Hardware
What you already own
GPU class and VRAM against your document volume and expected concurrency. Existing estates are often enough; “none yet” is a valid answer.
T4 · L4 · A10 · A100 · VRAM
Free 30-minute on-premises review
An Arc53 deployment engineer reviews your hardware, identity, network boundary and governance needs in thirty minutes. You leave with a hardware and governance checklist for your environment, whether or not you deploy DocsGPT.
No charge, no commitment, and no sales engineer on the call.
Last updated:
What the engineer reviews
Four questions decide whether an on-premises deployment is straightforward for you or not. The call goes through them in order; Deployment options covers the same ground in writing.
Identity
How people get in and out
Your identity provider, single sign-on and provisioning, so joiners and leavers are handled by the system of record.
OIDC SSO · SCIM 2.0 · RBAC
Network
Where the boundary sits
Hybrid, with retrieval local and prompts going to a cloud model, or fully air-gapped with no external egress at all.
hybrid · zero egress
Governance
What your auditor will ask
Retention, redaction, approval gates, and what the append-only audit log has to record for your evidence pack.
audit log · approvals · retention
Tell us about your environment
Anything you tell us about hardware and identity now saves time on the call.
Received
Thanks — that's with us.
The details are with the engineer who will take the call.
Pick a time that suits you — the invite goes to the work email you gave.
Pick a time30 min · video call
Booked
The invite is on its way to .
Need to move it? The invite carries a reschedule link.
Worth having ready
- Your GPU class and VRAM per card, or that there is no hardware yet.
- Which identity provider you run, and whether SCIM is in use.
- Any constraint that decides the shape of it — egress rules, residency, a go-live date.
In the meantime: Security · Pricing ·Quickstart.
What happens next
- 01
Pick a time
You choose a slot on the deployment engineer's calendar. Nothing is charged and nothing is committed.
30 min · free
- 02
Before the call
We read what you sent about hardware and identity first, so the thirty minutes go on your environment instead of on questions we can answer ourselves.
- 03
What you leave with
A hardware and governance checklist for your environment — GPU class and VRAM, identity and SCIM, network boundary, retention and audit — whether or not you deploy DocsGPT.
checklist · by email
Frequently asked questions
Do we need the hardware already?
- No. Sizing is part of the review, and an existing GPU estate is often enough. Hardware fit & sizing gives you a first estimate before the call.
Will you sign an NDA before the review?
- Yes, on request — ask for one in the message field. Until it is signed, keep the form to your hardware, identity and network details.
Can we run a proof of concept afterwards?
- Yes. The standard Proof of Concept is one week on your infrastructure, with your data and your model, against pass/fail checks agreed before the week starts.