Deployment
Draw the control boundary where your governance requires it.
The same platform, three control models. Your institution decides how much of the stack sits inside its own walls — and can change that decision later without re-training a single user.
Private Cloud
A dedicated tenancy in India, managed by Merzal Labs.
Best for
Institutions that want a live pilot this term
- No shared infrastructure with other institutions
- Managed patching, monitoring and backups
- Live in weeks, not quarters
- Clean migration path to on-premise later
Hybrid
Knowledge, documents and retrieval stay on campus; elastic capacity handles peaks.
Best for
Most engineering colleges
- Institutional index never leaves the campus
- Burst capacity for exam and admission peaks
- Lower on-campus hardware requirement
- Recommended starting architecture
On-Premise
The entire platform runs on institution-owned hardware inside your perimeter.
Best for
Universities and institutions with existing datacentre capability
- Nothing leaves the campus network
- Institution owns the full stack
- Operates without internet access
- Sized against concurrency, not headcount
Choosing
Pick on your binding constraint, not on preference
Almost every institution has one condition that decides this. Find yours below and the model follows.
Your governance rule says institutional records cannot leave campus
Hybrid or on-premise. Both keep the index, documents and retrieval inside the campus network; the difference is only whether inference also runs on your hardware.
You want something live this term
Private cloud. It is the fastest route to a working pilot and carries a clean migration path on-premise, so speed now does not cost you control later.
You have no datacentre capability and no appetite to build one
Private cloud or hybrid. Hybrid keeps your knowledge on campus with a smaller hardware requirement, because peak load is absorbed elsewhere.
Your campus network has to keep working when the internet does not
On-premise. The full stack runs inside the perimeter and operates without internet access at all.
You cannot predict adoption yet
Hybrid, the recommended starting architecture. On-premise sizing is driven by concurrency rather than headcount, and burst capacity covers exam and admission peaks while you learn your real numbers.
Architecture
What sits where
A hybrid deployment, drawn end to end. In an on-premise deployment every layer moves inside the campus boundary.
01 · Access
SSO, role mapping, department scope
02 · Ingestion
OCR, structuring, versioning
03 · Retrieval
Institution-owned index, on campus
04 · Reasoning
Private inference, isolated tenancy
05 · Governance
Audit log, attribution, analytics
Data ownership
The institution owns its documents, index and embeddings and can export them at any time.
No cross-tenant sharing
Nothing is pooled with other institutions, at any layer, in any deployment model.
Exit path
If the relationship ends, the institution keeps its knowledge base in an open format.
How those three guarantees are enforced — encryption, isolation, audit logging and zero training on institutional data — is set out on the security model.
Deployment questions
What IT asks before procurement starts
If your constraint is not covered here, it is worth a call rather than a guess.
Can we start in the private cloud and move on-premise later?
Yes, and it is a common path. You can start in a dedicated tenancy in India for speed and migrate on-premise afterwards without changing how staff or students use the platform. Nobody gets re-trained because the interface and the roles do not change.
How much hardware does on-premise need?
Sizing is driven by concurrency — how many people are querying at once — rather than by how many people are enrolled. That is scoped separately after a technical review, because the honest answer depends on your peak, not on your headcount.
In a hybrid deployment, what actually leaves campus?
Not the institutional index, the documents or the retrieval. Those stay on campus by design. What moves off is elastic compute capacity for peaks such as examinations and admissions.
Is anything shared with other institutions?
No. Nothing is pooled with other institutions at any layer, in any of the three models. Private cloud means a dedicated tenancy, not a shared multi-tenant service.