go only to the Ollama server address you enter; no third party is contacted.
Other services.
- Freemius (api.freemius.com) — only if you choose to opt in when activating the plugin (you can skip it). Sends your site URL, WordPress and plugin versions, and your admin name and email, so you can receive optional security and feature notices and manage a Pro license. Terms / Privacy
Knowledge base imports. When you import a web page or sitemap, your site fetches that address (and the pages a sitemap lists) directly. Only the address you enter is requested; downloads are size-limited and private/internal addresses are refused.
Cookies and browser storage. The widget keeps the conversation in this browser according to your Chat Persistence setting (memory only, this tab, or this browser). A sparxio_seen cookie remembers returning visitors. On WooCommerce stores with Sparxio Pro, a sparxio_attr cookie (a signed conversation number, not the conversation itself), set once the visitor has sent a message, lets an order placed later be counted as chat-assisted.
No data is sent to any AI service until you enter an API key and a visitor starts a chat. The plugin does not look up visitor locations with any outside service. API keys are stored in your own WordPress database.
Payments, licensing, and Pro updates are handled by our partner Freemius (checkout, invoicing, and license management). See the Freemius Privacy Policy.
Source Code and Build
This plugin ships its complete, human-readable source. No build step, bundler, or minifier is used.
- assets/js/widget.js — the chat widget script. This is the original hand-written source (short variable names are hand-authored, not the output of a minifier).
- All admin scripts in assets/js/ are likewise plain, hand-written source.
Bundled third-party code: the Freemius SDK (vendor/freemius, GPL), used for opt-in usage analytics and to offer the optional Pro version. Source: https://github.com/Freemius/wordpress-sdk. The free plugin bundles no other third-party JavaScript libraries; Sparxio Pro additionally bundles Chart.js (MIT, https://www.chartjs.org/) for its analytics charts. The widget uses your visitors’ system fonts, so no web fonts are downloaded.
Screenshots
Installation
- In your WordPress admin, go to Plugins Add New and search for “Sparxio”, or upload the
sparxiofolder to/wp-content/plugins/. - Activate the plugin through the Plugins menu.
- Open Sparxio Setup Guide and follow the live checklist.
- Enter your AI provider API key in Sparxio AI Settings.
- Customise the look in Sparxio Appearance.
- The chat widget appears on your site automatically.
FAQ
-
Do I need an API key?
-
Yes. Sparxio uses your own key from one of the supported providers (Anthropic, OpenAI, Google Gemini, DeepSeek, OpenRouter, or a local Ollama model). This keeps you in control of usage and cost. Free keys or trials are available from most providers.
-
Is the plugin free?
-
Yes. This plugin is free and fully functional, with no limits, time restrictions, or license checks. A separate plugin, Sparxio Pro, adds WooCommerce chat, live agent takeover, advanced analytics, and proactive engagement for those who need them.
-
How much does the AI cost to run?
-
You pay your provider directly for API usage. On the most affordable models this is typically a fraction of a cent per conversation. Built-in history limits, summarisation, and timeouts keep usage low.
-
Will it slow down my site?
-
No. The widget loads asynchronously after your page content and uses Shadow DOM, so it never conflicts with your theme’s CSS or JavaScript.
-
Does it work with WooCommerce?
-
Yes, with Sparxio Pro. The AI can search products, add items to the cart, and look up order status (with identity verification) directly in the chat.
-
What languages are supported?
-
The widget interface ships in 30 languages and the AI replies in whatever language the visitor writes in. Turn on multi-language, choose a default language, and enable browser auto-detection in Sparxio Behavior.
-
Can I customise the AI’s personality?
-
Completely. Write a system prompt that defines the assistant’s identity, tone, knowledge, and rules. Sample templates for e-commerce, SaaS, travel, healthcare, education, and real estate are included.
-
How do I remove all data when uninstalling?
-
By default your data is preserved across reinstalls. To wipe everything on uninstall, enable “Delete all data on uninstall” in Sparxio AI Settings Security before removing the plugin.
-
Is my visitors’ data safe?
-
API keys are stored in your database and messages are sent only to the AI provider you choose. No visitor location lookup service is used. See the External Services section above for the full list of connections.
Reviews
There are no reviews for this plugin.
Contributors & Developers
“Sparxio – AI Chatbot, Live Chat & Lead Generation” is open source software. The following people have contributed to this plugin.
Contributors
Translate “Sparxio – AI Chatbot, Live Chat & Lead Generation” into your language.
Interested in development?
Browse the code, check out the SVN repository, or subscribe to the development log by RSS.
