For Instructure Partners · 2026
AI Agents for Canvas, Owned by You
Three ways to add AI to Canvas by Instructure: agents that launch inside your courses over LTI 1.3, agents that author whole courses through the Canvas API, and Canvas wired into your whole-campus ontology. It all runs inside your own firewall — no student record ever leaves your perimeter — model-agnostic, no per-seat pricing, and full ownership of the code and data for maximum security.
The Short Answer
ibl.ai adds AI agents to Canvas three ways. First, agents launch inside Canvas courses over LTI 1.3 — roster-aware, grade-passback capable, and sitting beside your existing workflow, not replacing it. Second, a Course Authoring agent builds Canvas courses through the REST API — pages, assignments, quizzes, and modules assembled in dependency order and left unpublished for a human to review. Third, Canvas becomes one node in a campus-wide ontology: over an MCP interoperability layer, ibl.ai unifies Canvas with your SIS, CRM, and ERP into a single knowledge graph that every agent reasons over. Because ibl.ai is model-agnostic and you receive full source code and data ownership, the entire system runs inside your own firewall — there is no third-party cloud holding your students’ records.
LTI 1.3
Agents inside Canvas courses
One graph
Canvas unified with SIS, CRM, ERP
Any LLM
Model-agnostic, hot-swappable
You own it
Full code + data, in your firewall
Three Ways ibl.ai Works With Canvas
The same platform, met at three different layers of your campus
Application 1
Agents inside Canvas, via LTI
Your tutor, teaching assistant, and support agents launch directly inside Canvas courses over LTI 1.3 — no rip-and-replace, no new tab to teach students. The agent sees the course context and can pass grades back.
Application 2
Agents that author Canvas courses
Describe a course in plain English and an agent builds it through the Canvas REST API — pages, assignments, quizzes, modules — in dependency order and unpublished, for a human to review before students see it.
Application 3
Canvas in the campus ontology
Canvas becomes one system in a unified knowledge graph. Agents reason across the LMS and the SIS, CRM, and ERP at once — because they draw from the same ontology, not from Canvas in isolation.
The Benefits, Not Just the Features
What Canvas × ibl.ai delivers to your institution — weighed as heavily as what it does
Maximum security & control
Everything runs inside your own firewall — no third-party cloud holds your students' records, so there is nothing external to breach. Your keys, your IAM, your perimeter.
Lower cost at scale
Flat license plus actual usage — no per-seat penalty. Offer AI to every student and faculty member without the bill ballooning with headcount as enrollment grows.
No vendor lock-in
Model-agnostic by design and you own the source code. Run Claude, GPT, Gemini, or a local model and switch anytime — never trapped on one vendor's price or roadmap.
Better student outcomes
Context-rich tutoring grounded in real course and campus data means faster help, earlier at-risk flags, and fewer drop-offs — the outcome every Canvas institution is after.
A partner, not a vendor
As an official Instructure partner with source access, hard, specific requirements get built rather than declined. The relationship behaves like an extension of your team.
Future-proof by architecture
Model Canvas into the ontology once and every future agent inherits it. Adopt new model capabilities as they ship — the investment compounds instead of expiring.
Agents that launch inside your Canvas courses
ibl.ai integrates with Canvas over the standard LTI 1.3 protocol, so agents appear right where students and faculty already work. It sits beside your LMS — it does not replace it — so there is nothing to migrate and no workflow to relearn.
Standard LTI 1.3 launch
Add ibl.ai to a course the same way you add any external tool. No custom Canvas build, no fragile scraping — the certified interop path.
Roster- and course-aware
The agent knows the course, the enrolled roster, and the content it is embedded in, so answers are grounded in that specific class.
Grade passback
Agent-assisted assignments and assessments can return scores to the Canvas gradebook through the LTI Assignment & Grade Service.
Beside, not instead of
Faculty keep Canvas as the system of record. ibl.ai adds AI on top — a tutor for students, a real teaching assistant for instructors.
A Course Authoring agent that builds Canvas courses end to end
Canvas has no “create a whole course” endpoint — a course is a dozen independent resources, each with its own publish state. Describe the course you want and the agent assembles all of it through the Canvas REST API: the shell, the pages, the assignments and quizzes, the modules, and every module item.
Built in dependency order
Shell, sections, assignment groups, content, modules, then module items — so nothing is wired to an ID that doesn't exist yet.
Unpublished by default
Publishing a course with enrollments emails real people. The agent builds everything unpublished and leaves the last step to a human.
Safe to run twice
A manifest maps every object to the ID Canvas assigned it, so a re-run updates in place instead of duplicating the course — and a failed build resumes.
Revised in conversation
Ask for a longer page or a restructured module and the agent edits it in place. Reviewing a course and revising it are the same interface.
Canvas as one node in your campus knowledge graph
An agent that only sees Canvas can only answer Canvas questions. The larger win is connecting Canvas into an organizational ontology — a unified graph where Canvas data lives alongside the SIS, CRM, and ERP, so a single agent reasons across all of it with a per-learner memory.
