No AI inside. Your agent, on your machine.
Xogogy contains no model, no prompt and no API key. If you already use an AI agent, it can work on your courses through the app, on your computer, and you watch every change land.
What an agent can do
- Read the course outline, a slide, or the text and answer keys, using stable ids.
- Add a lesson, a slide, text, images or questions; change text, theme colours and settings; check the course for problems.
- Render a slide as a learner would see it and run the accessibility scan.
- Work with the app open, so you watch the change happen, or with it closed, on the project folder.
Build interactions, not just pages
An agent does not only fill in text. It can wire up how a course behaves, using the course's own variables and actions, the same ones you use by hand.
- Variables. Add a text, number or true/false variable, give it a starting value, rename it (fixing every place the name is used in your text), and read it back.
- Triggers. Add, change or remove a trigger on a slide, a layer or the whole course: when the learner clicks something, when a variable changes, when the timeline reaches a point, with conditions such as “only if the score is above 70”.
- Actions. A trigger can set a variable, show or hide a layer, go to a slide or move on to the next one, and more, from the same list of actions the editor offers.
- Layers, cue points and states. Add a layer for a pop-up or feedback message, add a cue point on the timeline, and set which layers and objects start visible.
Example: “Add a slide where the learner picks a tool, store the choice in a variable, and show a different feedback layer for each choice.” The agent creates the slide, the variable, the layers and the triggers in one labelled batch. You watch it appear, check it in Preview, and undo it in one click if it is wrong.
Agents build with structured triggers and actions. A script is added only if you ask for one; it is stored as text and never run by the agent tools. Our agent guide tells it to check the result for errors and accessibility problems and never to invent an answer key.
How it stays safe
- Every request is applied as one batch with a label such as “Claude: added lesson 4”. If it would leave the course invalid, nothing changes and the agent is told why.
- Changes appear as they land, with a notice naming the batch. “Undo the agent's last change” is one click, or one tool call. Review mode holds changes until you accept them.
- You approve the first connection to the open app: allow, read only, or deny. With the app closed, the agent can open only courses in the folders you allowed.
- The connection is local only: a process on your computer. The app opens no network port, and the agent can only read files in your project and images you name.
- Text, scripts and actions inside a course are treated as data, never run by the agent tools.
Which agents
The connection uses the open Model Context Protocol, with a one-click extension for Claude Desktop, and there is a plain command line (xogogy) for agents that can run commands. Setup guides cover Claude Desktop, Claude Code, Codex, GitHub Copilot in VS Code and Cursor. Browser-only chat products that cannot start a local program cannot connect, because the app runs no network service.
Who pays for the AI
You choose and pay for your own agent. Xogogy never sends your course to a model by itself.
Build your next course. Bring the ones you have.
Building, importing, editing and Preview are free. Pay only when you're ready to publish.