Local infrastructure, beautifully simple

One gateway.
Every model.

A resilient local control plane for your AI stack. Put up to 50 provider keys behind one clean endpoint, keep your SDKs native, and let intelligent failover handle the rest.

Windows 10 / 11Unsigned preview · open source
AI Load Balancer— □ ×
LIVE CONTROL PLANE
Gateway overview
ACTIVE ROUTES4 +2.8%
API KEYS18/50
OpenAI ProductionHEALTHY38ms
Claude FailoverSTANDBY52ms
Gemini ResearchHEALTHY71ms
99.98%route uptime
Failover ready4 strategies online
BUILT FOR THE WAY YOU SHIP
OpenAI SDKAnthropic SDKNative SSE50 accountsLocal-first

01 — What it does

Make every key
work harder.

AI Load Balancer gives your application one stable local endpoint while it shares traffic across provider accounts, protects unhealthy keys, and keeps your code compatible.

Balance traffic

Spread requests across multiple API keys instead of sending everything through one account. Choose priority, round robin, weighted balance, or least-used routing.

Recover automatically

If a key hits a rate limit, billing issue, network timeout, or upstream error, the gateway cools it down and tries a healthy account.

Keep your SDKs

Use the official OpenAI or Anthropic client with environment variables. OpenAI Chat Completions and Anthropic Messages stay familiar.

50

Grow without rewrites

Start with one provider key and add up to 50 accounts in a gateway. Set weights, priorities, balance floors, and retry policy per gateway.

Fair use, clearly:this is for legitimate redundancy, uptime, and budget management—not bypassing provider policies, billing controls, or rate limits.

02 — The advantage

Quietly powerful
under the hood.

All the routing muscle you need, without another hosted dashboard, another bill, or another abstraction to learn.

01

Smart routing

Priority, round robin, weighted, and least-used strategies make every key count.

ABC
02

Graceful failover

Cooldowns, Retry-After awareness, and bounded retries keep requests moving.

49max retries
03

Truly compatible

OpenAI and Anthropic clients work natively, including cross-protocol translation.

OpenAIAnthropic
04

Private by default

Keys stay encrypted on your device. The gateway is local, inspectable, and yours.

SAFE
STORAGE

03 — Drop-in by design

Keep your SDK.
Change nothing else.

Point your official client at the local base URL. Your application keeps its familiar types, tools, streams, and errors.

OpenAI SDK

Chat Completions compatible

const client = new OpenAI({
  apiKey: process.env.AI_GATEWAY_API_KEY,
  baseURL: "http://127.0.0.1:42891/v1"
});
Native endpoint · /v1/chat/completions

Anthropic SDK

Messages API compatible

const client = new Anthropic({
  apiKey: process.env.AI_GATEWAY_API_KEY,
  baseURL: "http://127.0.0.1:42891"
});
Native endpoint · /v1/messages

04 — Made for momentum

From keys to clarity.

Three calm steps from a fresh download to a production-ready local route.

01

Add your keys

Store provider credentials in the encrypted local vault.

02

Choose a strategy

Set priorities, weights, floors, and failover rules.

03

Ship confidently

Use official SDKs while the gateway balances the load.

05 — Ready when you are

Your local gateway
starts here.

Download the portable Windows preview. No installer, no account, no cloud control plane.

LATESTWindows x64Unsigned preview · open source