OpenModex
OpenAI-compatible · Drop-in replacement

Any model
behind one key.

The unified supply layer for AI apps and agents. Models, compute, and tools behind one metered key — bring your own keys, or run on our pooled capacity.

Routing across every major provider and compute channel

OpenAIAnthropicGeminiDeepSeekQwenMistralSiliconFlowchannelNovitachannelDeepInfrachannelFalchannelTogetherchannelFireworkschannel
console.openmodex.com
OOpenModex
Overview
API Keys
Models
Routing
Provider Keys
Tools
Usage
Billing
Logs
Routing Decisions · liveConfigure
RequestStrategyRouted toLatency
chat · gpt-4o
cost_optimized
openai · direct
412ms
chat · deepseek-v3
cost_optimized
siliconflow · channel
287ms
chat · claude-sonnet
quality_first
anthropic · direct
598ms
chat · qwen-max
latency_first
novita · channel
203ms
embed · bge-m3
cost_optimized
deepinfra · failover
158ms

One endpoint, every model

Cost, latency, or quality-first routing with per-model circuit breakers and automatic failover.

Learn more

Supply is the product.

Others sell routing. OpenModex operates the supply underneath it — capacity, keys, and tools.

Compute

Direct channels, not just official APIs

We contract inference capacity across direct channels and arbitrage in real time. You get the same models at structurally lower cost — without managing a single vendor relationship.

Keys

Hybrid BYOK, done right

Bring your own keys and keep your negotiated rates. The platform pool catches rate limits, outages, and overflow. No other gateway treats both as first-class.

ToolsIn development

The MCP gateway

Agents need tools the way apps need models. One URL for hosted MCP tools — same key, same meter, same audit trail as your model traffic.

Built like payments infrastructure

Every request clears through a billing pipeline designed for money, not logs.

Pre-deduct, settle, refund

Estimated cost is reserved before the request leaves, settled on completion, and the difference refunded — atomically.

Idempotent by design

Idempotency keys, atomic Lua balance operations, and sliding-window rate limits on every tier.

Reconciled continuously

Billing events stream through Kafka into a warehouse that reconciles every token against every invoice.

6+

providers behind one endpoint

<150ms

added latency at the median

1

invoice for models, compute, and tools

Everything a production gateway needs

Shipped and running today — not a roadmap.

Smart routing

Cost, latency, or quality-first strategies per request.

Circuit breakers

Per provider-model health with automatic failover.

Prompt caching

Exact-match cache with configurable TTL.

Rate limiting

Tier-based sliding windows backed by Redis.

Idempotency

Safe retries with 24-hour deduplication.

SSE streaming

Full streaming support across every provider.

Usage analytics

Daily and per-model breakdowns, exportable.

Teams & key vault

Encrypted provider keys, roles, and usage policies.

Change one line. Keep your stack.

Point your existing OpenAI SDK at api.openmodex.com and the whole supply network is behind it.

Get your API key
from openai import OpenAI

# Just change base_url  your existing code works
client = OpenAI(
    api_key="omx_sk_...",
    base_url="https://api.openmodex.com/v1",
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}],
)

print(response.choices[0].message.content)

Get up and running in minutes

Free tier included. No card required. Your first key is one click away.