The AI Tutor automatically reviews students’ assignments, and you can configure it to follow your course methodology. It analyzes each answer, compares it with the reference answer and your course materials, and makes a reasoned decision using logic similar to that of a human instructor.
The AI Tutor can reduce repetitive grading work. Use test cases and manual-review outcomes to check its accuracy and consistency for your courses.
You decide how much the AI Tutor does on its own. On Autopilot it sets the result and replies to the student when it is confident. In With Human Review mode it only prepares a recommendation, and a person on your team confirms it. See Review Mode.
How to Set Up the AI Tutor
Go to Settings → AI Agents. The page lists all the AI agents in your account.
Click "Create Agent".

In the "What should this agent do?" window, select "AI Tutor". AXL creates the agent and opens its settings page. The new agent starts out turned off.

On the settings page, complete the following fields in the General Settings card:
Name. Enter an internal name that your team can recognize.
Description. Describe what this agent is for.
Display Name. Enter the name students will see.
Courses. Select the courses whose assignments the AI Tutor will review, or leave Courses empty to cover all courses. Only one tutor can cover all courses, and two tutors cannot select the same specific course. To have it skip a specific assignment, select "Disable AI tutor review" in that assignment’s settings.
Review Mode. Choose who gives the verdict. "Autopilot": the tutor sets the verdict and replies to the student on its own when it is confident. "With Human Review": the tutor only prepares a recommendation, and a person applies or corrects it on the Assignment Review page. See Review Mode.
Instructions. Describe the AI Tutor’s tone, role, and any extra rules.
Analysis Strategy. Choose how the AI Tutor balances speed, cost, and quality of analysis.
Knowledge Base. Upload TXT, PDF, DOCX, or JSON reference files. Public pages can also be added through Knowledge Links.
Leave a comment for the student. Shown when Review Mode is "Autopilot". When a live review result is applied automatically, this option allows the tutor to post its generated feedback. It does not post a comment for a result requiring manual review or for a test; for a manual-review result, the reviewer sees the comment as a draft in the tutor's recommendation. Turning it on may increase usage costs. In "With Human Review" mode the switch is hidden, because the tutor always drafts a comment for the reviewer.
Analyze attached files. The AI Tutor considers the images and documents that students attach to their answers. Turning this on may increase usage costs.
Use lesson content (theory) only when needed. When this option is on, the tutor skips the published lesson theory if the assignment has a reference answer or the tutor has knowledge-base files or links. If none of these are configured, it still uses the published theory. When this option is off, available published theory is included.
Confidence Threshold. If the AI Tutor's confidence is below this threshold, the assignment is automatically sent for manual review, and the reviewer sees the tutor's suggestion as a recommendation.
Click "Save Changes" in the bar at the bottom of the page.

These controls are at the top of the page:
Enabled. If this switch is off, the AI Tutor won't review student submissions.
AI Tutor Reviews. Shows the history of assignments the AI Tutor has reviewed.
Delete (trash can icon). Deletes the agent. Results from previous reviews are kept.

Here's a closer look at each section.
Testing
Go to Settings → AI Agents. Your AI Tutor appears in the list. Click its name to open its settings.

The AI Tutor's settings page includes a Testing card where you can see how the agent reviews answers.
To test the AI Tutor, choose the following:
Select course. Choose the course you want to use for testing.
Select a lesson. Choose a suitable lesson in that course.
- Select an assignment. Choose the assignment whose answers you want the AI Tutor to review during the test.

In the Test Answer field, enter a sample answer to the assignment for the AI Tutor to review.

After you submit a test answer, the AI Tutor previews its analysis using the selected task and tutor settings. The test does not change a real student’s answer or post feedback to a student.
Based on its analysis, the AI Tutor determines whether the assignment was completed correctly, shows its confidence level as a percentage, and gives detailed feedback on how well the test answer matches the expected answer.
Use Testing to assess the suggested result, confidence and feedback. Live reviews also have safeguards for repeated rejections and unreadable attachments, so a test result does not guarantee the same automatic outcome for a real submission.

