# AI Agents for Canvas, Owned by You

**For Instructure Partners · 2026**

**✓ Official Instructure Partner**

Canvas × ibl.ai

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**.

[Book a Live Demo](https://cal.com/iblai/30min) · [Browse Agents](/solutions/higher-education#agents)

## 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Canvas courses through the REST API** — pages, assignments, quizzes, and modules assembled in dependency order and left unpublished for a human to review.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.

## At a glance

- **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

## How It Works: 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and an agent builds it through the Canvas REST API — pages, assignments, quizzes, modules — in dependency order and unpublished, for a human to review before students see it.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.

## Why It Matters: 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.

The features get you in the door. Ownership, security, cost, and outcomes are why institutions stay.

## Application 1 · Inside Canvas: 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.

How it fits: the LMS stays the LMS. [Read how ibl.ai sits beside your LMS](/blog/ibl-ai-with-your-lms-sits-beside-not-instead-of).

## Application 2 · Course Authoring: 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.

See it build a course: the walkthrough, the build order, and the full skill file the agent reads. [Read the Course Authoring agent update](/updates/canvas-course-builder-agent).

## Application 3 · The Ontology: 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](/product/agentic-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-user 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-user 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.

Learn more: [Agentic OS for Higher Education](/product/agentic-os/higher-education) · [AI Data Unification for Higher Education](/service/ai-transformation/higher-education)

## The Agents: 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

## Security & Ownership: 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.

Ownership is the durable differentiator. Any vendor can match a feature. Only ibl.ai hands you the full source code and keeps every record inside your firewall. [See the full-code license](/full-code-license).

## Proof · Case Study: 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

[Read the West Coast University story](/case-study/west-coast-university) · [More customer stories](/case-studies)

## Open Source & Model-Agnostic: 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.

- [github.com/iblai/ontology](https://github.com/iblai/ontology) — The knowledge-graph layer Canvas connects into
- [github.com/iblai/os](https://github.com/iblai/os) — The Agentic OS — connectors, memory, policy engine
- [github.com/iblai/claws](https://github.com/iblai/claws) — The library of deployable campus agents
- [github.com/iblai/base](https://github.com/iblai/base) — Build and customize agents visually

## For Instructure Partners: 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.

[Book a Live Demo](https://cal.com/iblai/30min) · [Reach out to arrange a walkthrough](/contact)

## Questions: 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.

## Get Started: 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.

[Book a Demo](https://cal.com/iblai/30min) · [Contact the team](/contact) · [Explore Agentic OS](/product/agentic-os/higher-education)
