The campus AI gateway
One secure AI API for learning, building, and research
Sentinel Hub sits between your students and the frontier — an OpenAI-compatible gateway that routes every request to the models your institution approves, on keys you control, with limits that enforce themselves. Compatible with modern apps. Governed by your campus.
Student dashboard · design-210
| key sent_live_9f2c… | ACTIVE · expires Dec 18 |
| model | sentinel/general-fast |
| device studio-mac-04 | CONNECTED · private network |
| Tokens this month | 412,380 / 1,000,000 |
| Requests today | 214 · 0 rejected |
| Provider credential | never leaves the gateway |
The idea
Familiar to build with. Governed by design.
Most AI tools, libraries, and course projects already speak the same dialect: OpenAI-style endpoints, a bearer key, a model name. Sentinel Hub presents exactly that interface — so the apps students already use, and the ones they build, connect with a base URL swap instead of a rewrite.
Behind the interface, your institution stays in charge. Sentinel routes each approved request to the provider you’ve configured — OpenAI, Anthropic, Gemini, local models through Ollama, or whatever comes next — without ever handing students the underlying credentials.
Every student and application gets its own Sentinel key: scoped to specific models, bound to enrolled devices, set to expire, and revocable in one click. Every request passes a server-side policy check — identity, device, model permission, rate, token, and budget — before it reaches a provider.
Students get a real production API. Your campus gets the controls. Nobody gets the master keys. Here’s the full request lifecycle.
Getting started
Connect. Create a key. Build.
Connect a device
Students enroll a laptop or workstation from the dashboard and join the private network with a one-time profile — a guided checklist, not a networking course.
Create a scoped key
Each app gets its own Sentinel key, limited to approved models, tied to enrolled devices, and stamped with an expiration date. Shown once, revocable always.
Point an app and go
Any application that accepts a custom OpenAI-compatible base URL can connect — coding tools, class chatbots, research prototypes, LMS integrations.
What students build
A production API, not a chat window
Course applications
Class chatbots, writing assistants, study helpers, and coding tutors — built against the same API patterns students will meet in industry.
Real software
Web apps, mobile apps, and desktop tools that need AI responses — with streaming, embeddings, and structured output where the model supports them.
Controlled experiments
Research prototypes on approved models with tracked usage — reproducible, budgeted, and reportable to whoever funds the work.
Security, in one breath
Private network. Scoped keys. Server-side limits.
Network access alone can’t invoke AI, and a stolen key is useless off the private path. Usage is logged as metadata only — model, tokens, cost, request ID — never prompts or responses by default.