Your data never enters a shared model.
The Cyborg’s working memory, indexes, logs, and reasoning traces all live on its VM. We never use them to train anything that benefits another customer. Yeh contractual hai, technical bhi hai.
— CONCEPT · 01 OF 04 · ISOLATION MODEL
One Cyborg. One machine.
One customer. Forever.
Every Cyborg you hire gets its own dedicated Windows VM in the Aliens data-centre. Not a shared inference endpoint. Not a multi-tenant cluster. A real, isolated, named machine — with your data on it, your software on it, your tools authorised on it. When you fire the Cyborg, the machine is wiped. When you keep it, your data never touches another customer’s — ever.
The architectural choice most vendors won’t make.
Multi-tenant inference is cheaper to operate. That’s why almost every AI company uses it. One model serves a thousand customers; the unit economics are beautiful for the vendor.
It’s also the reason their security pages dance around the question of data isolation. Once your prompts go through a shared model, “your data” becomes a fuzzy thing.
We picked the harder model on purpose. One Cyborg, one machine, one customer. The unit economics are worse for us — passed on transparently in the salary — but the data story is unambiguous.
— 02 / WHY DEDICATED
— THREE CONCRETE REASONS
Three things you only get with a dedicated machine.
If you leave, you take everything.
Memory, reports, configs, integration credentials — all exportable in 7 days. The VM gets wiped + recycled. There’s no “residual fine-tune” left behind to argue about.
Every action is logged on your side.
Pull the audit log from your dashboard, or have it streamed to your SIEM. Every command the Cyborg ran, every API it called, every document it touched — on a single timeline.
— 03 / VERSUS THE ALTERNATIVES
— HOW THIS COMPARES
Side by side — what isolation actually means.
Property
Multi-tenant AI tool
A’nil Cyborg (dedicated VM)
Each customer’s data on its own host?
No — shared inference cluster
Yes — named, isolated VM
Memory survives re-deploys?
Sometimes — per-vendor caveats
Always — persisted on your VM disk
Cross-tenant fine-tuning?
Often opt-out, not opt-in
Never — technically prevented
Real audit log of every action?
Partial — redacted or rolled-up
Full — per-API-call, exportable
Exit data takeout
Varies — ranging from clean to hostile
7 days · structured + portable
Cost
Cheaper — paid in privacy fuzziness
Higher — paid in clear isolation
— 04 / WHAT’S IN THE BOX
— STANDARD CYBORG VM SPEC
The default spec sheet.
This is the baseline for every Cyborg, regardless of role. Heavier roles (Data Analyst, Software Engineer) can scale up. None scale down below this.
- Compute
- 8 vCPU on AMD EPYC (3rd gen+) · bursts to 16 vCPU during peak workload windows
- Memory
- 32 GB DDR5 · reserved (not over-committed)
- Storage
- 500 GB NVMe SSD · encrypted at rest with per-customer key · daily snapshot
- OS
- Windows 11 Pro · hardened image ·
WebOSruntime pre-installed - Network
- Private VPC · 10 Gbps internal · egress only via vetted proxy with full request log
- GPU / inference
- Shared GPU pool for inference calls only · no model state retained on shared pool · per-call audit
- Region
- India (ap-south-1, Mumbai) by default · EU / US / UAE on enterprise plans
- Backups
- Daily snapshot retained 30 days · weekly retained 12 months · you can request takeout anytime
- Uptime SLA
- 99.9% (P0) · 99.95% (P1+, enterprise)
- Decommission
- On contract end · 7-day cooling-off · cryptographic wipe + certificate of destruction issued
— 05 / FAQ
— THE QUESTIONS YOU ACTUALLY HAVE
FAQ · straight answers.
Can I bring my own cloud?
Enterprise plan only. We can deploy the Cyborg VM into your AWS / Azure / GCP account, with the runtime under your IAM. P0 customers run on Aliens-managed infra by default.
What if I need a Cyborg in the EU specifically?+
Available on enterprise plans. Frankfurt (eu-central-1) for GDPR; London (eu-west-2) for FCA workloads. The whole VM lifecycle, including backups, stays in-region.
Do Cyborgs share the underlying language model?+
Inference calls go through a shared GPU pool, but no model state, prompt cache, or conversation history persists on that pool. Every request is stateless from the GPU’s point of view; state lives on your dedicated VM.
What happens to the VM when I cancel?+
7-day cooling off. Then cryptographic wipe, then VM termination, then certificate of destruction issued to your billing email + audit log. The 7-day window lets you change your mind or extract anything you missed.
Can two of my Cyborgs share a VM?+
No. The architecture is one role per VM, deliberately. Different roles have different tool authorisations, different memory schemas, and different blast radii. Mixing them re-introduces the multi-tenancy we’re avoiding.
Is there a way to see the VM live?+
Yes. From your dashboard you can see the VM status, current task, recent logs, and (on enterprise plans) RDP into it directly. Most customers don’t — the daily report covers what they need to know.
— 06 / NEXT IN CONCEPT
— THE FOUR-PART CONCEPT WALKTHROUGH