You can run the same test from the settings of a specific assignment in Memberships → Courses. Click "Test with AI Tutor".

How the AI Tutor Works
When reviewing an assignment, the AI Tutor analyzes the student’s answer and compares it with several sources:
The description of the correct answer in the assignment settings
The AI Tutor's knowledge base, if you uploaded one
The published lesson theory, according to the Use lesson content (theory) only when needed setting described below
The student's attachments, including audio messages, images, and documents
The AI Tutor considers the context of the assignment. It doesn't just mark an answer correct or incorrect; it also explains the reasoning behind its decision.
Review Mode
Review Mode decides who gives the verdict on an answer: the AI Tutor or a person on your team. You set it once for the tutor, and it applies to every course that tutor reviews. It cannot be set for a single assignment.
- Autopilot ("The tutor reviews and replies on its own"). When the tutor is confident, it sets the answer status itself and, if Leave a comment for the student is on, posts its comment under the tutor's name. When it is not confident, or a safeguard stops it, the answer stays in Pending Review for your team, and the reviewer sees the tutor's suggestion as a recommendation. Autopilot is the default, and tutors you set up earlier keep working this way.
- With Human Review ("The tutor recommends, a person confirms"). The tutor posts nothing to the student. For each answer it reviews, it prepares a recommendation: a suggested verdict and a comment draft. The answer stays in Pending Review until a team member applies or corrects the recommendation in Assignment Review. The student then sees the verdict and the comment from that team member, not from the tutor.
Choose the mode
- Go to Settings → AI Agents and click the tutor's name.
- In the General Settings card, find Review Mode below Courses and select Autopilot or With Human Review.
- Click Save Changes in the bar at the bottom of the page.

The mode applies to answers the tutor reviews after you save. Answers it has already reviewed stay as they are.
What the reviewer does in With Human Review mode
- Go to Memberships → Assignment Review. An answer the tutor has reviewed keeps the Pending Review status and shows an AI Tutor Recommendation card between the student's answer and the verdict buttons.
- Read the suggested verdict, the confidence, and the Comment for the Student draft. Click What the Tutor Checked to see the tutor's reasoning and its checklist of required results.
- Click Apply to set the verdict and send the comment to the student as your own comment. To change the verdict or the text first, click Edit, make the changes, and then click Apply.
The card, the tags that show how a recommendation was used, and rating the tutor are described in "Checking Assignments".
Keep in mind:
- While an answer waits for a reviewer, the student sees the same as with manual review: the answer is pending review.
- In With Human Review mode the Leave a comment for the student switch is hidden, because the tutor always drafts a comment for the reviewer. The comment reaches the student only when a team member applies the recommendation.
- The Confidence Threshold still matters. When the tutor's confidence is below it, the recommendation carries the note "The tutor is not confident in this verdict."
- A tutor on Autopilot shows the same card whenever it leaves the verdict to a person. When it sets the verdict itself, the answer shows a compact AI Tutor Verdict line instead.
- Applying, editing, and rating a recommendation requires the same right as giving verdicts on the Assignment Review page.
- Reviewers' corrections and ratings are saved with each review so that the tutor's instructions can be improved later. They do not change the tutor's instructions automatically.
Flexible Review Settings
You can adapt the AI Tutor to different course formats and learning models.
These settings are available:
Turn AI Tutor reviews on or off for a specific assignment

Choose all courses or only specific ones

Test the AI Tutor on individual assignments before launch

With these settings, you can roll out AI-assisted reviews gradually and monitor their quality along the way.
Instructions for the AI Tutor
You can give each AI Tutor its own instructions to define its review style and rules.
Examples include:
How the AI Tutor should communicate with students
The tone of its comments: strict, neutral, or friendly
Whether humor or informal explanations are allowed
Any other instructions

