Working with the Assistant

The AI assistant is a chat panel that opens beside the lab editor. You ask it questions about a lab or tell it what to change, and it edits the lab directly: content, activities, tabs, and sandbox configuration. Its changes land on the same working branch as your own, so you review and publish them the same way.

What you can do with the assistant

Use case	What the assistant does	Try asking
Create a lab from scratch	Builds the sandbox, instructions, and checks from a description. Start from the Create a lab with AI dialog, or ask in the chat on a new lab.	”Build a 20-minute lab that teaches a beginner to deploy a container and expose it on a port.”
Edit an existing lab	Changes only what you ask about and keeps the rest of the lab as it is.	”Add a quiz at the end of chapter 2.” / “Rename chapter 1 to Getting Started.”
Validate that a lab works	Plays through the lab in a real session — starts the sandbox, then runs every task and quiz in order — and reports what failed.	”Test the lab and tell me what breaks.”
Debug a lab	Reads the lab’s configuration and scripts, finds the cause of a failure, and fixes it. Combine it with a test run to find the failure first.	”The check on page 3 fails even when I do the task correctly. Find out why and fix it.”
Ask questions about a lab	Answers from the lab as it stands and changes nothing. Useful when you take over a lab someone else wrote.	”What does this lab cover?” / “Which resources does the sandbox use?” / “Is the difficulty right for beginners?”
Learn how to build something	Explains how to configure lab resources, such as sandboxes, tabs, tasks, and quizzes, and can search the web for product details.	”How do I give users a terminal and a code editor side by side?”
Work from a screenshot or file	Reads images and PDFs you attach, and can add an attached file to the lab.	”Build a task that matches the steps in this screenshot.”

What the assistant does not do

It does not publish. You review its changes and publish them yourself.
It does not run commands in a sandbox or read logs. To find runtime failures, it uses a test run, which reports the step that failed and its error.

Open the assistant

Open a lab and enter edit mode.
In the header, click Assistant.

The AI Assistant panel opens to the left of the page you are editing. It stays open as you move between pages of the lab, so you can reach it from anywhere in the editor. Click Assistant again to close it. The panel closes when you leave edit mode — for example after you publish or discard your changes.

If your team has no active Anthropic API key, the panel shows AI is not set up instead of the chat. See Using your own API key.

Editing an existing lab

The assistant works on any lab: a lab you created with AI, a lab you built by hand, or a lab someone else wrote. It reads the lab as it stands and makes the change you ask for, touching only what you asked about.

On a lab with no conversation yet, the panel asks What would you like to work on? and offers suggestions you can click to send:

Build — for example Add a chapter, Add a quiz, or Wire up a sandbox.
Improve — for example Tighten the instructions or Refresh anything out of date.
Ask about this lab — for example What does it cover?, Where are the gaps?, or Is the difficulty right?

You can also type your own request, such as “shorten chapter 2” or “which resources does this lab use?”. A question gets an answer from the lab, and the assistant changes nothing.

The assistant is idle until you send your first message. Opening or closing the panel does not start or stop it.

The chat

Type into the message box to steer the assistant. While it works, an activity line at the bottom of the conversation shows what it is doing and how long the turn has run.

You steer the assistant between turns. While a turn runs, the message box does not accept a new message. Turns are short and each one reports what it did, so if the assistant takes a wrong direction, correct it in your next message.

Attach files

You can attach files to a message, for example a screenshot or a diagram that shows what you want. Paste a file into the message box, or drag it onto the message box.

Each file can be up to 3 MB.
A message can carry up to 10 files.
Large images are scaled down before they upload.

A message can be files only, with no text. If the assistant cannot use a file type, the file shows an error and is not sent.

Answer questions and approve changes

Sometimes the assistant needs input before it can continue. These prompts appear above the message box:

Needs your input — the assistant asks a question and offers answers. Click one to reply. When you can pick more than one, select them and click Accept.
Change to review — the assistant asks to change a specific file. Click Approve to let it continue, or Deny to refuse. To explain what it should do instead, send that in your next message.

Review what changed

When a turn ends, the chat shows a summary of what the assistant added, updated, and removed. Click an item to open it in the editor. Click View unpublished changes to see every change on the working branch.

Run and test the lab

After a turn that changed the lab, the chat offers two ways to check the result:

Play — starts the lab in a fresh sandbox so you can try it yourself. If a lab session is already running, the button reads Restart lab, because the running session still has the version it started with.
Test lab — asks the assistant to play through the lab and report what breaks. The test starts a real session on your unpublished changes and runs every task and quiz, so it takes several minutes. The assistant then fixes what it can and tells you what it changed.

A test failure is not always caused by the lab. The assistant reports what the test run found, so read its report before you change the lab.

What you can do during a turn

Your edits and the assistant’s edits collect on the same working branch, so you can mix chat requests with changes you make yourself.

While a turn runs:

Publish and Discard changes are disabled. They become available again when the turn ends.
The file editor in Files is read-only, so your edits do not clash with the assistant’s.
Review changes stays available.

When you are happy with the result, review and publish the changes through version control, as you do for any lab.

Stop or restart the session

The menu in the panel header has two actions:

Stop session — ends the session and leaves the assistant idle. Your next message starts a new session.
Restart session — ends the session and starts a new one straight away. The old conversation stays visible above a Previous session divider.

The Stop button that replaces the send button during a turn also stops the session. There is no way to stop only the current turn.

Instruqt asks you to confirm both actions, because the assistant does not remember anything from an earlier session. Everything it already wrote to the lab is kept. If something from the earlier conversation still matters, mention it in your next message.

Use these actions when a session has gone off track, or when it works from context that is no longer current — for example after you discarded the changes it was building on.

Recover from problems

If the assistant cannot start or a turn fails, the chat shows an error with a Try again button. Try again starts a new session. Sending a message does the same.

If the connection drops during a turn, the chat reconnects on its own and replays the conversation, so you do not need to refresh. Refreshing is also safe: the chat reconnects to the running session.

If the error is about your API key, for example the key is revoked or out of credit, see When the key fails.

Usage

The assistant runs on your team’s own Anthropic API key. The key’s name shows at the bottom of the message box. Your spend is visible in the Anthropic console, not in Instruqt. See Usage & limits.