A gateway filters requests. A policy engine proves what happened.
LiteLLM routes across many providers, tracks spend, and adds simple guardrails. Swiftward enforces rules, versions them, and proves what happened. Different layers, and many teams run both.
What LiteLLM is
A widely used open-source proxy in front of many model providers, with one-click integrations and a large community. It routes a request and it filters a request.
What Swiftward does differently
Swiftward remembers across events: counters, windowed counters, labels, reputation, per account or per agent. A gateway judges each request on its own.
A gateway's filters are settings you edit in place. A Swiftward policy is a versioned artifact: draft, candidate, frozen, archived, rolled back by naming the previous version. You backtest a change against your own recorded traffic and read every decision that flips before it enforces, then run it in shadow or A/B on live traffic.
When something goes wrong, Swiftward names the rule that decided and the frozen version it ran on, in a hash-chained record with signed checkpoints. A gateway gives you a log. And a flagged call can wait for a person instead of failing.
Running both
Swiftward goes in front of, or alongside, a proxy you already run, and it governs more than model calls: MCP tool calls, network, FIX and blockchain come through the same policy and land in the same records.