內容簡介
Zybroz AI Agents 將您的 WordPress 控制台轉變為 AI 自動化平台，透過建立可重複使用的 AI 代理來執行各種任務，如撰寫元描述、生成圖片替代文字及網站維護等，提升網站管理效率。
【主要功能】
• 自動化 SEO 任務，批量撰寫元描述
• 改寫與改善文章，檢查可讀性
• 自動生成結構化內容，避免虛假資訊
• 審核內部連結，找出斷鏈
• 自動備份資料庫，確保安全性
• 支援多種 AI 模型，如 OpenAI GPT、Claude
外掛標籤
開發者團隊
② 後台搜尋「Zybroz AI Agents – AI Automation」→ 直接安裝（推薦）
原文外掛簡介
Zybroz AI Agents turns your WordPress dashboard into an AI automation platform. Instead of copy‑pasting prompts into ChatGPT, you build reusable AI agents that do real work on your site: writing meta descriptions, generating image alt text, auditing internal links, building FAQ schema, moderating comments, backing up your database, and more.
Each agent is a structured configuration with its own persona, scope, logic, safety rules and version history, so its behaviour is predictable and repeatable. Every change is previewed with a dry run and protected by an automatic backup and one‑click rollback before anything is written to your site.
It is built for site owners, bloggers, marketers, agencies, developers and small businesses who want practical AI automation for WordPress – not just another chat box.
Automate WordPress SEO Tasks
Draft SEO meta descriptions in bulk, audit your SEO plugin setup for conflicts, generate FAQ blocks for rich snippets, and flag stale content that needs refreshing.
Automate Content & Accessibility
Rewrite and improve posts, check readability, and generate structured content with clear guardrails against invented facts. Scan for images missing alt text and auto‑generate descriptive, SEO‑friendly tags.
Automate Site Maintenance & Links
Find orphan pages with no inbound links and audit internal links for broken targets. Report available plugin updates and site‑health issues, find unused or oversized media, and run full database and uploads backups.
Automate Comment Moderation
Review and act on comment queues with an automatic backup taken before any bulk action is applied.
Works with OpenAI GPT, Claude, Gemini and local AI models
One OpenRouter API key routes every cloud AI call through a curated fallback chain, with free models prioritised so you can start at zero cost. Through OpenRouter you can reach Claude, GPT, Gemini and dozens of other models without adding a separate key for each. Prefer to keep everything on your own hardware? Point the plugin at a local Ollama server and no data leaves your network.
The logic‑only InbuiltEngine runs maintenance and admin agents with no API key at all.
Privacy first, no lock‑in
No account required. No telemetry on activation or automatically, ever.
Nothing is sent to any AI provider until you add your own key and run an agent.
API keys are encrypted before storage and never exposed to the browser.
Export any agent as JSON or a plain prompt template and reuse it on any site or in any external AI tool.
Why choose Zybroz AI Agents?
Safety by default – dry‑run preview, automatic post snapshots, and one‑click rollback before every write operation.
You stay in control – agents move through draft, active and failed states; write agents always show you exactly what they found before asking for approval.
Genuinely free – all core features are fully functional with no usage caps, quotas or hidden paywalls.
Repeatable, not random – structured agent JSON with versioning means the same agent behaves the same way every time.
Who is it for?
Bloggers automating on‑page SEO, agencies standardising site audits across many clients, developers who want scriptable AI actions with a REST surface, and store owners who need bulk content and maintenance tasks handled safely.
Key Features
Visual Agent Builder – two‑column builder with tabbed panels for Persona, Scope, Logic and Safety, plus a live JSON preview. Save and qualify is gated behind a successful dry run.
AI Prompt Generator – describe what you want in plain English and let AI write a structured system prompt; compare, accept, keep or regenerate.
Chat‑to‑Create Wizard – a guided flow that suggests an agent name, role, goal, system prompt and category, creates a draft, and runs a dry run for you.
Agent Versioning – automatic patch bumps on save, stable and major release actions, plus full version history and restore.
Export / Import – export agents as JSON or as a prompt‑template text file for external AI tools; import either format back into the builder.
Multi‑model AI – one OpenRouter key routes every cloud call through a curated fallback chain (free models first), plus local Ollama support.
BYOK security – API keys are encrypted before storage and never exposed to the browser.
InbuiltEngine – logic‑only AI Actions that run with no API key for maintenance and admin tasks.
Agent Library – ready‑to‑use starter agents across content, SEO, images, schema, social, comments and admin health.
AI Action Library – vetted atomic operations for real WordPress changes, including read and write actions.