What the ontology unifies with Canvas
LMS
Canvas, Blackboard, Brightspace, Moodle
SIS
Banner, PeopleSoft, Colleague
CRM
Slate, Salesforce Education Cloud
ERP / HR
Workday, Oracle, Dynamics
One MCP interoperability layer
ibl.ai normalizes access to Canvas and every other system over MCP-based adapters — read-only, role-scoped, and with no data extraction. Records stay where they live.
A secure, per-learner memory
Advising history, enrollments, grades, deadlines, and goals — federated from Canvas + SIS + CRM into a policy-guarded memory the agent can draw on.
Build once, every agent benefits
Model Canvas into the ontology once, and every current and future agent that touches coursework inherits it. The tenth agent is a fraction of the cost of the first.
What Your Campus Gets
Three roles, all grounded in the Canvas-connected ontology
For Students
A context-rich tutor
- Answers grounded in the actual course content in Canvas
- Proactive nudges on deadlines and study plans
- Privacy by design — fine-grained consent per interaction
For Instructors
A real teaching assistant
- Pulls rosters, outcomes, and rubrics to draft materials
- Office-hour triage and assignment hints, citing sources
- Department-level policy and oversight dashboards
For Administrators
A digital aide that sees across systems
- Reasons across Canvas + SIS + CRM in one place
- Flags at-risk students earlier with fuller context
- Runs in your environment — your VPC, your keys
It runs inside your firewall — and that is the whole point
The more student data an AI system touches, the more it matters where that data lives. With ibl.ai, the runtime, the connectors, and the memory all execute on infrastructure you own — on-prem or in your own cloud tenant. There is no third-party SaaS holding a copy of your students’ records. When the system lives behind your perimeter, there is simply no external custodian to breach.
Managed AI SaaS
- Your student data sits in a vendor's cloud
- You trust someone else's security posture
- No access to the source code running on your data
- Per-seat pricing that scales with headcount
- A breach of the vendor is a breach of you
ibl.ai — you own the stack
- Data never leaves your VPC or data center
- Your keys (KMS), your IAM, your controls
- Full source code to connectors, policy engine, agents
- Flat license + usage — no per-seat penalty
- Nothing on a third-party cloud to compromise
Your infrastructure
On-prem or your cloud (AWS, Azure, GCP). No external data dependencies.
Your keys & policy
KMS, IAM, RBAC, PII redaction, audit logs, data-residency options.
FERPA by deployment
Student records stay inside the institution's perimeter — not by promise, by architecture.
West Coast University: screen-aware AI support in production
WCU deployed ibl.ai agents that can see the user’s screen and guide students through registration, financial-aid, and learning portals in real time — on a model-agnostic platform with no per-seat pricing. The exact kind of partnership an Instructure institution can stand up.
“Add a flexible, model-agnostic platform backed by a team genuinely on the leading edge of AI, and you have a partner, not a vendor. That’s real partnership.”
Marwan Alamat
Chief Information Officer, West Coast University
The code is yours, and it runs any model
The ontology, the operating system, and the agent library are open source. Run Claude, GPT, Gemini, Llama, or a local model — and switch as capability, cost, and privacy needs change, without re-platforming.
Meeting us at the 2026 conference? See AI that keeps every record inside your firewall.
As an official Instructure partner, ibl.ai works alongside Canvas — and it runs entirely inside your own perimeter. Nothing leaves your firewall, no third-party cloud ever holds your students’ records, and you own the full source code and data. Bring us the systems your institution actually runs — Canvas plus your SIS, CRM, and warehouse — and we’ll show you agents reasoning across all of them, live, in a deployment you would own end to end. Maximum security, no per-seat pricing, your campus with AI on top.
Canvas × ibl.ai, Answered
How does ibl.ai integrate with Canvas?
Three ways. Agents launch inside Canvas courses over the standard LTI 1.3 protocol (with roster context and grade passback); a Course Authoring agent builds whole courses through the Canvas REST API; and Canvas is unified into a campus-wide ontology over an MCP interoperability layer so agents can reason across the LMS alongside your SIS, CRM, and ERP.
Can an agent build a Canvas course for us?
Yes. Describe the course in plain English and the Course Authoring agent assembles it through the Canvas REST API — the course shell, pages, assignments, quizzes, modules, and every module item — in dependency order. It builds everything unpublished and writes a manifest, so a re-run updates the course in place instead of duplicating it and publishing stays a deliberate human decision.
Does it replace Canvas?
No. ibl.ai sits beside Canvas, not instead of it. Canvas stays your system of record; ibl.ai adds AI agents on top over LTI and MCP, so there is nothing to migrate.
Where does our student data live?
Inside your own firewall. The runtime, connectors, and memory all execute on infrastructure you own — on-prem or in your cloud tenant — with your keys and controls. No third-party SaaS holds a copy of your records, which is what makes FERPA a deployment fact rather than a vendor promise.
Which AI models can we run?
Any of them. ibl.ai is model-agnostic — Claude, GPT, Gemini, Llama, or a local model — and you can switch as capability, cost, and privacy needs change without re-platforming.
Do we really own the code?
Yes. You receive full source access to the connectors, policy engine, and agent interfaces — no vendor lock-in. The ontology, OS, and agent library are open source on GitHub.
Add AI to Canvas — on infrastructure you own
Agents inside your courses, Canvas wired into your campus ontology, any LLM you choose, and full ownership of the code and data — all behind your own firewall.