Sentient AI — AI-Powered WordPress Chatbot with Semantic Search, Autopilot & MCP Server
| 开发者 | maidulcu |
|---|---|
| 更新时间 | 2026年7月10日 20:08 |
| PHP版本: | 7.4 及以上 |
| WordPress版本: | 6.9 |
| 版权: | GPLv2 or later |
| 版权网址: | 版权信息 |
详情介绍:
- Multi-Provider AI — Supports OpenAI (GPT-4o, GPT-4, GPT-3.5), Claude (Anthropic), Google Gemini, OpenRouter, and Hugging Face. Switch providers without changing your workflow.
- Semantic Vector Search — Two-tier architecture: Local MySQL engine for shared hosting (no custom PHP extensions needed) and cloud engines (Pinecone, Supabase) for scale. Finds content by meaning, not just keywords.
- Browser-Based FAQ Matching — Answers common questions instantly in the visitor's browser. Zero API costs, works offline, no data leaves the site. Falls back to cloud AI when needed.
- WordPress Autopilot — AI agent that manages your WordPress site through chat commands: create posts, check for updates, scan for broken images, clear cache, and more. Administrator-only access.
- MCP Server — Expose your chatbot, FAQs, and site analytics to external AI agents via the Model Context Protocol. Connect Claude Desktop, Cursor, or custom AI agents directly to your WordPress installation.
- GEO Scanner — Analyze content for Generative Engine Optimization. Score each post/page across 10 factors that determine how AI search engines (Google SGE, Perplexity, ChatGPT Search) evaluate your content. One-click fixes for schema markup.
- Empathy Engine — Real-time emotion detection adapts the AI's response tone. Automatically escalates frustrated users to human support.
- Glass Box Transparency — Every AI response shows a confidence score (0-100%) with source attribution. Admins can review low-confidence queries and improve their FAQ database.
- Shared hosting (Bluehost, SiteGround, GoDaddy, HostGator) — Local MySQL vector engine works without SSH access or custom PHP extensions
- Cloud / VPS — Pinecone or Supabase for high-scale vector search
- Privacy-first mode — Local Brain + Privacy Mode = zero external API calls
- Semantic search that understands meaning, not just keywords
- Full site management through chat commands (Autopilot)
- External AI agent integration via MCP (standard protocol)
- Content optimization for AI-powered search engines (GEO)
- Emotion-aware responses with automatic escalation
- Transparent confidence scoring and source attribution
- Privacy-first local processing with zero API costs
- Multi-provider flexibility — use OpenAI, Claude, Gemini, or others
安装:
- Upload the
sentient-aifolder to/wp-content/plugins/ - Activate the plugin from the Plugins menu
- Navigate to Sentient AI → Settings to configure
- Select your AI provider and enter your API key in the AI Providers tab
- Enable the chatbot in the General tab
- Customize appearance and behavior in the Widget and Features tabs
- Place
[sentient_ai_chat]on any page, or use the floating widget
屏幕截图:
常见问题:
Can this work without an API key?
Yes. The browser-based FAQ matcher and Local Brain feature answer common questions using your FAQ database without any API calls. For AI-generated responses beyond existing FAQs, an API key from a supported provider is required.
Which AI providers are supported?
OpenAI (GPT-4o, GPT-4, GPT-3.5), Claude by Anthropic, Google Gemini, OpenRouter (multi-model access), and Hugging Face. You can switch between providers at any time.
How does the semantic vector search work?
The Vector Database (v2.8.0) converts your content into mathematical embeddings and stores them for similarity search. The Local MySQL engine works on any shared hosting using standard WordPress tables. For higher scale, Pinecone and Supabase cloud engines handle millions of vectors with SOC 2/GDPR compliance.
Which vector engine should I choose?
For shared hosting with under 10,000 posts, the Local MySQL engine works without any additional setup. For larger sites, Pinecone or Supabase offer free tiers and enterprise-grade scalability.
What is the MCP Server?
The MCP Server implements the Model Context Protocol, a standard for connecting AI agents to external systems. It allows tools like Claude Desktop, Cursor, or custom AI agents to interact with your WordPress site — querying the chatbot, searching FAQs, managing content, and accessing analytics — all authenticated via API key.
How does the GEO Scanner help my content?
Generative Engine Optimization (GEO) ensures your content is cited by AI-powered search engines. The scanner analyzes each post across 10 factors — structured data, FAQ sections, citations, statistics, author expertise, content structure, and more — providing a score (0-100) with actionable suggestions and one-click fixes.
Does this work with WooCommerce?
Yes. The chatbot supports product lookup, order status checking, recommendations, and cart information. WooCommerce data is accessible through chat and is included in the Autopilot's site management capabilities.
Is my data sent to external services?
Only when you explicitly configure an AI provider API key and a user query requires AI-generated response. FAQ-matched queries, Local Brain responses, and Privacy Mode interactions never leave your server. The plugin holds no data on its own — you control all API keys and provider choices.
Can I customize the chatbot appearance?
Yes. Six preset themes (Indigo, Ocean, Emerald, Rose, Amber, Slate) plus custom color pickers, border radius adjustment, and a live preview in settings. You can also configure quick replies, canned responses, and a pre-chat form.
What is WordPress Autopilot?
An administrator-only AI agent that executes WordPress tasks through the chat interface. Supported commands include creating posts and pages, listing content, checking for plugin/theme/core updates, finding broken images, clearing cache, and retrieving site statistics. All actions are logged for review.
