FortaifyFor educators and institutions
Course material that stays on campus.
Faculty and staff can work with student material, source texts, and repeatable AI workflows on models that run on institutional hardware — with an explicit boundary around anything that would otherwise reach an outside provider.
Managed organization deployment and regulated workflows below describe planned capabilities. Account administration and policy authoring are implemented; client delivery, destination approvals, SIEM integration, and matter controls remain in development. Review current availability.
The operating reality
Students and faculty are already using consumer AI tools. The institutional problem is not access — it is that graded work, unpublished research, and student records reach an outside provider with no record that it happened and no rule that could have stopped it.
Student records and unpublished work
Run the work on a local model when the material belongs to a student.
Course workflows involve material an institution cannot casually export: graded submissions, letters of recommendation, admissions files, exam banks, and research that has not been published yet. Glyph runs open models directly on the machine, so that work can be evaluated without a network request at all. Cloud models remain available for the work that genuinely benefits from them — as a deliberate choice rather than the default.
What Glyph brings to it
Local models
Run open GGUF and MLX models on the desktop, or directly on iOS and Android. No network request leaves the device for a local turn.
Knowledge bases
Index a syllabus, reading list, or department handbook once and reuse it across chats, so answers cite the course material instead of the open web.
Project workspaces
Keep each course or cohort in its own workspace with its own files, chats, and personas rather than one undifferentiated history.
Signed organization policy
Administrators publish one policy; enrolled devices enforce it locally and fail closed when it is missing or expired.
A practical way to begin
Keep the first workflow narrow enough to evaluate, but real enough to matter.
- 01
Start with one course
Pick a single course with real material and a real workload. Narrow enough to evaluate honestly, real enough that the result means something.
- 02
Build its knowledge base
Index the syllabus, readings, and rubric so the assistant answers from the course rather than from the internet.
- 03
Set the destination rule
Decide which models student material may reach — often none — and publish that as policy instead of as guidance in an email.
What good looks like
- Student work evaluated without leaving institutional hardware
- One approved destination list instead of per-instructor choices
- An audit trail that records decisions, not student text
Availability
Local model workflows are available. Organization administration and policy authoring are implemented in the account control plane. Managed client rollout, destination approvals, SIEM delivery, and regulated workflows remain in development; see the Enterprise availability page.
Make education the starting point.
Tell us about the workflow and the boundary around it, and we will help scope a practical first step.