FAQ

Do you need access to our production systems?

No — and that's deliberate. Most work is designed to be 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 a temporary, key-only account you create, that expires on its own and that you can revoke with one command, after you sign our client authorization. It is reached over your VPN or a WireGuard/Tailscale tunnel you control, only during your window; we never ask you to expose SSH to the internet. Every command is recorded, and you get a tamper-evident copy. 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, carried inside your VPN or a tunnel you control. Your cluster is just another set of nodes we work on exactly the way we work on our own. There's nothing to screen-share — no GUI, no desktop. Every command we run is recorded with its reason and output, and you get a sealed copy of the record at the end of each session.

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, minimum 4 hours, scheduled ahead of all other work; for now the emergency channel is email to contact@anantus.net, as we don't yet have a business phone line. 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?

Paid up front — the clock starts when payment lands and materials are in hand. Invoices in USD, from contact@anantus.net; payment details are on the invoice.

Is our data safe? Can you work offline?

Your configs, logs, and model details go to one firm, under NDA — never into a third-party talent platform. The outside services we use today include Proton Mail (email) and Cloudflare (hosting, cookie-free analytics); any others that process engagement information are bound by confidentiality and named in your engagement or on request. We don't sell your information. Lab-only work on your configs can run on our nodes that have no internet route, fully offline. Our lab is not a certified air gap: most of our eight nodes have no internet route; one is also our desktop and is our audited way out, and one other can reach the internet through our office network to download software packages. 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. The principal's identity and credentials are disclosed under mutual NDA 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.