更新日志:
- WooCommerce Deep Integration — HPOS, product vectorization, real-time webhooks
- HPOS (High-Performance Order Storage) support — reads orders from custom wc_orders tables when enabled, auto-falls back to legacy wp_posts
- get_order_stats_hpos() queries either HPOS tables or legacy storage based on what's active
- hpos_enabled() detection method for plugin-wide awareness
- Product vectorization — auto-generates rich embeddings from product title, description, attributes, categories, tags, reviews, FAQs, variations, and upsells
- on_product_save() hooks into woocommerce_new_product and woocommerce_update_product
- reindex_all_products() bulk reindex with woocommerce_product_import_end integration
- Real-time WooCommerce webhooks — REST endpoint at /wp-json/sentientai-woo/v1/webhook with HMAC signature verification
- 8 webhook events: order created, completed, cancelled; product updated; stock status changed; review created; low stock; no stock
- Event data includes structured payloads with formatted messages for logging
- Product webhooks trigger automatic re-vectorization
- Secret key configuration with optional signature validation
- All events logged to query log with source='webhook'
- New WooCommerce Deep Integration card in Features tab with 4 toggles and HPOS/vectorization info
- Fixed: $wpdb global used before declaration in handle_webhook_event()
- Fixed: Missing register_setting() for sentient_ai_woo_webhooks_enabled and sentient_ai_woo_webhook_secret
- Added: sentient_ai_woo_webhooks_enabled to checkbox preservation array
- Added: 72 test cases covering all WooCommerce integration methods
- Conversational Commerce Engine — AI shopping assistant with Vibe Search, Visual Shop, Gift Concierge
- Vibe Search: natural language product matching via Vector DB with tokenized match reasons
- Visual Shop: image URL → AI description → visual product matching with domain security
- Gift Concierge: 5 guided questions → personalized gift guides with budget/personality scoring
- Admin toggles for each feature + allowed image domains list
- Fixed: Vibe search used get_post() instead of wc_get_product() for vector results
- Fixed: Visual Shop didn't check WooCommerce is active before loading products
- Fixed: Match reasons checked full query against product name instead of individual tokens
- Fixed: Gift Concierge questions missing translation wrappers
- Fixed: Extra PHP bracket in Commerce Engine feature card array
- New: Vector Database with two-tier semantic search — Local MySQL (shared hosting) and Pinecone/Supabase cloud engines
- New: Auto-generated embeddings for posts, pages, and FAQs with OpenAI API or built-in hash fallback
- New: Semantic search integrated into chatbot responses for meaning-based content matching
- New: Admin settings page with engine selector, cloud provider configuration, and index status dashboard
- Combined 2.7.0 features: MCP Server (8 tools, SSE transport, API key auth, CORS headers)
- Combined 2.7.0 features: GEO Scanner (10-factor content analysis, scoring dashboard, one-click schema fixes)
- Combined 2.7.0 features: Browser-based FAQ matcher with keyword and title matching
- Fixed: Settings reset when saving across tabs (Enable Chatbot, Pre-Chat Form, role limits)
- Fixed: Pre-Chat Form data not persisting across page loads (sessionStorage)
- Fixed: Welcome message not displaying when chat opens
- Fixed: Role limit field blocking form submission when feature is disabled
- Fixed: SSE endpoint infinite loop (timeout added)
- Fixed: Missing content format in MCP manage_wp tool
- Fixed: Unhelpful error message when no API key configured
- Fixed: FAQ matching skipped when conversation history exists
- Chore: Added Help tab with complete feature documentation
- MCP Server — Model Context Protocol support for AI agent integration
- 8 tools: chat, FAQ search/list/create, contacts, stats, low confidence, WordPress management
- 3 resources: FAQ database, usage stats, contacts
- 2 prompts: site overview, FAQ improvement
- API key authentication, SSE transport, CORS, rate limiting
- GEO Scanner — 10-factor content analysis for AI search visibility
- Factors: content length, structured data, FAQs, citations, statistics, E-E-A-T, structure, meta, lists, titles
- Scoring dashboard with auto-fix actions
- Browser-Based FAQ Matcher — Local FAQ search without API key
- Bug fixes — Multi-tab settings persistence, pre-chat form stability, welcome message display
- Help tab — Complete feature reference with documentation
- Security Hardening — Remediated Assistant ID leak, added output sanitization
- Design Refresh — Modernized UI with CSS variables, improved SVG rendering
- Bug Fixes — Resolved ReferenceError, history persistence, transient counting, approval flow
- WooCommerce Deep Integration — Product lookup by name/SKU, order status, recommendations, cart info
- Performance: FAQ query caching, Options_Cache helper
- Widget Branding & Theming — 6 themes, custom colors, border radius, live preview
- Prompt Injection Protection — 20+ injection pattern detection
- Local Brain — Fuse.js-powered fuzzy FAQ search in the browser, zero API costs, offline support
- Privacy Mode — Restrict all processing to local database
- WordPress Autopilot — 10 commands: updates, broken images, posts, security, cache, stats, health
- Sentient Memory — Topic clustering, feedback tracking, auto-FAQ generation, learning dashboard
- Empathy Engine — Emotion detection and dynamic tone adaptation
- Glass Box Transparency — Confidence scoring and source attribution
- Multi-turn conversation context and server-side history
- Security and performance hotfixes
- Session-based chat history, TXT export, configurable position
- Initial release with multi-AI support