Agenistrator is an AI-powered automation tool that brings the full power of large language models directly into your WordPress admin dashboard. Through a simple chat interface, you can delegate complex administrative tasks to an intelligent assistant that acts on your behalf, eliminating repetitive work and dramatically accelerating your workflow. Instead of hunting through menus and settings pages or getting bogged down with manual data entry, just tell the agent what you need and it handles the rest.
When given a complex task, the assistant automatically plans its approach, creates a step-by-step strategy, and executes it autonomously. It monitors its own progress and course-corrects if it gets stuck, so you can hand off multi-step workflows and trust that they will be completed reliably.
The assistant can search across your website content, modify WordPress settings through natural conversation, manage users and permissions, write and delete files inside your site’s uploads directory, conduct online research and generate content from it, analyze user comments, and much more. All of this is accomplished through a single, unified interface: a natural language conversation. The full scope of what you can achieve is limited only by your imagination.
This plugin is experimental. It provides powerful capabilities by equipping large language models with tools for not only retrieving information from your website, but also for making changes to it and interacting with the outside world.
By default, only read-only tools are enabled to significantly reduce the risk of unintended consequences. However, the full power of LLMs can only be unleashed by enabling write tools, which requires explicit action on your part.
The performance of your assistant is highly dependent on the quality of the AI model used. If you are not seeing the desired results, try switching to a different model.
Risks: Usage of this plugin exposes users and other parties to both obvious and non-obvious risks, including but not limited to:
Disclaimer: The developers of this plugin accept no responsibility for actions taken by any AI model through this plugin, whether due to prompt injection attacks, AI model hallucinations, or any other reasons. Use at your own risk.
The assistant comes with over 100 built-in tools organized into categories:
Each tool can be individually enabled or disabled. Read-only tools are enabled by default; tools that modify data require explicit activation.
Some tools can fetch external URLs or send emails if you enable them.
This plugin connects to external services to provide chat assistant functionality. You must configure at least one AI provider for the plugin to work. Your conversations and site content are only sent to providers you explicitly configure. The plugin also retrieves model metadata and tokenizer data from external services (see models.dev and OpenAI Tokenizer Data sections below).
OpenRouter is an API gateway that provides access to hundreds of AI models from multiple providers (OpenAI, Anthropic, Google, Meta, and others) through a single API key.
Data sent: When you send a message to the assistant, your conversation history, tool outputs, and relevant site content are sent to OpenRouter’s API, which routes requests to the underlying model provider. If you enable semantic search with an OpenRouter embedding model, your site’s posts and pages are sent to generate embeddings.
When: Data is sent each time you interact with the chat assistant (if OpenRouter is your configured provider) and when embedding synchronization runs (if you’ve selected an OpenRouter embedding model).
Anthropic provides large language models (Claude) for chat completions. Anthropic does not offer embedding models.
Data sent: When you send a message to the assistant, your conversation history, tool outputs, and relevant site content are sent to Anthropic’s API.
When: Data is sent each time you interact with the chat assistant (if Anthropic is your configured provider).
OpenAI provides large language models for chat completions and embedding models for semantic search.
Data sent: When you send a message to the assistant, your conversation history, tool outputs, and relevant site content are sent to OpenAI’s API. If you enable semantic search, your site’s posts and pages are sent to generate embeddings.
When: Data is sent each time you interact with the chat assistant (if OpenAI is your configured provider) and when embedding synchronization runs (if you’ve selected an OpenAI embedding model).
Ollama Cloud is a hosted inference service operated by Ollama that runs large open-weight models on datacenter GPUs. It exposes an OpenAI-compatible API and offers a free tier. Ollama Cloud does not offer embedding models.
Data sent: When you send a message to the assistant, your conversation history, tool outputs, and relevant site content are sent to Ollama Cloud’s API.
When: Data is sent each time you interact with the chat assistant (if Ollama Cloud is your configured provider).
The plugin uses the tiktoken library for token counting, which downloads tokenizer vocabulary files from OpenAI’s public blob storage.
Data sent: Only the HTTP request itself (your server’s IP address). No user data, conversation content, or site information is sent.
When: When the plugin first needs to count tokens (typically on first chat message). The vocabulary file is cached locally in the WordPress content directory and reused for subsequent requests.
models.dev is an open-source database of AI model specifications, used to retrieve model metadata such as context window sizes and display names.
Data sent: Only the HTTP request itself (your server’s IP address). No user data, conversation content, or site information is sent.
When: When the plugin needs to determine context window limits during a chat session (cached locally for 24 hours).
Multisite: Site administrators can manage their own sites, including writing and deleting files inside their site’s uploads directory. Tools that read files outside the uploads directory or access server-level diagnostics (arbitrary file reads, debug logs) require super admin access.