# ComputeGpu

> Domain: https://computegpu.com
> Type: Cloud GPU rental platform (instances, serverless, clusters)
> Language: English
> Publisher: Tyga.Cloud Ltd (UK#14643275)
> Contact: joe@tyga.cloud
> Sitemap: https://computegpu.com/sitemap.xml
> API Docs: https://computegpu.com/docs
> Agent Skill File: https://computegpu.com/api/docs/skill.md
> Agent Discovery: https://computegpu.com/.well-known/ai-plugin.json
> OpenAPI Spec: https://computegpu.com/.well-known/openapi.json

## Agent Session Pattern (START HERE)

npm i -g computegpu
computegpu signup you@co.com my-agent     → gpu_ API key instantly
computegpu types                          → 47 GPU models with pricing
computegpu deploy --gpu "RTX 4090"        → GPU running in ~30s
computegpu pods                           → list active instances
computegpu stop POD_ID                    → pause billing
computegpu destroy POD_ID                 → done

If 402: owner must verify with $1 at /auth/verify-payment (credited to wallet)
MCP: computegpu mcp-serve (stdio, for Claude Code / Cursor)

REST API fallback: POST /api/signup, GET /api/v1/gpu/types, POST /api/v1/gpu/deploy

## What ComputeGpu Is

ComputeGpu is a cloud GPU platform offering 47 GPU models from RTX 3060 ($0.22/hr) to H200 ($5.38/hr). Per-second billing. Three deployment modes: GPU instances (Ollama API, Docker, SSH), serverless endpoints (scale to zero), and multi-GPU clusters (2-32 nodes).

Two unique features no competitor has:
- Shared Memory (agenticmemory.ai) — persistent AI memory across instances, restarts, and GPU upgrades. Zero code changes.
- AI Compliance (a2ainfrastructure.com) — 4-gate safety firewall with OCSF audit trail. EU AI Act, US state laws, NIST, GDPR, SOC 2 ready.

Both bundled free with every rental.

## Core API Endpoints

POST /api/signup — get API key instantly (email required, $0 to start)
GET /api/v1/gpu/types — 47 GPU models with live pricing
GET /api/v1/gpu/cheapest?vram=24 — cheapest GPU for a VRAM requirement
POST /api/v1/gpu/deploy — deploy on-demand GPU instance
POST /api/v1/gpu/deploy-spot — deploy spot instance (50% off, preemptible)
GET /api/v1/gpu/pods — list your active instances
PUT /api/v1/gpu/pods/:id/stop — pause instance (billing pauses)
DELETE /api/v1/gpu/pods/:id — destroy instance permanently
POST /api/v1/gpu/billing/request-topup — email account owner to add credits

## Authentication

Get a key: computegpu signup you@co.com my-agent (or POST /api/signup)
Key format: gpu_{tenantId}_{random}
CLI stores key in ~/.computegpu/config.json
REST API: Authorization: Bearer gpu_your_key

## Pricing

47 GPU models. Per-second billing. 20% markup over community cloud pricing.
Consumer GPUs: $0.22-$1.24/hr (RTX 3060 to RTX 5090)
Datacenter GPUs: $0.64-$10.76/hr (A40 to H200)
Spot pricing: 50% off (preemptible with 5-min warning)
Serverless: per-second compute, $0 when idle
Minimum wallet top-up: $5