The platform, in depth
One console for every service, a purpose-built OS on every VM, a CLI baked into each instance, and an on-device AI shell. Here's what's actually under the hood.
Management Console
A full management UI — dashboards, inline function editor, VM console, IAM policy builder. Every service, one login.
11 services in the sidebar — one login covers everything
Inline function code editor — write, deploy, test without leaving the browser
Live VM console — SSH access directly from the UI
Aether OS
Most private clouds give you a bare VM and walk away. KoshCloud ships Aether OS — a purpose-built Linux that knows it's running on KoshCloud, and gives every instance first-class cloud identity from the moment it boots.
Direct kernel boot — no BIOS, no firmware, no grub. Apple Silicon HVF acceleration. The instance is live before you finish typing the command.
Every instance gets a unique IAM credential injected at boot via IMDS — auto-rotates every 12 hours. No secrets in your code, ever.
chetu is pre-installed. SSH in and immediately run chetu store put, chetu fn invoke, or chetu whoami — no setup required.
Type plain English: chetu create folder myproject, chetu why is it slow, or chetu ask <anything> — answered by a local LLM. No cloud. No latency.
No emulation. Aether OS runs natively on arm64. Combined with HVF, you get bare-metal performance inside a managed VM with full SSH PTY support.
IMDS credentials auto-rotate every 12 hours
Served at 10.0.2.100/latest/meta-data/iam/security-credentials/aether
Cloud CLI
SSH into any Aether instance and the full KoshCloud API is right there — no SDK install, no config files, no extra auth. chetu is pre-installed on every VM.
Cloud service commands
Gateway: http://10.0.2.2:8080 · auth via IMDS token (auto)
Aether OS · AI Shell
Every Aether instance ships with 30+ natural-language intents built into the shell — plus a local LLM (llama3.2 via Ollama) for anything that goes beyond the built-ins. No extra setup. No cloud API calls. Runs fully on-device.
30+ intents · 10 categories
Navigation & Files
chetu where am I · chetu show files · chetu go to work · chetu create folder myproject · chetu rename X to Y
System Intelligence
chetu why is it slow · chetu how much memory · chetu disk space · chetu free up space · chetu stop program nginx
Network & Search
chetu am I online · chetu download <url> · chetu find file config.json · chetu find largest files
Local LLM — anything else
chetu ask how do I reverse a string in python · chetu ask explain this error
LLM runs on-device via Ollama (llama3.2) — zero data leaves your perimeter
Interpreted, not just dumped
Responses include a verdict, a visual bar, and a "what to do next" — not raw numbers. The OS reads and reasons about system state.
Differentiators
Most "private cloud" products are Kubernetes with a dashboard. KoshCloud is something different — purpose-built services, a custom OS, and capabilities you won't find anywhere else.
Not containers. Not sandboxes. Full hardware-isolated virtual machines with their own kernel — booting in under a second thanks to direct kernel boot and HVF.
Run standard SQL queries on your NoSQL database via DuckDB — a feature no major cloud platform offers natively. Plain JSON, no typed attribute nonsense.
Every VM gets its own rotating credential at boot — via IMDS. No static keys in env vars. No secret sprawl. Tenant isolation enforced at the gateway.
KoshCloud runs with zero internet dependency. Ship it to an aircraft carrier, a hospital, or a factory floor. All 11 services work completely offline.
Full management UI at port 3000 — dashboards for every service, inline function code editor, VM console, IAM policy builder. Plus the chetu CLI for automation.
Your existing SDKs, infrastructure code, and CLI tooling work without modification. Change one endpoint URL — zero code changes, zero retraining, zero rewrites.
Run LLM inference, vector databases, and ML pipelines on your own hardware — with the full cloud primitives (queues, functions, storage, secrets) needed to build production AI systems. No GPU tax. No data leaving your network.
How it works
Run docker compose up on any machine you own.
All 11 services start in under a minute. No cloud account required.
Point your apps at your gateway URL. Same APIs. Same SDKs. Same config your team already knows. Zero migration effort.
Your data stays inside your perimeter. Your infrastructure, your rules. No invoice arrives at the end of the month.