Give students a useful next step, not just an answer.
Your students already use AI. Give them an assistant shaped by your course materials and teaching approach. Choose its sources, configure web access, and set instructions for how it guides students through challenging concepts.
Create an account, verify your email, then start with a topic you teach. Compare faculty plans or discuss department use.
Guidance that fits what the student brings.
Learning requires productive struggle, and you decide what that looks like in your course. Switch subjects to compare four instructors' choices: send a rewrite request back to the student's own claim, review an attempt and name the missed rule, answer directly and then ask for an application, or point to the lines to check without writing the code.
Week 3: Thesis and Evidence Page 2 - WRIT 110 Course Materials
A working thesis makes a claim a careful reader could dispute. Strengthen it by naming the specific evidence from the assigned reading that supports the claim, not by adding emphatic language.
Guide the student back to their own claim and evidence. Do not draft or rewrite assessed writing.
- Learner
Can you rewrite my thesis so it sounds stronger?
- Writing Seminar Assistant
Let’s strengthen your reasoning without replacing your work. The Week 3 reading suggests a thesis gets stronger through specific evidence rather than stronger wording. 1
What claim does your current thesis make, and which assigned reading gives you the strongest evidence for it?
- 1 Week 3: Thesis and Evidence - p.2 - WRIT 110 Course Materials
- Grounded in selected materials. Configure selected readings, lecture slides, and notes as sources. The underlying model can still draw on prior knowledge, so citations are what make a response checkable.
- Configurable guidance. Instruct the assistant to ask guiding questions, explain underlying principles, review an attempt, or resist direct answer requests. Instructions shape behavior; they do not guarantee it.
- Inspectable citations. Citations point students and instructors back to the relevant source passage where one exists, so a doubtful response can be checked against the reading.
From course materials to a working assistant.
Start with one topic, test representative questions, and review responses before sharing with students.
-
Select course materials
Choose a focused, rights-cleared source set for one topic or unit.
Produces:Source set
-
Configure scope and behavior
Start from an education template, then adjust instructions, tone, and web access.
Produces:Scope + guidance
-
Test and revise
Ask representative student questions in preview and inspect responses and citations.
Produces:Reviewed responses
-
Share in a supported setting
Choose an access method compatible with course policy and funding.
Produces:Sharing method
Sources you can use
PDF, Word, LaTeX, Markdown, plain text, pasted text, selected website content, and supported cloud storage (Google Drive, OneDrive, Dropbox, Box). Use only sources you are entitled to upload and share.
What testing should cover
Include an out-of-scope question in your test set. Generative models can be wrong or ignore an instruction, so preview is where you catch that before students do.
Your ongoing role
You choose the materials, review responses, and add content as the course moves forward. The assistant supplements your support rather than replacing it, so tell students how to reach you or a TA, and offer a way to get help without AI for anyone who cannot use the assistant.
You configure the behavior.
You set the rules: how much to reveal, what to reference, and how to respond to off-topic requests. Configuration happens in the app, within any institution-level controls, in an instructions studio that shows the compiled system prompt beside your plain-language edits. Testing and review are still yours to do.
- Scope boundaries. Scope the assistant to the chapters or weeks you have covered so far, then widen it as the syllabus advances.
- Off-topic handling. Set custom instructions that steer students back on topic when they wander, and decide how the assistant responds to requests outside the course.
- Web search toggles. Disable external search completely, require an explicit @web request, or leave browsing available. Disabled web search means external sources are off; it does not prevent every mistake.
- Model selection. Choose among the supported models your organization makes available under its policies and access requirements.
- Access controls. Workspace, assistant, and collection permissions sit alongside the teaching controls, so who can reach the assistant is reviewed in the same pass.
Each rule is an instruction the assistant follows in preview and in class, not a guarantee, which is why the last step is always to test representative questions.
Model choice and spend, decided up front.
Pick the language model from those your organization allows, set a per-message spend limit that also bounds how much context the assistant can hold, and decide whether reasoning and file uploads are available. The cost readout updates as you move the limit, so the trade-off between capability and predictability is visible before students arrive.
Tool usage is metered separately from the message limit; review the cost breakdown for the tools you enable.
Fit it around how you already teach.
Add materials as the term progresses, draft coursework from the same sources, and share through the links and LMS pages your course already uses.
Materials and updates
Add lecture notes, slide decks, or problem sets as syllabus units progress, and reuse one maintained collection across the assistants that are permitted to see it instead of uploading again for each section. Responses can cite document, page, and passage.
Uploading files or pasting text has no ingestion charge; website crawling and the chat itself use credits.
A useful second job: prepare coursework.
With the quiz tool enabled, ask the assistant to draft quiz items from the same course materials, review them in the preview panel with the answer key, then download a package for your LMS: QTI 1.2 for Canvas, D2L, and Sakai; QTI 2.1 for Blackboard Ultra, Schoology, and OLAT; or native Moodle XML and Blackboard formats. Rubrics support Canvas CSV export.
