Get the right answer.
Pay the right price.
The tools you already use, sending every prompt through Flux to the model that does it best. Frontier quality, intelligently allocated. One key. One decision: send.
The router that reduces your exposure.
Routing sounds like your data goes to more places. It's the opposite. One scrubbed gateway, one policy, one contract, and a guarantee you can't get a worse answer than going direct.
One gateway is safer than five
Go direct to five providers and your data lives in five places, under five privacy policies, behind five breach surfaces. Flux is one scrubbed gateway. Prompts are PII-stripped before they ever leave, and nothing you send trains a model. Ever.
A smart default, never a cage
Pin any model in one line. Every answer shows which model wrote it. Auto is a default you overrule on any call. We do the boring benchmarking so you don't have to, and you keep the wheel.
Same answer or better. Never worse.
A junior task doesn't need a senior model. But when the job needs frontier, it gets frontier. If a pick ever misses your bar, Flux escalates to a stronger model at our cost, not yours. Cost-effective stops being a risk and becomes a free win.
Run it on your own keys
Bring your own provider keys and Flux routes through your own accounts, under your own contracts, never sitting in your data path. You hold the keys; we just pick the model.
Drop-in OpenAI or Anthropic API · one line in, one line out · cancel anytime · your keys, your contracts.
Four lanes. One decision: send.
Not every job needs your best model. Auto reads each prompt, sizes the job, and sends it down the right lane. When the work needs frontier, it gets frontier. Pin a lane any time, or let Auto drive.
Classify, extract, tag, short chat. The work that just needs to be right and efficient, at scale.
pull fields from JSON
tag support tickets
Draft, code, summarize, refactor. The daily-driver lane for most of what you ship.
summarize a contract
write the email
Architecture, proofs, gnarly debugging, strategy. When the prompt earns frontier, it gets frontier.
design the system
untangle the bug
Iterate fast, then render the final on a frontier generator. You set the bar; your eye is the judge.
render the hero on a frontier model
one key for both
Stop picking models. Start shipping.
Every prompt is a different job. Flux reads each one and hands it to the model that does it best, live, on every request. This is that decision, happening now.
Every specialist, always on.
Forty-plus frontier models stay warm and online. You bring the prompt; Flux fields the one that wins it. New models join the bench the day they land.
models on the bench
Your stack shouldn't.
Integrate FluxRouter once. New models join the bench the day they ship, you get them automatically, and you never migrate again.
Frontier power, used with judgment.
You wouldn't put your principal engineer on data entry. Flux doesn't spend frontier prices on junior work. It reserves the heavy models for the prompts that earn them, and routes the rest to specialists that nail the job. Same answer. Sharper allocation.
Your whole toolbelt, through one router.
Any tool that talks an OpenAI or Anthropic endpoint plugs straight in. Point it at one base URL and Flux sits in the middle, picking the best model for every call it makes.
ContinueIDEIf it speaks an OpenAI or Anthropic endpoint, it speaks Flux. Official logos shown in the live build.
Route your whole machine, not just your code.
A menu-bar app that finds every AI tool you already run and connects it to FluxRouter in one click. Claude Code, Codex, Cursor, aider and the rest. Fully reversible: one button puts it all back.
Get Flux Desktop →Pay for routing, not seats.
Top up a balance and send. The Fast lane from $0.20 / M input tokens. Standard from $1.50. Reasoning from $4.00. You only pay Reasoning rates when the prompt earns it.
Get the right answer.
Pay the right price.
Every tool you use, every model worth using, one key. Start free in under a minute.