AI gateway
What is an AI gateway?
An AI gateway sits between your app and the model providers. Every request goes through it on the way to a model, and every response comes back through it. That position is what makes it useful: one way to reach every model, one place to control what you spend, and one bill.
The clearest way to picture it
Like a router
It sends each request to the right model, in the format that model speaks. Change models by changing an id; nothing else in your code moves.
Like a firewall
It enforces your rules. Rate limits, spend caps, and a permissions policy that blocks the tool calls you have not allowed, so a prompt injection or a runaway agent cannot run up your bill or do damage. This part is not a metaphor; it is the job.
Like a bank
It keeps the ledger. Every request is recorded, what it cost and who made it, and settled on one bill, with checks against bad provider charges.
Most gateways stop at routing
A basic AI gateway routes and logs. Switchboard is also the ledger that bills your usage, per person, and the firewall that protects it, and it gives you the data to pick the right model for the cost and the risk. Routing is the start, not the point.
So the short answer: an AI gateway is the layer that makes inference safe, accountable, and affordable, from one place, no matter how many models or providers sit behind it.
Yes, it sees everything. That is the point.
A gateway sees every request, the same way your bank sees every transaction or a firewall sees all your traffic. That is how it can bill you, protect you, and show you what is happening. Seeing the traffic is not reading your messages: Switchboard meters the request and enforces your rules, it does not read or store the content of your prompts or replies. More on that in how we handle your data.
Common questions
Is an AI gateway the same as a proxy?
It is a proxy with a job. It sits in the path like a proxy, and uses that position to route your requests, enforce your limits, manage keys, and keep the bill.
Do I need one?
If you call more than one model, care what inference costs, or ship AI to your own users, yes. It replaces juggling provider accounts, keys, and separate bills with one place to route, control, and pay.
Does it slow requests down?
The gateway adds a thin hop and speaks each provider’s native format, so streaming, tool calling, and caching all work at the model’s full speed. You trade a small amount of latency for routing, control, and a single bill.
Does the gateway read my prompts?
A gateway sees every request, the same way your bank sees every transaction. Seeing the traffic is not reading the content: Switchboard meters the request and enforces your rules, it does not read or store the content of your prompts or replies.
Switchboard is an AI gateway that bills, protects, and helps you choose.
See Switchboard