Agent confirmations and memory
What the AI agent in the course builder checks with you before it acts, and how it remembers the preferences you ask it to keep.
What will the AI agent ask me to confirm, and what does it remember?
Before the agent in the new course builder publishes a course, changes your brand voice, saves a workspace-wide memory, reverts a course, or deletes a course, collection, translation, folder or several screens at once, it stops and asks you to confirm or cancel. It only goes ahead when you confirm. When you ask it to remember a preference, it keeps it for this session, this course or your whole workspace, and it checks with you before saving one for the whole workspace.
Which actions need my confirmation?
The agent asks you a question with a confirm option and a cancel option, and tells you what will happen:
- Publishing a course: it names the course and explains that publishing makes your current draft live for learners.
- Changing your brand voice: it describes the new brand voice and explains that it replaces the one your workspace uses for future AI writing.
- Saving a workspace memory: it quotes the exact memory and explains that it applies to every later session in this workspace.
- Reverting a course: it explains that this discards every unpublished change since the last publish, and that it can't be undone.
- Deleting a course: it names the course. Its translations are deleted with it.
- Deleting a collection: it names the collection. The courses in it are kept.
- Deleting a translation: it names the language.
- Deleting a folder: it names the folder. The courses in it move to the top level of your dashboard and aren't deleted.
- Deleting two or more screens at once: it tells you how many screens are affected.
Everyday edits, such as rewriting text, adding a screen, or deleting a single screen or block, happen straight away without a question.
Does one confirmation cover everything?
No. Each confirmation covers the one action you agreed to. If you send another message before the agent acts, or ask for a second publish or delete later on, it asks you again.
What happens if I cancel?
The agent doesn't carry out the action. Your course, brand voice and workspace stay as they were.
What does the agent remember?
When you ask the agent to remember something, such as your preferred tone, your audience or how you name things, it saves it at one of three levels:
- This session: only for the conversation you're in.
- This course: whenever you work with the agent on this course. The course needs to exist first.
- Your workspace: for every later session with the agent in this workspace. This is the only level it asks you to confirm first, and it quotes the exact wording so you can check it.
While it saves, the chat shows Saving context, then Saved context. When it saves more than one at once, you'll see a count instead, such as Saved 2 memories.
Can I see or delete what the agent has remembered?
There isn't a screen in Coassemble to view or remove saved memories. Because a workspace memory applies to every later session in your workspace, read the quoted memory carefully before you confirm it, and cancel if the wording isn't quite right.
Frequently asked questions
Does the agent ask the same questions when I use Coassemble from my own AI assistant?
No. When you connect an AI assistant such as Claude or Microsoft Copilot to your workspace over MCP (the Model Context Protocol), Coassemble doesn't add these confirmation questions. Your assistant decides when to check with you before it runs a tool, and Coassemble tells it which tools make changes. Saved memories are only used by the agent in the course builder.
Will the agent publish my course without asking?
Not in the course builder. It always asks before publishing, names the course, and waits for you to confirm.