Instructions guide the review and feedback. Test representative answers and exceptions before relying on automatic grading.
AI Tutor Knowledge Base
You can upload your own knowledge base in the AI Tutor settings. The AI Tutor uses it when reviewing assignments.
Supported upload formats are:
- TXT (text files)
- DOCX
- JSON

A knowledge base supplies your terminology, methodology and review criteria. Check the tutor's interpretation during testing.
Knowledge Links. Add a full HTTP or HTTPS address for a public page and an optional title, click Add Link, then Save Changes. A new row says it will be read after saving. Check its sync status before relying on that source. If a later refresh fails, a previously synced version can remain in use; the row identifies that version. To stop using a link, remove its row and save the changes.
Keep reference materials aligned with the course assignments.
Comments for Students
After reviewing an assignment, the AI Tutor can automatically leave a detailed comment for the student that:
Points out errors
Explains why the answer was accepted or rejected
Recommends materials for the student to review

On Autopilot, feedback is posted only when Leave a comment for the student is enabled, the live result passes the automatic-review safeguards, and the tutor generates a nonempty comment. Manual-review results and Testing do not automatically post feedback to the student. When the switch is on and a live result needs manual review, the reviewer sees the drafted comment in the AI Tutor Recommendation card and can send it as their own.
In With Human Review mode this switch is not shown. The tutor always drafts a comment, and it reaches the student only when a team member applies the recommendation in Assignment Review, as that team member's own comment. See Review Mode.
Analyzing Attached Files
The AI Tutor can review assignments that include attachments such as:
Images, including screenshots and photos
Documents
Audio messages

This is especially useful for hands-on exercises, homework, and assignments with visual results.
Confidence Threshold
To help maintain review quality, you can set a minimum confidence threshold for the AI Tutor.
Here's how it works:
On Autopilot, if the AI Tutor's confidence is equal to or above the threshold, the result can be applied automatically. For example, 80% meets an 80% threshold.
If its confidence is below the threshold, the assignment is automatically sent for manual review. The reviewer sees the tutor's suggested verdict in the AI Tutor Recommendation card with the note "The tutor is not confident in this verdict."
A real submission can still require manual review at a higher confidence level. Safeguards cover repeated rejection, unreadable attachments, requests for human help, and a result that contradicts the assignment requirements identified during review. The recommendation card names the reason.
In With Human Review mode no result is applied automatically at any confidence level. The threshold only decides whether the recommendation carries the low-confidence note.

This reduces the risk of errors on complex or ambiguous assignments.
Analysis Strategy
You can choose a strategy for the AI Tutor based on your priorities:
Review speed
Quality of analysis
Processing cost

This setting works well for both small courses and large learning programs with many students.
Usage Costs
AI Tutor usage is billed based on the tokens actually used.

Token usage depends on your settings and the complexity of each review.
Review the usage cost when testing different strategies and attachment settings.
Business Impact
The amount of work the AI Tutor can handle depends on the assignments, reference answers, source materials, and review settings. Measure the results on your own course before expanding its scope.
As a result, you get:
Less repetitive manual work
Faster assignment reviews
Consistent grading standards
Use test results and manual-review outcomes to decide which assignments are suitable for automated review.
Keep ambiguous or high-impact assignments in the manual-review workflow until the results meet your course standards.
Roll out reviews for a course
- Create the AI Tutor and leave it disabled while you set the course scope, reference materials, instructions, and confidence threshold. Save Changes before Testing; save again after changing those settings.
- In Testing, select the course, lesson, and assignment. Try a correct answer, an incomplete answer, and an answer that needs human judgment.
- Review the decision, confidence, and student-facing feedback for each test. Adjust the instructions or reference answer if the result does not match the grading criteria.
- Save with Save Changes, then enable the tutor for the intended courses. Use AI Tutor Reviews to review its decisions after launch.
The confidence setting determines when an answer is sent for manual review; it does not guarantee that every automatic decision is correct.
To check the tutor on real answers before it replies to students on its own, set Review Mode to With Human Review first. Your team confirms or corrects each recommendation in Assignment Review, and you can switch to Autopilot later.