# Canvas LMS + Agentic OS Integration > Source: https://ibl.ai/resources/integrations/canvas-agentic-os *Deploy purpose-built AI agents inside Canvas LMS — on your infrastructure, under your control.* Canvas LMS powers learning for millions of students worldwide. With ibl.ai Agentic OS, institutions can layer a full AI agent platform directly onto their existing Canvas environment — no rip-and-replace required. Agentic OS connects to Canvas via LTI 1.3 and REST APIs, enabling purpose-built agents to act on real course data. Agents can surface insights, automate workflows, and personalize experiences without leaving the Canvas ecosystem. Unlike generic AI add-ons, every agent you deploy through Agentic OS runs on your own infrastructure. Your data stays yours — fully FERPA-compliant and free from vendor lock-in. ## Capabilities ### Purpose-Built Agent Deployment Deploy agents with defined roles — advisor, tutor, grader, analyst — directly inside Canvas. Each agent has a specific scope and does not behave as a generic chatbot. ### Real-Time Canvas Data Access Agents read live Canvas data including enrollments, submissions, grades, and discussion activity to make contextually accurate decisions and recommendations. ### Automated Grade Passback Agents can evaluate open-ended responses, simulations, or custom assessments and write scores directly back to the Canvas gradebook via LTI Advantage. ### Institution-Owned Infrastructure All agents, models, and learner data run on your servers. No student data is sent to ibl.ai or third-party clouds without explicit institution authorization. ### Agent Orchestration and Monitoring The Agentic OS dashboard gives administrators full visibility into agent activity, task queues, error logs, and performance metrics across all Canvas courses. ### Extensible Agent API Build custom agents using the Agentic OS SDK and connect them to Canvas data. Extend or fork existing agents to match your institution's unique workflows. ## Setup ### Step 1: Provision Agentic OS on Your Infrastructure (~2–4 hours) Deploy the ibl.ai Agentic OS stack to your cloud environment or on-premises servers. ibl.ai provides Helm charts for Kubernetes and Docker Compose configurations for rapid provisioning. Requirements: - Kubernetes cluster or Docker-compatible host - Minimum 8 vCPU, 32 GB RAM for production - Outbound network access to Canvas API endpoint ### Step 2: Register Agentic OS as an LTI 1.3 Developer Key in Canvas (~30 minutes) In Canvas Admin, create a new LTI Developer Key using the JSON configuration provided by Agentic OS. Enable the key and note the Client ID for the next step. Requirements: - Canvas Admin account - Agentic OS LTI configuration JSON (provided post-provisioning) - Canvas instance URL ### Step 3: Configure Canvas API OAuth Credentials in Agentic OS (~20 minutes) Generate a Canvas API token or configure OAuth 2.0 client credentials. Enter these into the Agentic OS admin console to authorize agent read/write access to Canvas data. Requirements: - Canvas API token with appropriate scopes - Agentic OS admin console access ### Step 4: Connect Your Identity Provider for SSO (~1 hour) Link your institution's SSO (SAML 2.0 or OIDC) to Agentic OS so that Canvas user sessions map seamlessly to agent contexts without requiring separate logins. Requirements: - SAML 2.0 or OIDC-compatible IdP (e.g., Shibboleth, Azure AD, Okta) - IdP metadata or OIDC discovery URL ### Step 5: Select and Configure AI Agents (~1–3 hours per agent) From the Agentic OS agent marketplace, choose purpose-built agents for your use cases — such as a course advisor agent or an at-risk learner detection agent. Configure each agent's role, data access scope, and behavior policies. Requirements: - Defined use cases and agent roles - LLM endpoint configured (institution-hosted or approved provider) ### Step 6: Deploy Agents to Canvas Courses and Test (~2–4 hours) Add the Agentic OS LTI tool to target Canvas courses or at the account level. Run end-to-end tests with pilot users to validate data flows, agent responses, and grade passback accuracy before full rollout. Requirements: - Pilot course or sandbox Canvas environment - Test user accounts across student, instructor, and admin roles ## Technical Requirements **Canvas LMS Requirements:** - Canvas LMS (Cloud or self-hosted) version 2022 or later - Canvas Admin access to create LTI Developer Keys - LTI 1.3 / LTI Advantage enabled on Canvas instance - Canvas REST API access with scopes: courses, enrollments, assignments, submissions, grades - Canvas Data 2 or Data Streaming enabled (optional, for batch analytics) **Agentic OS Infrastructure Requirements:** - Kubernetes 1.25+ or Docker Engine 24+ on institution-managed infrastructure - Minimum 8 vCPU, 32 GB RAM, 200 GB SSD for production deployment - PostgreSQL 14+ for agent state and interaction storage - Vector database (pgvector, Weaviate, or Qdrant) for semantic agent memory - LLM endpoint: institution-hosted (Ollama, vLLM) or approved API (Azure OpenAI, AWS Bedrock) - TLS-terminated reverse proxy (NGINX or Traefik) with valid SSL certificate **Security and Compliance Requirements:** - SAML 2.0 or OIDC identity provider for SSO integration - Network firewall rules permitting Canvas-to-Agentic OS HTTPS traffic - Data residency controls configured to keep PII within institution boundaries - Role-based access control (RBAC) policies defined for agent data scopes - Audit logging enabled on Agentic OS for FERPA compliance documentation ## FAQ **Q: Does the Canvas + Agentic OS integration require replacing our existing Canvas setup?** No. Agentic OS integrates with your existing Canvas environment via LTI 1.3 and the Canvas REST API. There is no migration required. Your courses, data, and workflows remain in Canvas — Agentic OS adds an AI agent layer on top. **Q: Where does student data go when agents process Canvas information?** All data stays on your institution's infrastructure. Agentic OS is deployed on your servers, and no learner data is transmitted to ibl.ai or external clouds unless your institution explicitly configures an external LLM endpoint. The integration is designed for FERPA compliance by default. **Q: What is the difference between Agentic OS and a generic AI chatbot added to Canvas?** Generic chatbots have no defined role or institutional context. Agentic OS agents are purpose-built with specific roles — advisor, grader, analyst — and operate on real Canvas data. They follow defined behavior policies and are scoped to authorized tasks, not open-ended conversation. **Q: Can we build custom agents for our institution's specific Canvas workflows?** Yes. The Agentic OS SDK allows your development team to build, fork, or extend agents. Custom agents can connect to Canvas data, internal SIS systems, and other institutional tools. ibl.ai also offers professional services for custom agent development. **Q: How long does it take to go from zero to live agents in Canvas?** A standard deployment with pre-built agents typically takes 1–2 weeks, including infrastructure provisioning, LTI configuration, SSO setup, and pilot testing. Custom agent development timelines vary based on complexity. **Q: Is the integration compatible with Canvas Data 2 and Canvas Data Streaming?** Yes. Agentic OS can ingest Canvas Data 2 exports and streaming event data for batch analytics use cases. This enables richer longitudinal analysis beyond what the real-time REST API provides, such as cohort-level curriculum analytics. **Q: What compliance certifications does ibl.ai Agentic OS support?** Agentic OS is designed to support FERPA, HIPAA, and SOC 2 compliance requirements. Because the platform runs on institution-owned infrastructure, compliance controls — data residency, access logging, encryption — are managed by the institution with ibl.ai's architectural guidance. **Q: Can Agentic OS agents write back to Canvas, such as posting grades or announcements?** Yes. Agents can perform write operations in Canvas including LTI Advantage grade passback, posting announcements, sending Canvas inbox messages, and creating calendar events — subject to the permission scopes configured during setup.