Changelog
3.7.5
- An empty or damaged server response is no longer treated as an answer: the widget fetches the real answer for the same message instead, so it is never sent twice
- A message the server received but could not answer is now labelled “Answer unavailable” instead of “Not sent”
3.7.4
- Faster replies: a message is sent the moment the visitor presses Send and the typing indicator appears at once; the artificial pauses before sending and after the answer are gone (the Pre-Typing Delay and Typing Speed settings were removed)
- Every chat and session request now has a time limit. If a reply is slow or the connection drops, the widget keeps asking for the same message’s answer for up to about 2½ minutes, without the message being stored or answered twice
- A message the server is still answering is picked up automatically when ready, instead of showing a “still being answered” error
- Sending a message again updates its existing bubble instead of adding a duplicate; each visitor message shows Sending… / Waiting for reply / Not sent until it is answered
- A message sent just before a page reload gets its answer after the reload (Tab and Browser persistence modes)
- AI providers updated for current models: Claude 5.5 family, GPT-6 / GPT-5.x, Gemini 3.x and 2.5, DeepSeek V4 (deepseek-flash, deepseek-v4-pro), current OpenRouter models. Reasoning models get the right options (low effort, room for thinking), so replies are fast and never empty; thinking text is never shown to visitors
- Retired models are switched to a current one on upgrade and, if a model disappears later, at runtime (with a notice in AI Settings)
- New providers: xAI (Grok), Mistral AI, Groq, and Custom (any OpenAI-compatible API)
- New: backup AI provider, custom model ID for every provider, OpenRouter live model list and backup models, “Only answer from what you know” accuracy setting (on by default)
- Gemini API key is sent in a header instead of the URL; OpenRouter errors returned inside a successful response and Ollama HTTP errors are now reported in AI Settings
- Test Connection shows the model used and how long it took
- Every widget request (live-agent updates, lead form, add to cart, summaries) now has a time limit, so a stalled connection can’t freeze the chat
3.7.3
- Retrying a message whose answer is still being written keeps the same identity (also across a page reload in Tab/Browser mode), so it is never saved or answered twice; a retry gets the original answer back, including product cards
- A saved conversation is tied to its key: if the key had to be replaced, earlier messages stay visible and a new conversation starts instead of a broken one
- Knowledge base: page addresses keep their exact path and query (for example /Guide and /guide are separate pages); only the scheme and host are case-insensitive
- Pro: opening an empty chat no longer replaces the order attribution of an earlier conversation — it is set only after a visitor message is accepted
- Pro: the conversion rate now compares the same conversations (started in the period) instead of mixing two groups
- Packaging: removed a stray runtime file
3.7.2
- Live agent: agent replies are now saved as one message in the conversation transcript, so the admin screen and the visitor see the same messages with the same IDs; your own reply is no longer shown twice when the screen refreshes at the same moment
- Saving problems are reported: if an AI answer could not be stored it is still shown, with a note; hand-off and “release to bot” report failures instead of claiming success
- Retrying after a lost connection no longer saves a lead or message twice, and does not pay for a second AI answer
- Product cards now also appear when a chat returns from a human agent to the AI; the visitor’s language is kept
- Trusted proxy list: malformed entries (such as a wrong /prefix) are rejected and reported
- Rate limiting is now exact under simultaneous requests, also on sites without an object cache
- Conversations end on the server after the Session Timeout; “None” and “Tab” persistence now behave as described; old session cookies are migrated or removed
- Knowledge base: each web page is identified by its address (two pages with the same title stay separate), a failed re-import keeps the previous version, imports are size-limited, and sitemap imports can continue where they stopped
- Pro: office hours support overnight ranges and fall back to the site timezone; analytics use the selected period, site timezone, paid orders minus refunds and each order’s currency
- Pro: order lookup understands short order numbers and an email sent in a following message (both are still required)
3.7.1
- Fixed: the “Start free trial” button opened a blank page
- Chats switched back from a human agent to the AI now show the AI’s reply instead of losing it
- Live agent screen: fewer duplicated or skipped messages
- Lead form: double-clicking while the chat is starting can no longer send the lead twice
- Visitor messages and agent replies report an error if they could not be saved
- Proxy setting: choose None, Cloudflare (verified against Cloudflare’s published IP ranges) or your own trusted proxy IPs
- Removed the optional ip-api.com geolocation. With Cloudflare, the visitor’s country comes from Cloudflare’s own header; no visitor data is sent to a third party
3.7.0
- Privacy: visitor geolocation is now off by default and must be switched on by the site owner; the widget no longer loads Google Fonts
- Language settings (multi-language, default language, auto-detect) are now available to everyone under Sparxio Behavior
- Lead form only confirms success after your details are actually saved, and keeps what you typed if saving fails
- Rate limiting now uses the real visitor IP by default, with an option for sites behind Cloudflare or a proxy
- Knowledge base search now understands Urdu, Arabic, Chinese and accented languages
- CSV exports are protected against spreadsheet formula injection
- Deleting a chat now also removes its related live-agent messages
- Safer session keys and clearer messages when the chat cannot connect
3.6.5
- First public release on WordPress.org.