Kong AI Gateway
:::info Status: roadmap (stub) The Kong adapter is scaffolding. Capabilities below are intended scope, not yet implemented. The reference adapter today is LiteLLM. :::
Intended integration
| Capability | Mechanism (planned) | Notes |
|---|---|---|
| Request transform | custom Kong plugin (Lua/Go) → POST {OPTIMIZER}/v1/optimize | Full request rewrite. |
Cache hit (canShortCircuit) | custom plugin | Served directly — Kong's plugin model can return a response from the hook. |
| Response transform | plugin → POST {OPTIMIZER}/v1/optimize-response (+ /v1/cache) | Expected to work. |
Kong is a strong target because its plugin model can report canShortCircuit: true and
serve cache hits directly (unlike LiteLLM, which delegates to its built-in cache). The
plugin will be its own package and depend on Kong's core — never a fork.
Want this sooner?
See adding a new gateway and track status in
integrations/README.md.