# RAW

> Dedicated GPU and CPU servers for AI workloads, provisioned in 3 seconds over a 100% REST API with flat pricing and zero egress fees.

RAW is a bare-metal cloud platform built specifically for AI inference, training, and agent workloads. Founded in 2025 by Michael Jakob — who previously built Eulerpool, a financial data platform — RAW grew out of a real $7,000 monthly cloud bill and the frustration of AWS complexity. The platform offers dedicated NVIDIA GPU and AMD EPYC CPU servers across five global regions, provisioned via a single REST API in as little as 3 seconds.

## What It Is

RAW is a server-creation and fleet-scaling API designed to replace AWS, DigitalOcean, and GPU clouds for AI teams. Instead of navigating a 200-service console, users POST to a single endpoint to get a running server with full root SSH, CUDA, NVMe storage, and a public IP. The platform targets LLM inference, fine-tuning, AI agent hosting, and vector database deployments. RAW positions itself as "the anti-AWS console" — one API, flat pricing, and zero egress fees.

## How the API Works

The entire product surface is a REST API with 24 endpoints. The workflow is intentionally minimal:

- **Sign up** with a single POST to get a Bearer token — no charge
- **Add a card** via `/billing/setup` (Stripe-backed, $0 until deploy)
- **Deploy** with `POST /deploy` passing `type` and `region` — server is live in ~13–30 seconds for CPU, 1–48 hours for GPU (dedicated hardware reservation)
- **Scale** by repeating the same call; destroy with `DELETE /servers/:name`

A CLI (`npm install -g rawhq`) and a web dashboard wrap the same endpoints. The dashboard provides 14 management tabs per server: metrics, networking, firewalls, volumes, backups, snapshots, rescale, rebuild, power, rescue, terminal, and an AI prompt helper for Claude/ChatGPT.

## Infrastructure and Regions

RAW runs on dedicated hardware — no hypervisors, no shared tenancy — in Tier-3 datacenters across five regions:

- 🇩🇪 Frankfurt, Germany (`eu`)
- 🇮🇪 Dublin, Ireland (`eu-fi`)
- 🇺🇸 Ashburn, Virginia (`us`)
- 🇺🇸 Hillsboro, Oregon (`us-west`)
- 🇸🇬 Singapore (`sg`)

CPU servers use AMD EPYC and Intel Xeon processors with NVMe SSDs. GPU servers use NVIDIA A100, L40S, and RTX cards with full CUDA. GPU SKUs are currently EU-only. All servers include unlimited bandwidth with no egress metering.

## AI Stack Support

RAW is designed to run the full AI stack on dedicated hardware:

- **LLM inference**: vLLM, Ollama, TGI with OpenAI-compatible endpoints; supports Llama, Mistral, DeepSeek
- **Training and fine-tuning**: Full CUDA access, up to 96 GB VRAM, NVMe for checkpoints; LoRA or full training
- **AI agents**: Persistent, always-on CPU servers with root access; scale workers via API from CI
- **Vector databases**: Qdrant, pgvector, Milvus on dedicated NVMe
- **Pre-install apps**: Docker, PostgreSQL, Redis, Node.js, Nginx, Caddy, Python, Go, Bun, MySQL, OpenClaw

## Deployment Model and Company Background

RAW is bootstrapped, headquartered in Munich, Germany, and describes itself as "AI-native" — product development, infrastructure automation, and operations are run primarily by AI systems rather than a large human team. The founder, Michael Jakob, migrated Eulerpool's entire stack from managed cloud to bare metal before building RAW as a product. The platform holds SOC 2 Type II certification and offers GDPR-compliant EU GPU regions. Per-second billing is available, and the 99.9% uptime SLA applies across all regions.

## Features
- 100% REST API with 24 endpoints
- GPU servers with full CUDA (NVIDIA A100, L40S, RTX)
- CPU servers from 2 vCPU to 48 vCPU
- 3-second server provisioning for CPU
- Full root SSH access on every server
- NVMe SSD storage on all servers
- Unlimited bandwidth with $0 egress
- 5 global regions (EU, US East, US West, Singapore)
- Per-second billing
- CLI (npx rawhq / npm install -g rawhq)
- Web dashboard with 14 management tabs
- Live metrics: CPU, RAM, disk, network
- Point-in-time snapshots and automated daily backups
- Persistent block storage volumes (10 GB–10 TB)
- SSH key management (import from GitHub)
- Web terminal (VNC console in browser)
- Pre-install apps: Docker, PostgreSQL, Redis, vLLM, Ollama, etc.
- Team access with Owner/Admin/Member roles
- SOC 2 Type II certified
- GDPR EU GPU regions
- Flat pricing — one price per server, no hidden fees

## Integrations
vLLM, Ollama, PyTorch, CUDA, Llama, Mistral, DeepSeek, Qdrant, pgvector, Milvus, Docker, PostgreSQL, Redis, Nginx, Caddy, Node.js, Python, Go, Bun, MySQL, OpenClaw, Stripe, GitHub (SSH key import)

## Platforms
LINUX, WEB, API, CLI

## Pricing
Paid

## Links
- Website: https://rawhq.io
- Documentation: https://rawhq.io/docs
- EveryDev.ai: https://www.everydev.ai/tools/rawhq