LLMdex
The brain behind the browser
AlohaJet is the browser your agent drives. LLMdex is the brain behind it: for every step of a task it decides which model should answer, and hands that model exactly what it needs to get the step right. Built for AlohaJet, proven with AlohaJet — and open to any agent that speaks the OpenAI API.
The promise
Big-model results, small-model bills
Frontier models are brilliant — and expensive. Most steps of a web task don't need one. LLMdex closes the gap: it already knows where things live on a page and how a task tends to go, so instead of making your model dig through raw HTML, it hands over the address and the route. A model that doesn't have to hunt doesn't have to be huge — which is why AlohaJet runs so well on cheap, open-source models.
↓ one base URL
LLMdex
self-hosted · picks the model · brings the route
↓ the cheapest model that can do the job
How it works
Cheap on day one. Cheaper every day after.
01
The ladder
Every request is answered by the cheapest model that's up to the job. Routine steps go to small, affordable models; only the genuinely hard ones climb higher. You stop paying frontier prices for routine clicks.
02
The flywheel
Tasks you repeat become routes LLMdex can replay. The first run charts the course; every run after simply flies it — faster and cheaper each time. Routes are saved on the server, so every connected agent benefits, not just the one that flew first.
03
One brain, shared by all
Your agent never has to decide which model answers — LLMdex makes that call, in one place. So when the routing gets smarter, every agent connected to it gets smarter at the same moment. Nothing to update, nothing to redeploy.
FAQ
What you'll want to know
What do I have to change in my agent?
One line: the base URL. If your agent speaks the OpenAI chat completions API, that's the whole migration. No SDK to install, no framework to adopt, no rewrite.
Can page content stay inside my network?
Yes. LLMdex is self-hosted, and every model on the ladder can be one you host too. Do that and the whole path is yours — nothing about the pages you browse ever leaves your network.
Who decides which model answers?
LLMdex does, request by request. That's the point of putting the brain in one place: every routing improvement reaches every connected agent the moment it lands — no release required, on your side or anyone else's.
The result
The proof is on the board
All of this shows up where it counts: 88% task success on WebArena, 2.2× faster, and 54% fewer tokens burned.
See the benchmarks →