Lux · latin for light

Every model,
in one light.

One key enters. Lux refracts it to every provider in its native format, holds the secret, and returns the answer the same way.

one key
OpenAI
Anthropic
Gemini
OpenRouter
Ollama
Moonshot
Grok
GLM
Local
The path of one call

Six steps. One request.

Your appone base URL
Authenticatevirtual key or login
Limitsallowlist, cap, rate
Routepick the provider key
Providernative wire format
Recordtokens, cost, latency
The model control plane

Every call.
One pane of glass.

See every model request the moment it happens: who called, which provider answered, the tokens, the latency, the cost. Set the limits once and watch them hold.

lux.latere.ai/dashboardLIVE
Requests today
18,432
Tokens
9.64M
Spend
$412.80
ModelCodeTokensCostLatency
claude-3.7-sonnet2001,842$0.0139742ms
gpt-4o-mini200604$0.0058311ms
local/llama-3.1-8b2002,317$0.01751094ms
gemini-2.0-flash429415blocked208ms
mixtral-8x7b2001,268$0.0101866ms
gpt-4o-mini200933$0.0072428ms
One key per app

A key for every app, agent, and environment.

ProductionACTIVE
lux_live_••••••••••
Spend cap$3.2k / $5k
gpt-4oclaude-3.7+3
StagingACTIVE
lux_test_••••••••••
Spend cap$180 / $500
gpt-4o-minilocal/*
Agent swarm82% CAP
lux_agent_•••••••••
Spend cap$820 / $1k
claude-3.7gpt-4o20 req/min
Agent swarm, monthly spend$1,000 cap
$820 used of the $1,000 monthly cap
Spend caps that hold

Overruns stop before they spread.

A budget lives on the key, not the caller. Fork an agent, spawn a subprocess, or fan out a swarm: the cap is the same one, and when it is reached the next call is refused, not silently paid for.

Two lines of difference

Keep your SDK. Change two lines.

from openai import OpenAIclient = OpenAI(- base_url="https://api.openai.com/v1",- api_key=OPENAI_API_KEY,+ base_url="https://lux.latere.ai/openai/v1",+ api_key=LUX_VIRTUAL_KEY,)

The rest of the call site doesn't change.

Building something new? The Lux SDKs for Go, TypeScript, and Python speak one native format for every model. Meet the Lux API →

Run it yourself

Your machine, every Lux client.

Your machine, behind NAT
latere lux servedials out, no inbound port
Ollama, vLLM, MLXlocalhost runtime
outbound wssrequests stream back
lux.latere.ai
Lux gatewayidentity, limits, log
Any callercalls local/<model>
Works withOllamavLLMLM Studiollama.cppApple MLXany OpenAI-compatible server

One light.
Every model.

Secrets held for you, one sign-in, one record for spend and audit, and a check on every call.