Advanced Backup Module – full database and uploads backups with list, delete and restore, stored in a protected per‑site folder.
Quality Precheck – validates agent JSON before qualification and highlights missing fields, scope issues or missing provider settings.
Safety Layer – automatic post snapshots, dry‑run preview mode, one‑click rollback, and a consent gate on write-capable steps that halts and requires your approval in every run mode before anything is written.
Workflows – chain agents, actions and conditions into multi‑step automations with a visual builder and run history.
Community Library – browse, install, and optionally publish agents to/from the Zybroz Hub, with ratings, favorites, and update notifications for Hub‑sourced items.
More AI Providers – optionally configure a direct key for Anthropic, OpenAI, Google, or WordPress’s own Core AI Client, as an alternative to OpenRouter.
Run History & Metrics – track total runs, success rate and favourites for each installed agent.
Dashboard – site health, recent runs, provider connection status and quick actions in one place.
WooCommerce Product Copy – read full product grounding context (name, SKU, price, stock, categories, attributes, images/alt text) and update a product’s full and short description without touching price, stock, variations or tax/shipping settings.
Achievements & Share Cards – run-count and streak milestones unlock automatically, and any run or milestone can be turned into a branded, AI-captioned share-card image from the Runner.
Starter Agents
Ten simple, low‑risk agents are installed and ready to run the moment you activate the plugin:
Accessibility Scanner – finds published content with images missing alt text.
Alt Text Architect – generates descriptive alt text for images missing descriptions.
Meta Description Writer – drafts SEO meta descriptions for a post.
FAQ Generator – builds FAQ pairs from a post for rich‑snippet eligibility.
Link Health Checker – audits internal links for broken targets.
Media Library Auditor – reports unused and oversized media.
Site Health Reporter – reports which plugins have updates available and summarises site health.
SEO Plugin Auditor – checks your SEO plugin setup for conflicts and gaps.
Orphan Page Finder – lists published pages with no internal links pointing to them.
Outdated Content Scanner – flags stale posts that may need refreshing.
This bundled set replaces the five starter agents shipped in 0.9.391 (Traffic Healer, Alt Text Architect, Internal Link Weaver, Schema Guardian, Social Sniper). If you installed any of those on 0.9.391, they are not removed from your site on upgrade — only the newer bundle above is pre‑installed for new activations. Traffic Healer, Internal Link Weaver, Schema Guardian and Social Sniper remain available to install from the Community Library.
More advanced agents – multi‑step rewriters, an autonomous site orchestrator, taxonomy and schema tools, comment moderation and more – are available in the Community Library under the “Testing” filter. Install one to try it and help review it.
Requirements
PHP 8.1 or higher.
WordPress 6.4 or higher.
An OpenRouter API key for cloud AI models, or Ollama running locally. Neither is required for logic‑only (InbuiltEngine) agents.
Bundled Libraries
Action Scheduler 4.0.0 (GPLv3) — used for the plugin’s background job queue (backups, scheduled scans). Because Action Scheduler is licensed GPLv3, the combined work is distributed under GPLv3 terms; see License above.
Privacy
No data is sent to third parties unless you configure an external API key, or unless you explicitly connect a Zybroz Hub account under Agent Library -> Contribute.
Depending on which AI Agents and AI Actions you run, the content sent to your configured AI provider can include WordPress content such as post titles/body, and – for comment‑moderation agents – commenter names and comment text. See “External Services” below for exactly what is sent and when.
API keys and Hub credentials are stored encrypted.
Ollama runs locally with no external calls.
No telemetry is sent on activation, and none is sent at all unless you explicitly connect a Zybroz Hub account under Agent Library -> Contribute. Once connected, the plugin sends: basic site info (once, to link your site to your account); the actions you explicitly take (voting, favoriting, submitting an agent, etc.); a daily automatic sync of per-agent run metrics (success/failure counts, no post content); and, after each agent run completes, an automatic report of that run’s outcome (success/failure only, no post content). A random anonymous install identifier (no personal data) is generated and sent alongside these so the Hub can associate them with your connected account. See “External Services” below for exactly what is sent and when.
External Services
This plugin can optionally connect to external AI services to run agents: OpenRouter (a unified API that re‑serves Claude, GPT, Gemini and other models) by default, or — if you choose to configure one in Settings -> More AI Providers — a direct connection to Anthropic, OpenAI, or Google using your own key for that provider. It can also talk to Ollama, which you run yourself, and to whatever provider you’ve separately registered through WordPress’s own Core AI Client. All connections are user‑initiated and require you to supply your own API key for the provider you choose. No data is sent to any provider without a key being configured and an agent run being triggered.