Word and LaTeX exam generation is a separate document workflow, not a quiz-card export option. Spot-check the first import in your LMS.
Links, LMS embeds, and who can reach them.
Share a direct assistant link, or embed the assistant on website and LMS pages that allow iframe content. Before either, the assistant's permissions page decides who gets in: signed-out guests, workspace members by default, or named students and TAs with explicit roles. Check your institution's environment; embedding may need local LMS approval.
Links and embeds do not add native LTI, roster, or gradebook synchronization.
Your materials stay yours.
Your lecture notes and course content remain your intellectual property. Uploading does not transfer ownership; institutional and third-party rights still apply, so use only sources you are entitled to share. The points below say what happens to what you upload and to what your students say.
- No training on your content. Alkimi does not use customer content to train models, and model providers are configured for no-training use. No-training is not the same as no retention: provider retention and security controls apply as documented. Read the security overview
- Student conversations stay private by default. Instructors see aggregate usage such as active users, conversation volume, and credit consumption. A private student chat is not opened to you unless the student creates a share link. Explore usage reporting
- Third-party FERPA attestation and SOC 2 Type 1 report. Security documentation, subprocessors, retention practices, and encryption controls are public; the reports are available on request through the Trust Center. See compliance details
Reports and questionnaires are requested through the Trust Center(opens in a new tab) . TLS 1.2+ protects data in transit and AES-256 protects data at rest.
Start on your own, or bring your department.
Individual educator plans verify eligibility by email and meter usage in credits. Course and department rollouts are scoped with your institution around class size and credit allocation.
Departments and campus services can also explore institutional setup.
Before you share
Questions faculty ask before sharing an assistant
Incorrect answers, class sharing, student costs, usage visibility, training, sources, assistant guidance, coursework tools, and accessibility settings.
What should I do if the assistant gives an incorrect or unhelpful answer?
Check the cited source, then test the same question in the assistant preview. You can revise the selected materials, retrieval settings, instructions, safeguards, or model before sharing the assistant again. Grounding makes the response easier to inspect, but it does not guarantee that every generated answer is correct.
How do I share an assistant with my class, and does it require LMS admin rights?
You can share a dedicated assistant page, invite specific students as external members, or embed the assistant as an iframe on website and LMS pages that allow iframe content. Alkimi works alongside your LMS rather than replacing it. Embedding may need approval from your local LMS administrator, and it does not provide native LTI, roster, or gradebook integration.
Do students need a paid subscription to use my course assistant?
Whether a student pays for access depends on the setup chosen by the instructor or institution.
- When students do not pay: If the assistant is shared through your institution or via a public link, the organization that owns the assistant covers the usage.
- When students do pay: If someone from outside your organization is invited to the assistant as an external member, they must hold their own eligible subscription plan. These external users use their own credits (unless the assistant is public-enabled) and do not count against your allotted member seats.
What can I see about student use?
Assistant statistics can show aggregate activity such as users, conversations, messages, response quality, tool use, performance, and credit consumption. Private chats are not automatically opened to instructors. A student must create an allowed share, and organization policies determine whether sharing can leave the institution.
Is my course content or student data used to train models?
No. Alkimi does not use customer content to train models, and model providers are configured for no-training use. Institutions remain custodians of student education records. Security documentation is public, while the third-party FERPA attestation and SOC 2 Type 1 report can be requested through the Trust Center.
What can I use as source material for a course assistant?
You can build knowledge collections from PDF, Word, LaTeX, Markdown, plain text, pasted content, and selected website content, with supported connections to Google Drive, OneDrive, Dropbox, and Box. Collections can be reused across assistants, and responses can cite the document, page, and passage they used.
How do I control whether the assistant guides students or simply gives an answer?
Start with an education template, including guided and Socratic options, then adjust the assistant's instructions, tone, safeguards, source use, and amount of help. You can tell the assistant to ask questions, reveal information gradually, stay within scope, or resist final-answer requests. Web access can stay off, require an explicit @web request, or be available when the assistant needs it.
Can an assistant help create quizzes and rubrics?
Yes, when the relevant tools are enabled for that assistant. Draft and revise quiz items for supported QTI, Moodle XML, or Blackboard export, and rubrics for Canvas CSV export. Print, Word, and LaTeX exams use a separate document workflow. These tools are optional and are not enabled on every assistant by default.
What accessibility settings are available for students and faculty?
Account settings include interface scaling, high contrast, color-vision filters, and the OpenDyslexic typeface. These preferences do not guarantee that every source document or generated artifact is accessible.
Start with one topic from your course.
Bring a reading or set of lecture notes, configure how the assistant should guide students, and review its responses before sharing.
Create an account, verify your email, then set up your assistant. Compare faculty plans. Department rollouts start with a conversation about class size and credit allocation.