# AI Sovereignty at Syracuse University > Source: https://ibl.ai/case-study/syracuse-university ## Case Study How Syracuse University deployed a full-stack AI platform it fully owns and controls — achieving deep integration with campus systems, complete data sovereignty, and dramatically lower costs compared to per-seat SaaS alternatives. [See Syracuse's AI Tools page](https://ai.syracuse.edu/ai-tools.html) | [Read Syracuse's own write-up](https://ai.syracuse.edu/stories/ibl-google-cloud.html) Published by Syracuse University ITS at ai.syracuse.edu > "We wanted full ownership of our AI platform — the code, the data, the infrastructure. ibl.ai delivered exactly that. We're running on our own GCP, with our own models, fully integrated into our campus systems." > > — Jeff Rubin, Senior Vice President & Chief Digital Officer, Syracuse University [Watch the interview](https://www.linkedin.com/feed/update/urn:li:share:7399157983726706693/) ### Quick Stats - **30,000+** Students served - **85%** Lower AI costs - **100%** Code ownership - **Any LLM** Model flexibility --- ## The Challenge ### Why "just buy a SaaS" wasn't good enough **Vendor lock-in risk** — SaaS providers can change terms, raise prices, or shut down features overnight. A university serving 30,000 students cannot afford that dependency for a critical academic resource. **Data sovereignty** — Student data, research, and institutional knowledge cannot live on a third-party vendor's infrastructure. FERPA compliance demands control over where data resides and who can access it. **No real customization** — Off-the-shelf AI tools offer one-size-fits-all experiences. Syracuse needed agents integrated with its specific SIS, LMS, SSO, and RBAC systems — not a generic chatbot. --- ## AI Sovereignty ### Your code. Your data. Your infrastructure. Syracuse University received the complete ibl.ai source code with a perpetual license, deployed on its own infrastructure. No black boxes, no API keys pointing to someone else's servers, no exit fees if priorities change. **Deployed on Syracuse's own Google Cloud Platform environment.** The entire ibl.ai stack runs within the university's GCP project — data never leaves infrastructure that Syracuse controls, and the university's cloud team manages access, networking, and compliance just like any other institutional system. #### Typical SaaS AI - Vendor owns the code and can change it - Data stored on vendor infrastructure - Locked into vendor's LLM choice - Features disappear if vendor pivots - Per-seat pricing punishes growth - Exit = data migration nightmare #### ibl.ai at Syracuse - Full source code with perpetual license - Data stays on university infrastructure - Connect any LLM — swap freely as pricing changes - University controls the roadmap - Flat-rate pricing for unlimited users - If you part ways, you keep everything **Build vs. Buy resolved:** Syracuse didn't have to choose between building from scratch (12–24 months, specialized AI team) or renting a black-box SaaS. ibl.ai delivered a production-ready platform in weeks with full source code — the speed of buying with the control of building. [Read the full Build vs. Buy analysis](/build-vs-buy) --- ## Deep Customization ### Wired into the campus ecosystem Generic AI chatbots sit outside your systems. Syracuse's deployment integrates directly with the institutional technology stack, giving AI agents access to the context they need to be genuinely useful. **SSO & Identity** — Native integration with university SSO (Shibboleth, SAML, CAS). Faculty, students, and staff authenticate once — no separate AI login. **RBAC & Permissions** — Role-based access maps to existing university roles. Department chairs, advisors, and students each see agents and data appropriate to their role. **Custom UI/UX** — The interface matches Syracuse branding and UX standards. Not a white-labeled vendor portal — a product that looks and feels like it belongs. --- ## In Production ### Clementine: the platform students actually use Syracuse runs its ibl.ai deployment under its own name — **Clementine**, the university's private AI platform for teaching, learning, and AI innovation. It sits alongside Claude Enterprise, Microsoft Copilot, and Google Gemini on Syracuse's list of University-approved AI, and it is the only one the university owns outright. Its first campus-wide agent, **Clementine Class Search**, launched in Spring 2026 and landed straight into Fall 2026 registration — the single heaviest week of the academic year for advising staff. #### First 37 days (March 25 – April 30, 2026) - **2,038** Students in the first 37 days - **8,217** Questions asked - **2,401** Conversations - **56** Turns in the longest session **Demand spikes on registration day** — The five-day undergraduate registration window (April 9–15, 2026) drove 41% of all questions in the 37-day period. On April 14 alone, students sent 1,055 questions across 313 conversations. **37% of usage is after hours** — More than a third of all activity happened outside weekday business hours — late nights and weekends, when advising offices are closed and students are staring at a draft schedule. **Two-thirds are multi-turn** — Two out of three conversations went past a single question, and students returned days later to keep refining — treating the agent as a planning partner rather than a search box. **Students asked in their own words** — no query syntax, no catalog cross-referencing. Real questions from the first registration season: > "what 3 credit art classes or classes about fashion can I take as an elective as a sociology major that are on monday wednesdays at 2:15 fall 2026" > "no 8am and nothing past 5pm, and no fridays" > "Build a 14 credit schedule, avoiding Fridays" > "classes that align with someone who wants to go to law school" **What ownership buys next:** The clearest signal from the first season was students asking degree-path questions — "does this count toward my requirement," "what should I take next." Because Syracuse owns the platform, making the agent student-aware is a roadmap decision the university makes on its own timeline, connecting the agent to each student's degree audit. On a SaaS product, that is a feature request. [Read Syracuse's full report on Clementine's first registration season](https://ai.syracuse.edu/stories/clementine.html) --- ## Use Cases ### What faculty, students, and staff build on it Owning the platform means every Syracuse student, faculty, and staff member can build agents on it — not just consume a vendor's fixed feature set. Each agent is grounded in the material its author uploads, so answers cite real sources instead of inventing them. **A course agent built from a professor's own lectures** — Jeff Rubin loaded every lecture transcript and slide from his 200-student intro IT course into an agent instructed to answer from his materials first and reach for a general model only when a student needs more. Seeing the questions students ask tells him where they get stuck. **Practice exams generated from the lectures** — The platform generates practice exams straight from course lecture content, so the questions test what was actually taught rather than a generic question bank. **Branching "pick-your-path" simulations** — Faculty build scenarios that respond to a student's answer and push deeper, surfacing how students actually reason. It is the one-on-one attention a 200-person lecture could never offer — critical thinking, at scale. **Student-built study agents** — Students create a tutor around their own courses, textbooks, and goals, then work with it at whatever hour they actually study. It follows their pace, gives targeted feedback, and pulls together practice questions — grounded in course material, with citations back to the source. **A teaching assistant that scales** — Faculty and staff hand repeat questions and course FAQs to an agent built on approved material, keeping full oversight of what goes out. Lesson plans, quizzes, and first-pass feedback draft in minutes. **Embedded agents and an institutional API** — Finished agents embed directly in a campus webpage, and the platform exposes an API for working with agents and their datasets in the university's own code — so campus teams build on top of it rather than around it. **Grounded in whatever the campus already has** — Agents read PDFs, Word documents, PowerPoint decks, spreadsheets, and plain text, or connect Google Drive, OneDrive, Dropbox, a website, a GitHub repository, or a captioned video. **Guardrails set by the institution** — System prompts define purpose and voice. Moderation settings filter inappropriate input and stop agents from wandering into medical, legal, or financial advice, redirecting to the right campus resource instead. Authors control whether an agent is visible to students, administrators, or anyone, and whether a Syracuse sign-in is required. **Model-agnostic, in practice:** Every agent author at Syracuse picks the model behind their agent — Claude, ChatGPT, or Gemini — and switches at any time. An agent built for long-form writing feedback need not use the same