FAQ

Do you need access to our production systems?

No — and that's deliberate. Most engagements are reproduced on our GB10 lab from the config and logs you send. If it can't be reproduced, that itself is a finding. When live access genuinely speeds things up, it's scoped SSH or a reverse tunnel you open and can close, in a shared terminal you can watch command by command. We never need your production credentials or a standing door into your network.

How does the remote work actually happen — RDP? Screen sharing?

Neither. This work is entirely command-line over SSH. Your cluster is just another set of nodes we operate exactly the way we operate our own. There's nothing to screen-share — no GUI, no desktop. If you want to watch, we run inside a shared terminal session and you see every command as it runs.

What does it cost?

$1,500 fixed for a one-day diagnostic on a single or dual node. $7,500 fixed for a full production-cluster audit. Implementation is $400/hr or a fixed bid against a target you name. Emergencies (production down) are $1,000/hr with response in hours, not days. Full detail on the services page. And the diagnostic carries a guarantee: if we find nothing actionable, it's free.

What if the fix turns out to be simple?

Then it's in the report and you're done — no upsell, no invented follow-on work. We'd rather you tell three people the diagnostic was worth it than squeeze one engagement. If the fix is real work, the report ends with a scoped estimate you're free to decline.

How do we pay?

ACH or bank transfer (no fee, preferred by most clients), or card. Paid up front — the clock starts when payment lands and materials are in hand. Invoices come from a real mailbox, in plain USD, no platform in the middle.

Is our data safe? Can you work air-gapped?

Your configs, logs, and model details go to one firm, under NDA, on hardware we control — never into a third-party talent platform. We operate fully air-gapped on request, because we run air-gapped clusters ourselves every day. Full detail on how we work.

Who is behind Anantus?

anantus.net LLC is a founder-led U.S. company. Our main work is building private, self-hosted AI models; engineering services apply what we learned running our own GB10 cluster. The person who answers your email does the work, and company details, references and a mutual NDA are available before any engagement. See About.

What exactly is in scope?

Multi-node bring-up, throughput and latency tuning, weight-load stalls, unified-memory behavior, speculative decoding, quantization correctness, and benchmark forensics — on NVIDIA Grace-Blackwell (GB10 / DGX Spark) hardware. If it's on the trap index, it's in scope. If your problem is on a different platform, we'll say so and point you somewhere useful.