Open Smart Routing Infrastructure

Your agents run 24/7.
Your router should
think before it spends.

ROUTEXOR is the BYOK routing layer built for agentic workloads: cron jobs, sub-agent orchestration, and multi-step pipelines across 317+ models. Assign the right model to every task type, set per-job budgets, pay providers directly. 2.2% pay-as-you-go, or flat $67 with zero markup.

Free tier included. No credit card required. Using Claude Code? Set it up in one line →

routexor · live
incoming query
routexor grades itdeep-analysiscomplexity
routing across your keys · in parallelensemble
Claude Sonnet 5
GPT-5.5
Gemini 3.5 Flash
judge · Claude Opus 4.8synthesizing…
Best answer in 312ms
your keysflat plan · $0 markup
this run ≈ $0.96 · you keep$0.048 vs a 5% router
$928/moest · at ~1,000 agent calls/day
kept vs a
percentage fee
solo
ensemble
the ensemble beats
a solo model

Cheaper than OpenRouter, two ways

OpenRouter charges 5.5%.
We charge 2.2%, or a flat $67.

Agents don't sleep. They run cron jobs, spawn sub-agents, and process chains around the clock, and a 5.5% routing fee turns that volume into a six-figure annual tax. Pay-as-you-go on ROUTEXOR is 2.2%, less than half of that. Scale up and switch to a flat $67/mo with zero markup: your provider invoices stay exactly what they'd be without us.

Typical router
ROUTEXOR
Routing fee
5.5% of every token
2.2% pay-as-you-go, or $0 on flat
Fee at $10K/mo of spend
$550
$220, or $67 flat
Fee at $100K/mo of spend
$5,500
$2,200, or $67 flat
Annual fee at $100K/mo
$66,000
$26,400 (2.2%) or $804 (flat)
Who holds your provider keys?
They do
You do
Cron job budget controls
No
Per-job budgets
Task-aware routing
No
Task profiles

Based on publicly listed pricing. 5.5% figure from OpenRouter's documented fee structure.

Without Routexor

Your agent's cron jobs default to the most expensive model — silently burning cash at 3am

Sub-agents inherit the parent's model even when a cheaper one would do

No way to assign different models to different task types in your pipeline

Provider goes down, your entire agent fleet goes down with it

No per-job budget limits — a runaway loop drains your balance before you wake up

Your routing provider holds your keys and takes 5.5% of every token

With Routexor

BYOK: your keys, your provider accounts, at 2.2% pay-as-you-go or a flat $67, never a 5.5% cut

Task profiles assign the right model to every job type: cron, sub-agent, pipeline step

Per-job budget caps halt runaway agents before they drain your balance

Automatic failover — if Anthropic is down, your agent fails over to GPT, then Gemini

Real-time cost tracking per agent, per cron job, per task profile

Your keys stay encrypted in your vault. We route with them, we never own them.

Setup for agent builders

Three steps. Your agents route smarter tonight.

01

Plug in your provider keys

Add your Anthropic, OpenAI, Google, Mistral, or any provider API key to ROUTEXOR. We use them to make requests on your behalf. Billing goes straight to your provider account — we never see a cent of your token spend.

# Dashboard → Keys
anthropic_key = "sk-ant-..."
openai_key = "sk-proj-..."
google_key = "AIza..."
✓ Keys encrypted at rest. Billed to your accounts.
02

Define task profiles for your agents

Your nightly data extraction agent doesn't need Opus. Your reasoning chain does. Task profiles let you assign the right model to every job type — cron jobs, sub-agents, pipeline steps — each with its own fallback chain.

# routexor.config
{
"summarize": { model: "claude-4-haiku" },
"code_review": { model: "claude-4-sonnet" },
"deep_analysis": { model: "claude-4-opus" }
}
03

Budget every agent. Sleep through the night.

Every task profile gets a budget ceiling. When a cron job or sub-agent hits its limit, ROUTEXOR pauses it and alerts you — it doesn't silently keep spending. Your agents run at 3am. Your wallet doesn't have to worry.

# Per-job budget caps
{
"summarize": { budget: "$5/day" },
"code_review": { budget: "$50/day" },
"deep_analysis": { budget: "$200/day" }
}
! budget exceeded → job paused, alert sent
$750
burned in 3 days.

A real story from the agent builder community: an automated document processing agent running on a cron schedule defaults to the most expensive model. No task profile, no budget cap, no guardrails. The agent runs silently for three days. $750 gone.