model as one built for data questions. The platform runs on Google Vertex AI inside Syracuse's own Google Cloud, and the university can connect any LLM and swap freely as pricing changes. This is what model-agnostic looks like when you own all the code and the data: the choice belongs to the institution, not the vendor. [See how Syracuse documents building an agent](https://ai.syracuse.edu/clementine.html) --- ## Cost Savings ### 85% lower cost at scale Per-seat SaaS pricing was designed for small teams, not universities. At $20/user/month, 30,000 students means $600,000/month — $7.2M/year — before a single customization. With ibl.ai, Syracuse pays only for actual LLM token usage. #### Example: 30,000 students, 5 messages/day | Approach | Monthly Cost | Annual Cost | |----------|-------------|-------------| | SaaS AI ($20/seat) | $600,000 | $7,200,000 | | ibl.ai + Claude Sonnet 4.6 | ~$81,000 | ~$986,000 | | ibl.ai + GPT-4.1 Mini | ~$7,200 | ~$87,600 | | ibl.ai + Gemini 2.5 Flash | ~$10,080 | ~$122,640 | [Full LLM price calculator](/llm-price-calculator) --- ## Comparison: Build vs. Buy vs. ibl.ai | Dimension | Build In-House | Buy SaaS | ibl.ai | |-----------|---------------|----------|--------| | Time to production | 12–24 months | 2–4 weeks | 2–4 weeks | | Code ownership | Yes | No | Yes | | LLM flexibility | If you build it | Vendor-locked | Any LLM | | Customization | Unlimited | Limited | Unlimited | | Maintenance burden | 100% yours | 0% yours | Shared | | Scaling cost | Infrastructure only | $20–60/user/mo | Flat rate | | Data sovereignty | Full | Vendor-dependent | Full | | Integration depth | Whatever you build | What vendor offers | Deep & native | | Vendor dependency | None | Total | None | | Pre-built agents | 0 | Generic | 160+ | [Explore the full Build vs. Buy analysis](/build-vs-buy) --- ## For University Leadership ### Strategic implications for the C-suite AI is becoming core infrastructure for universities — not a nice-to-have tool. The decisions made today about ownership, integration, and cost structure will compound for years. **No exit fees, no migration projects** — If your institution changes direction, you keep the code, the data, and the integrations. There is no vendor to negotiate an exit from because you own everything from day one. **Your team builds capability, not dependency** — Licensing a SaaS AI tool trains staff to use a product. Working with ibl.ai trains your team to build and operate AI systems. When the engagement ends, institutional capability stays. **Costs stay flat as enrollment grows** — Per-seat pricing punishes success. The 30,001st student costs the same as the first under ibl.ai's model. Scale AI across new programs, departments, and use cases without financial anxiety. **Competitive differentiation** — A university with its own AI infrastructure can offer experiences that competitors using off-the-shelf tools cannot match. Custom agents, integrated workflows, and institutional knowledge become a moat. --- ## In Syracuse's Own Words ### Read the university's own account Syracuse University ITS publishes its AI program at [ai.syracuse.edu](https://ai.syracuse.edu/) — the tools it approves, how it built them, and what the campus does with them. Every number and quote on this page comes from those pages or from Jeff Rubin directly. **[AI Tools at Syracuse](https://ai.syracuse.edu/ai-tools.html)** — The four University-approved AI platforms available to every student, faculty, and staff member — Clementine (by ibl.ai) alongside Claude Enterprise, Copilot, and Gemini. **[Owning Its AI: Syracuse, ibl.ai and Google Cloud](https://ai.syracuse.edu/stories/ibl-google-cloud.html)** — Why Syracuse bought the source code instead of renting a platform, how it runs on the university's own Google Cloud, and what 85% lower cost looks like in practice. **[Clementine — Syracuse's own AI platform](https://ai.syracuse.edu/clementine.html)** — The campus-facing documentation: what you can build, how to ground an agent in your material, choosing the model, setting guardrails, and controlling who sees it. **[Clementine Takes On Registration Season](https://ai.syracuse.edu/stories/clementine.html)** — Syracuse's usage report from the Fall 2026 registration window: 2,038 students, 8,217 questions, and 37% of it outside business hours. --- ## Get Started Deploy a production-ready AI platform on your infrastructure in weeks — with full source code, unlimited users, and a team that works alongside yours. [Try ibl.ai](/contact) | [Calculate your savings](/llm-price-calculator)