The exact content sent depends on which agent or AI Action you run. This always includes the agent’s configured system prompt. Depending on the agent’s scope, it can also include WordPress content relevant to that agent’s task – for example, a post’s title and body for a content‑rewriting agent, or a commenter’s name and comment text for a comment‑moderation agent. The plugin does not send WordPress credentials, API keys, or payment information to any AI provider. If your site handles content containing personal data (e.g. names in comments or post content), that data may be included in the request when an agent processing that content is run – review each agent’s scope before running it on content containing personal data.
OpenRouter API (New default path since 0.9.392)
What it does: Routes your configured agent prompts to one of several underlying AI models (selected automatically from a fallback chain you configure in Settings) and returns AI‑generated text responses for WordPress automation tasks.
Data sent: The system prompt, plus whatever WordPress content the agent’s scope includes (e.g. post title/content, or commenter name/comment text for comment‑moderation agents). No WordPress credentials are sent.
When: Only when an agent run is triggered, or when you click “Test Connection” on the Settings page (checks key validity only – no WordPress content sent).
Provider: OpenRouter, Inc.
Terms of Service: https://openrouter.ai/terms
Privacy Policy: https://openrouter.ai/privacy
Direct provider connections (optional, opt-in under Settings -> More AI Providers)
0.9.391 and earlier connected directly to Anthropic/OpenAI/Google as the only path, with no OpenRouter option. 0.9.392 retires those original key fields (see the Migration note in the 0.9.392 changelog entry) and makes OpenRouter the new default — but if you’d rather not route through OpenRouter, a newly-added “More AI Providers” panel (New in 0.9.392) lets you configure a fresh direct key for any of the providers below instead. Only the provider(s) you configure a key for are ever contacted; the plugin does not share your key or your content with any provider you haven’t configured.
Anthropic PBC (Claude models) – What it does: runs your configured agent prompts directly against Anthropic’s API when a direct Anthropic key is configured and selected. Data sent: the same system prompt + WordPress content the agent’s scope includes, as described above. No WordPress credentials sent. When: only when an agent run is triggered using this provider, or when you click “Test Connection” for it. Terms: https://www.anthropic.com/legal/consumer-terms – Privacy: https://www.anthropic.com/legal/privacy
OpenAI, LLC (GPT models) – Same data/when terms as above, using your configured OpenAI key. Terms: https://openai.com/policies/terms-of-use – Privacy: https://openai.com/policies/privacy-policy
Google LLC (Gemini models) – Same data/when terms as above, using your configured Google key. Terms: https://ai.google.dev/gemini-api/terms – Privacy: https://policies.google.com/privacy
WordPress Core AI Client (Beta, WP 7.0+) – Delegates to whichever AI provider you’ve separately registered under Settings -> Connectors in WordPress core; this plugin never sees that provider’s key, only the response. Review the terms of whichever provider you’ve registered there.
If you also place one of these providers’ model slugs in your OpenRouter fallback chain instead of configuring a direct key for it, OpenRouter forwards that request to the provider on its own infrastructure, and that provider’s terms apply to the forwarded traffic — the same disclosures above apply in that case too.
Ollama (Optional, local only)
What it does: Runs an AI model on a server you control and returns generated text.
Data sent: The same prompt/content an agent’s scope includes – sent only to the Ollama host URL you configure (by default http://localhost:11434, i.e. your own machine).
When: Only when an agent run is triggered with Ollama selected as the provider, or when you click “Test Connection” for Ollama in Settings.
Provider: none – this is self‑hosted software, not a third‑party service. No data leaves your network unless you point it at a remote host yourself.
Project: https://ollama.com
Zybroz Hub / AI Agents Central Library (Optional — account-linked actions below are New in 0.9.392; catalog browsing existed in 0.9.391)
What it does: Fetches a catalog of community‑contributed starter agents from the central registry so you can browse and install them from Agent Library. Browsing the catalog is read‑only and does not require an account.
Data sent (browsing, no account connected): None beyond the request itself. Read‑only GET requests to the registry endpoint to retrieve the public agent catalog.
Data sent (only if you explicitly connect a Zybroz Hub account under Agent Library -> Contribute): your site URL, site name, WordPress version, and PHP version (sent once, to link your site to your Hub account); a random anonymous install identifier generated locally (no personal data); the content of any action you explicitly take – casting a vote, favoriting an item, submitting a change request, or submitting an agent to the library; a daily automatic sync of per-agent run metrics (success/failure counts only, no post content); and an automatic report of each agent run’s outcome (success/failure only, no post content) sent right after that run completes. No data of any kind is sent unless you have connected an account – the daily …