Why this happens

1

No task profiles

Routers treat every request the same. A cron agent summarizing receipts gets the same model as your reasoning chain doing deep analysis.

2

No per-job budgets

Without spend limits per agent task, a misconfigured loop or spawned sub-agent can drain your entire balance before you notice.

3

No cron-aware routing

ROUTEXOR is the first router built for autonomous agents. Tag a request with a task profile and it gets routed to the right model, budgeted, and monitored — whether it runs once or on a cron.

Your Keys, Every Provider

New model drops.
Your agents use it that day.

ROUTEXOR routes through your own provider keys across 13 providers — Anthropic, OpenAI, Google, xAI, Mistral, DeepSeek and more. The moment a provider ships a model to its API, it's available to your agents through the same single endpoint. No waiting for a middleman to add support, and we never mark up your token prices.

This isn't a reseller standing between you and the frontier. It's your direct line to it.

Introducing Ensemble

An ensemble of models is the best answer.
On your keys, that ensemble costs provider price.

Ensemble fans your prompt out to several frontier models in parallel, then a judge model synthesizes them into one answer. Every member runs through your own provider keys, billed at the provider's listed price. An ensemble spends 3–5× the tokens of a single call, and that multiplier is exactly where a big percentage fee turns painful.

The ensemble idea isn't ours alone: OpenRouter's own Fusion benchmark (June 2026) showed a budget ensemble scoring 64.7% on Perplexity's DRACO test, ahead of solo GPT-5.5 (60.0%) and solo Claude Opus 4.8 (58.8%). Here's what isn't on the slide: an ensemble multiplies your spend, and their cut rides along with it, 5% to use your own keys past their free tier, 5.5% to top up credits. On ROUTEXOR the same ensemble is 2.2%, or $0 on a flat plan. Less than half, either way.

A fraction of a 5.5% fee

A 5-model ensemble multiplies the tokens and any percentage fee together. On ROUTEXOR that fee is 2.2%, or $0 on a flat plan, while a 5.5% router multiplies its full cut over every member.

Your keys, your provider accounts

Each ensemble member is a normal BYOK call. The invoice lands at Anthropic, OpenAI, Google, never inflated by us.

One endpoint, full ensemble control

Pick the ensemble and the judge, or take the defaults. Same single endpoint your agents already call.

Transparent aggregate cost

Ensemble reports the true summed usage across every member plus the judge, no hidden orchestration fee folded in.

ensemble
// one call → ensemble + judge
r = client.chat.completions.create(
model="routexor/ensemble",
messages=prompt,
ensemble={ "models": [sonnet, gpt, gemini] }
)
ensemble: 3 models answered in parallel
judge: claude-opus-4.8 synthesized 1 answer
billed to your keys at provider price
routexor cut: 2.2%, or $0.00 on flat

DRACO figures are OpenRouter's own published Fusion results (June 2026), cited for the general finding that model ensembles can beat single frontier models. ROUTEXOR has not run its own eval and makes no benchmark claim of its own. Ensemble's contribution is the cost structure: BYOK, 2.2% or flat zero-markup.

Now works with Claude Code

Run Claude Code on your own keys.
Flat $67, or 2.2%. Never 5.5%.

Point Claude Code, or any tool that speaks the Anthropic API, at ROUTEXOR. Set one environment variable and your requests route through your own provider keys, billed at the provider's listed price. ROUTEXOR takes 2.2% of usage, or $0 on a flat plan.

OpenRouter's 5.5% grows with every token you spend, and a coding agent runs all day, so a big percentage fee is worst exactly where it hurts most. Start at 2.2%, and once you're coding at volume, a flat $67 doesn't move. The more you code, the more the math favors flat.

Flat $67 or 2.2%, never 5.5%

A 5.5% router scales its cut with your usage. ROUTEXOR is 2.2% pay-as-you-go, or a flat $67 with zero markup, and for a coding agent that runs all day, flat wins big.

Your keys, in your vault

Requests run on your own Anthropic (or any provider) key, encrypted server-side. Your invoice lands at the provider, never inflated by us.

One key, both surfaces

The same sk-at- key works for Claude Code (Anthropic API) and your OpenAI-format apps. One account, every tool.

