Agent Studio
A WordPress plugin · Free and open source

Build AI agents inside WordPress

An agent is a set of instructions, something that starts it, and a short list of things it's allowed to do. It runs in the background on your own site and asks you before it changes anything you didn't pre-approve.

Everything an agent needs, built on WordPress

Agent Studio uses what's already in core: the Abilities API for what agents can do and the AI Client for whichever model you've connected. Agent Studio never holds your AI keys.

Agents

Name, plain-language instructions, triggers, allowed abilities, skills and a model, with a sentence of help on every field. Start from a template: pre-publish reviewer, welcome-email writer or comment moderator.

Triggers

A post changes status, a user registers, a comment arrives, a schedule comes round, a signed webhook is called, someone messages the agent, or another plugin fires its own event. Or click Run now.

Abilities

Anything a plugin registers with the WordPress Abilities API can be given to an agent. Starter abilities cover posts, comments, users, private notes and email, so agents are useful on a plain site.

Skills

Reusable SKILL.md-style instructions. Agents see each skill's "when to use it" line and load the full text only when needed. Import, download, and restore earlier versions.

Connections

Paste an MCP server's address and its tools become abilities. Protected servers sign in with OAuth or an API key, and anything not marked read-only needs approval.

Channels

Talk to agents in Slack or Telegram, and get push notifications with Approve and Reject buttons through ntfy. Only people you let in can chat.

Approvals

Reads run freely. Writes, destructive actions and anything that reaches outside your site can pause the run until you approve, from the run screen, email, chat or WP-CLI.

Run log

Every run keeps its trigger data, steps, tool calls, results and token use. Old runs are pruned after 30 days.

For developers

WP-CLI commands, signed webhooks, and filters and actions for adding your own triggers, channels, connection types and model clients.

How a run works

Agents work in the background, so the page save, signup or webhook that started one returns right away.

  1. Something happens on your siteAn event matches an active agent's trigger. Duplicate checks, an hourly limit and a loop guard run first.
  2. The run is queuedIt's stored and handed to Action Scheduler if you have it, or WP-Cron otherwise.
  3. The agent acts as a real userEvery ability checks that user's permissions, so an agent can never do more than the account you gave it.
  4. The model works with only the tools you allowedIt sees the instructions, the event data and its allowed abilities, and calls tools until it's done or reaches its step limit.
  5. You approve what mattersA tool that needs approval pauses the run and notifies you. Approve or reject, and the run carries on.

Careful by default

AI on a live site needs guardrails. Agent Studio is built so that the worst case is a run waiting for your approval.

  • Least privilegeAgents call only the abilities on their allowlist, as the user they act as.
  • Event data is treated as dataPost content, comments and webhook bodies are labelled as untrusted, never as instructions.
  • No runaway chainsEvents caused by an agent's own actions never start agents.
  • Limits on every agentA step limit per run and a cap on runs per hour.
  • Secrets stay secretTokens use the WordPress Secrets API when available, and are encrypted until then.

Install

Once it's installed, new versions show up under Dashboard → Updates like any other plugin, and you can update with one click or turn on automatic updates.

From your dashboard

Download the latest zip, then go to Plugins → Add New → Upload Plugin, choose the file, and activate it.

With WP-CLI

Install and activate the latest release in one step:

wp plugin install https://github.com/wpscholar/agent-studio/releases/latest/download/agent-studio.zip --activate

After activating, connect an AI provider under Settings → Connectors if you haven't already, then open Agents → Abilities to check your site is ready to run agents.