The Short Answer
FERPA-compliant AI for higher education is a deployment fact, not a vendor promise: the only way student records stay inside the institution's perimeter during an inference call is if the AI runtime executes there. ibl.ai runs inside the campus VPC alongside Banner, PeopleSoft, or Workday Student and Canvas, Blackboard, or Moodle — you own all the code and the data, run it model-agnostic across any LLM, and deploy anywhere.
Because the runtime is co-located with the SIS and LMS, FERPA-protected records — transcripts, financial aid files, advising notes, IEP documentation — never traverse a third-party AI vendor's cloud. That answers the one question a general counsel cannot get a managed vendor past: where the data physically sits mid-call. 1.6M+ users across 400+ organizations run the platform this way, and ibl.ai is family-owned and operated from New York, NY.
What FERPA Compliance Actually Requires of AI
FERPA — the Family Educational Rights and Privacy Act — restricts the disclosure of student records to third parties without consent. Three structural questions every institution's general counsel asks of any AI vendor:
- Where do student records live during the AI inference call?
- Who has access to logs and intermediate state?
- What contractual + technical controls prevent the vendor from using student data for model training, evaluation, or quality-improvement?
A managed AI vendor can answer (3) with a strong DPA. They can answer (2) with role-based access controls. They can't answer (1) without the data physically transiting their cloud. Self-hosted on the institution's infrastructure makes question 1 a no — the records never leave.
How ibl.ai Ships FERPA-by-Deployment
The agent runtime executes inside the campus VPC. Same network as Banner / PeopleSoft / Workday Student (SIS), Canvas / Blackboard / Moodle / D2L (LMS), Slate / Salesforce Education Cloud / EAB Navigate (CRM).
Integrations terminate inside the campus. When the AI agent needs to pull a student's degree audit from Banner or check current LTI course progress in Canvas, the connector executes inside the same VPC. The API calls don't traverse a vendor's cloud.
The model can run on campus infrastructure. Open-weight models (Llama 4, DeepSeek-R1, Qwen 3 for multilingual) execute on the institution's GPU. For workloads that need frontier models (Claude Opus, GPT-5, Gemini Pro), API calls route through a campus-controlled proxy that enforces data residency and logs every request to the campus SIEM.
The control plane sees orchestration metadata, not student records. The Ed25519-signed WebSocket between the campus-hosted runtime and the ibl.ai platform carries which-agent-which-skill-which-model-class metadata. Student data never crosses that boundary.
For the full FERPA-aligned architecture (Banner / PeopleSoft / Workday Student + LMS via LTI 1.3 + APIs + MCP), see Higher Education AI Reference Architecture on ibl.ai.
Workloads Where FERPA Matters Most
Three classes of campus AI workload where the FERPA-by-deployment story is non-negotiable:
Academic advising. Every advising conversation contains FERPA-scope data — degree audit, registration status, GPA, financial-aid scenarios. Conversation transcripts are FERPA-protected student records. Self-hosted means the transcripts stay on the campus's SIS-adjacent infrastructure.
For the per-conversation cost math + vendor comparison: What AI Academic Advising Actually Costs in 2026.
Tutoring. Tutoring session logs contain FERPA-scope student-performance data — what the student struggled with, what accommodations were used, what the agent observed. Districts and campuses serving multilingual learners need locally-controlled language support (Qwen 3 for Spanish/Mandarin/Arabic).
For the cost math: What AI Tutoring Actually Costs in 2026 (K-12 + Higher Ed).
Financial-aid agents. FAFSA scenarios, aid-package decisions, family-income context. All FERPA-scope. The institution's general counsel reviews where this data lives before any AI deployment.
The Cost Math at Campus Scale
A 30,000-student university running advising + tutoring + course-content generation (~89M input + 120M output tokens/month):
| Approach | Monthly cost | Student-data location |
|---|---|---|
| ChatGPT Enterprise ($60 × 33K) | $1,980,000 | OpenAI cloud |
| ChatGPT Edu (~$25 × 33K) | $825,000 | OpenAI cloud |
| Microsoft 365 Copilot Edu ($30 × 33K) | $990,000 | Microsoft cloud |
| Direct Claude Sonnet API | ~$2,067 | Anthropic cloud |
| ibl.ai self-hosted (Llama 4 / Qwen 3) | ~$5,000–10,000 | Inside campus VPC |
ibl.ai self-hosted is ~100× cheaper than ChatGPT Edu for the same workload, with FERPA-protected records inside the institution's network.
For the segment cost math: AI Cost Math for Higher Education: Per-Seat vs Usage-Based in 2026.
FERPA Posture Differences That Matter
| Managed AI vendor (DPA) | ibl.ai self-hosted | |
|---|---|---|
| Student-record location during inference | Vendor cloud | Inside campus VPC |
| FERPA DPA scope | Renewed annually | None needed for the runtime |
| Audit log location | Vendor's infrastructure | Campus SIEM |
| Sub-processor changes | Trigger DPA review | N/A |
| Model swap | Vendor approval cycle | Config change inside campus |
| Multilingual model choice | Vendor's selection | Campus's choice (Qwen 3 for Spanish, etc.) |
| Air-gapped option (for special programs) | Rarely | Fully supported |
Deployment Tiers
Managed VPC — campus's existing AWS / Azure / GCP environment. Same VPC as SIS / LMS. Fastest path; suits 80% of campus workloads.
On-premise — campus data center (some R1 institutions with significant on-prem infrastructure prefer this).
Hybrid — Managed VPC for general faculty pilot + on-premise for institutional production. See Higher Ed AI Blueprint: Hybrid Rollout for FERPA Campuses.
Run the Numbers
- Higher Education AI Reference Architecture — full FERPA-by-design architecture (mirrors Syracuse + SUNY deployments)
- AI Cost Math for Higher Education — segment cost math
- What AI Academic Advising Actually Costs in 2026 — per-conversation math
- What AI Tutoring Actually Costs in 2026 — per-session math
- Self-Hosted AI vs ChatGPT Enterprise for Higher Education — deployment comparison
- Higher Ed AI Blueprint: Hybrid Rollout for FERPA Campuses — staged deployment recipe
- AI and FERPA Compliance: What Higher Ed Needs to Know — broader FERPA framework
- COPPA-Compliant AI for Schools — the parallel under-13 data-privacy regime for K-12
- Self-Hosted AI for Universities — the deployment-focused companion to this page
- The Complete Guide to AI Agents for Universities — the full catalog of campus agents
Why Family-Owned and New York Matters Here
A university's AI vendor relationship for workloads as central as advising and tutoring is a multi-year commitment that touches FERPA-protected records and student-success outcomes accreditors scrutinize. ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned, long-term partner with a perpetual platform license and no investor exit pressure. The runtime is open source. The FERPA-protected records stay inside the campus VPC. The math works at a 2,000-student community college or a 200,000-student multi-campus system like SUNY.
FERPA-compliant AI isn't an enterprise SKU. It's the architecture.
Related: ibl.ai on AWS: Seamless Integration with Bedrock, SageMaker, and the AWS Gen AI Stack
Related: Microsoft Fabric + ibl.ai: Unified Data Analytics Meets AI Tutoring via MCP