Usage tracking + fallbacks included

See every request and dollar in one dashboard, with automatic model fallback: the operational wins of a gateway, without a 5.5% token tax.

claude code · setup
# point Claude Code at ROUTEXOR
export ANTHROPIC_BASE_URL="https://api.routexor.com"
export ANTHROPIC_AUTH_TOKEN="sk-at-…"
claude
routed through your Anthropic key
billed at provider price, tracked per request
routexor cut: 2.2%, or $0.00 on flat

ROUTEXOR exposes the Anthropic Messages API at /v1/messages, so Claude Code and the Anthropic SDK route through it with your own keys. You supply a provider key for the model you call.

317
models available today

Every model your agents need. Every major provider. One endpoint, all at the provider's listed price. We never mark up your tokens: our fee is a separate 2.2%, or a flat $67.

Claude Fable 5
anthropic$50/1M out
Claude Opus 4.8
anthropic$25/1M out
Claude Sonnet 5
anthropic$10/1M out
GPT-5.5
openai$30/1M out
GPT-5.5 Pro
openai$180/1M out
Gemini 3.5 Flash
google$9/1M out
Grok 4.5
xai$6/1M out
GLM-5.2
zai$1.32/1M out
GLM-4.5 Air
zai$1.1/1M out
Claude Opus 4.6
anthropic$25/1M out
Claude Sonnet 4.6
anthropic$15/1M out
Claude 4 Haiku
anthropic$4/1M out
GPT-5.4
openai$15/1M out
GPT-5.4 Pro
openai$30/1M out
GPT-5.3
openai$8/1M out
GPT-5.2
openai$8/1M out
GPT-4.1
openai$8/1M out
GPT-4.1 Mini
openai$1.6/1M out
GPT-4.1 Nano
openai$0.4/1M out
OpenAI o3
openai$40/1M out
OpenAI o4-mini
openai$4.4/1M out
Gemini 3.1 Pro
google$10/1M out
Gemini 3.1 Flash Lite
google$0.2/1M out
Gemini 3 Flash
google$0.4/1M out
Gemini 2.5 Pro
google$10/1M out
Gemini 2.5 Flash
google$0.6/1M out
Mistral Large
mistral$6/1M out
Mistral Small
mistral$0.3/1M out
Codestral
mistral$0.9/1M out
Llama 4 Maverick
meta$0.85/1M out
Llama 3.3 70B
meta$0.88/1M out
DeepSeek R1
deepseek$2.19/1M out
DeepSeek V3
deepseek$1.1/1M out
Command R+
cohere$10/1M out
Grok 4.20
xai$15/1M out
Grok 3
xai$15/1M out
Grok 3 Mini
xai$0.5/1M out
Sonar Pro
perplexity$15/1M out
Sonar
perplexity$1/1M out
Llama 3.3 70B (Groq)
groq$0.79/1M out
Llama 4 Maverick (Groq)
groq$0.6/1M out
Gemma 2 9B (Groq)
groq$0.2/1M out
Claude 3.5 Sonnet
anthropic$15/1M out
Claude 3.5 Haiku
anthropic$4/1M out
GPT-4o
openai$10/1M out
GPT-4o Mini
openai$0.6/1M out
OpenAI o3-mini
openai$4.4/1M out
+270 more

You pay providers for tokens.
You pay us for the brain.

Two ways to pay, both cheaper than OpenRouter: 2.2% pay-as-you-go, or a flat $67 with zero markup. Either way, your provider invoices stay exactly what they'd be without us.

Free$0
1K requests/day · 3 provider keys · All models · Basic routing
Starter2.2% of usage
Pay-as-you-go, no monthly fee · Unlimited requests + keys · Ensemble · Less than half of OpenRouter’s 5.5%
Pro$67/mo
Flat, zero token markup · 100K req/day · Unlimited keys · Ensemble · Task profiles · Per-job budgets
EnterpriseCustom
Unlimited · Dedicated infra · SLA · Cognitive memory integration · Slack channel

The math: at $100K/mo of spend, OpenRouter's 5.5% fee is $66,000/yr. ROUTEXOR Starter (2.2%) is $26,400/yr, and ROUTEXOR Pro (flat) is $804/yr. Cheaper either way. Pick pay-as-you-go when you're small, flat when you scale.

Build agents that route smarter,
spend less, and never go down.

ROUTEXOR is live. Built for agent builders. Free tier included, no credit card, no strings. Paid plans are 2.2% or a flat $67.