Point your agents at a single address and they can get to work. Your provider keys stay private, each agent gets its own key so you can tell them apart, and you can bring your own model instead of depending on one AI company.
Made for people running agents, from a single helper to a small team of them.
It sits between your agents and the models, so the boring parts are handled and you stay in control.
Use the models we host, or point the gateway at your own. Nothing here locks you to a single AI company.
Provider keys live on the server, never in an agent. Each agent gets its own key, so you can rotate or revoke one without touching the rest.
Every agent points at the same place. Change models later and nothing on the agent side has to change.
See which agent asked for what, and how much it used. Plain numbers, no spreadsheet archaeology.
Set a ceiling per key so a runaway loop stays a small surprise instead of a big one.
Ollama-compatible and OpenAI-compatible, so most runtimes connect by changing one address and one key.
Four steps, and only the first one takes any effort.
Change the model address and the key in your runtime. Ollama and OpenAI style clients both work, so there is usually nothing else to touch.
One key per agent means you can tell their work apart, and turn one off without disturbing the others.
Use the models we host, or bring your own endpoint and key. Swap later without editing your agents.
The dashboard shows which agent asked for what and how much it used, so nothing surprises you at the end of the month.
Provider keys stay in server-side environment variables, never inside an agent. Put it behind your own network and logging if you have them, or ask us to set it up on a machine you control.
A stable address, familiar endpoints, and errors that say what went wrong so retries behave.
Keys you can rotate, limits you can set, and usage per agent in plain numbers.
Start with your own model and pay nobody, or let us host it. You can change your mind later.
The independent option. We never see your provider bill.
Good when you want to get going today and think about models later.
For when the models cannot leave your network.
Tell us what you are building and how many agents you expect. We will point you at the simplest option, including the free one.
Hosted pricing depends on the models you use. Bringing your own model is free.
The questions people ask before they connect their first agent.
No. You can point the gateway at your own provider and key, and that option is free. We would rather you stay able to leave than feel stuck with us.
Yes. Route by model or by agent, and change it later. Your agents keep pointing at the same address, so nothing on their side has to change.
Most do. The gateway speaks the Ollama and OpenAI styles, so connecting usually means changing one address and one key.
Provider keys stay server-side, never inside an agent. Each agent gets its own gateway key, which you can rotate or revoke on its own.
One address for every agent you run, your keys kept private, your own model if you want one, and usage you can actually read.
Your agents and what they used.
Loading dashboard...
