Unified Multi-Cloud Compute Brokerage

One OpenAI-compatible endpoint aggregating AWS, Azure, Google Cloud, CoreWeave, Lambda, Cerebras, and Groq with live spot arbitrage.

Integration Docs

Production Gateway Endpoint (OpenAI-Compatible):

Latency-Optimized Anycast Edge

POST https://api.appzed.com/v1/chat/completions

Routing Policy: Automatically selects the lowest cost GPU vendor per token with 99.99% fallback SLA.

Live Multi-Cloud API Playground

Test real-time routing, compare token economics, and inspect multi-vendor response telemetry live.

Model Target:

Prompt:

Gateway Telemetry & Response:

Click "Execute Request" to stream multi-vendor response telemetry.

Active Multi-Vendor Fabric Providers (7 Connected)

Cerebras CS-3 2,100 tok/s

Wafer-Scale cluster for ultra-high throughput reasoning.

$0.60 / 1M tokens • 35ms TTFT

Groq LPU Array 380 tok/s

Deterministic tensor streaming for live conversational voice.

$0.59 / 1M tokens • 75ms TTFT

CoreWeave / Lambda Spot GPU

NVIDIA H100 / H200 high-bandwidth memory clusters.

$0.35 / 1M tokens • Auto-failover

AWS Bedrock & GCP Hyperscaler

Trainium2, TPU v6 & enterprise multimodal nodes.

$0.06 - $0.80 / 1M • Enterprise SLA

1-Click Serverless Model Hosting

Deploy open-weights models on auto-scaling vLLM and TensorRT-LLM runtimes. Zero Kubernetes management, automatic scale-to-zero.

Recommended Foundation Model Templates

DeepSeek R1

Reasoning SOTA

Frontier reasoning model with deep Chain-of-Thought logic for STEM, math, and code generation.

Recommended: H200 NVL Spot Fabric

Engine: vLLM PagedAttention v2

Llama 3.3 70B

Versatile Flagship

Meta's 70B flagship foundation model for multilingual agent workflows, RAG, and tool calling.

Recommended: Cerebras CS-3 or H100

Engine: TensorRT-LLM FP8

DeepSeek V4 Flash

$0.09/1M Tokens

Ultra-low latency Mixture-of-Experts engine supporting 1M token context with high efficiency.

Recommended: Multi-Cloud Spot GPU Fabric

Engine: vLLM MoE Runtime

Phi-4 Mini (3.8B)

Ultra Lightweight

Microsoft's high-density reasoning SLM for fast, cost-effective structured JSON & tool use.

Recommended: Serverless LPU / Spot GPU

Engine: ONNX Serverless Engine

Active Serverless Deployments (3 Running)

Auto-scales to 0 when idle

Deployment ID / Name	Runtime Engine	Hardware Target	Status	Cold Start	Throughput	Replicas
prod-llama33-70b-vllm https://api.appzed.com/v1/deploy/d-892a	vLLM v2	CoreWeave H200 Cluster	Active	28ms	185 tok/s	1 / 4
deepseek-r1-serverless https://api.appzed.com/v1/deploy/d-741c	TensorRT-LLM	NVIDIA Spot GPU Fabric	Active	34ms	115 tok/s	2 / 8
phi4-mini-agent-endpoint https://api.appzed.com/v1/deploy/d-290f	ONNX Serverless	AppZed Auto-Spot LPU	Scaled to 0	18ms	240 tok/s	0 / 2

Hybrid Edge & On-Device SDK

Seamlessly execute lightweight prompts locally on Apple Neural Engine (CoreML) and Android AICore for $0.00 cloud compute cost, only bursting to cloud GPU clusters for complex reasoning.

Est. Savings: 64% Cloud Bill Offload

Intelligent Edge/Cloud Dynamic Routing Rules

Configure automatic threshold rules for client-side execution vs multi-cloud GPU dispatch.

Rule 1: Local On-Device ($0.00) Active

When Prompt Length < 500 tokens AND Task == "Summarize / Classify / Autocomplete"

→ Route to Apple Foundation 3B (CoreML) or Android AICore

Rule 2: Multi-Cloud GPU Burst Active

When Task == "Deep Reasoning / Code / Agent Tool Turn" OR Device Offline == false

→ Burst to Cerebras CS-3 / DeepSeek R1 / Claude 3.7

AppZed Edge SDK Quickstart

import AppZedClient

// 1. Initialize AppZed Hybrid Edge Client
let client = AppZedClient(
    apiKey: "zed_live_9f82c41a7d6e8b20c4e1a09d",
    routingPolicy: .hybridSmartOffload(
        localModel: .appleFoundation3B, // Apple Neural Engine ($0.00)
        cloudFallback: .cerebrasLlama33 // Wafer-Scale Cloud Burst
    )
)

// 2. Execute Prompt (Automatically chooses local ANE vs Cloud GPU)
let response = try await client.generate(
    prompt: "Summarize this offline user note...",
    taskComplexity: .autoDetect
)

print("Served by: \(response.telemetry.executedOn)") // "Apple Neural Engine (Local $0.00)"

API Keys & SDK Integration

Manage your production API keys and configure spending limits.

Active Secret Keys

Key Name	Key Secret	Created	Last Used	Monthly Quota	Actions
Production Default Key	zed_live_9f82c41a7d6e8b20c4e1a09d	Today	Just now	$500.00 / mo

cURL & Python Example

# 1. cURL Example
curl https://api.appzed.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer zed_live_9f82c41a7d6e8b20c4e1a09d" \
  -d '{
    "model": "cerebras/llama-3.3-70b",
    "messages": [{"role": "user", "content": "Hello AppZed Multi-Cloud Fabric!"}]
  }'

Real-Time Arbitrage Stream & Health

Live telemetry feed across all connected GPU clouds and inference clusters.

Live Stream Active

Cloud Provider	Region / Cluster	Active Workload	Current Latency	Spot $/1M Tokens	Arbitrage Spread	Status
Cerebras Wafer-Scale	us-east-supercluster	Llama 3.3 70B (2,100 tok/s)	32ms	$0.60	-45% vs AWS	Operational
Groq LPU Array	us-central-lpu-1	Real-Time Voice Streaming	71ms	$0.59	-35% vs Hyperscaler	Operational
CoreWeave H200	us-east-gpu-fabric	DeepSeek R1 / vLLM v2	128ms	$0.35	-62% vs Azure	Operational
Lambda Labs Spot	us-west-spot-h100	DeepSeek V4 Flash	145ms	$0.09	-78% SOTA Spread	Operational
Apple CoreML Edge	Client Device (Local)	Apple Foundation 3B	15ms	$0.00 (FREE)	100% Free Offload	On-Device

Usage Analytics & Invoices

Transparent monthly token breakdown and multi-vendor spot arbitrage savings.

48.2M

Total Monthly Tokens

$1,428.50

Arbitrage & Edge Savings

$32.18

Current Billing Cycle Cost

99.99%

Gateway Uptime