Back to engineering

4ROUTER / Routing

Inside our routing engine

Tracing a request through selection and dispatch.

Engineering note

Begin with the candidate set

The routing engine selects candidates from the gateway model pool. Model descriptions, task context and cost constraints inform the decision. Candidate scores explain selection; they do not guarantee the quality of an individual response.

Record decision and execution separately

Choosing a model and successfully calling it are separate events. Troubleshooting needs the candidate set, the selection and the attempts that followed. The engine page contains recorded decisions; its evaluation page states the pool, workload and measurement context.

Keep the engine inside the platform boundary

Routing is one part of the platform alongside integration, accounts, billing and audit. Clients can use model selection or pin a model where the interface supports it. Integration also needs error, stream-completion and usage-record checks, beyond a successful response.

Before implementation

  • Model pool and constraints
  • Candidate and execution records
  • Errors and stream completion

Further reading

4Router · Enthalpy

4ROUTER / SYSTEMS RESEARCH

Mechanisms and experiment design

Understand how operators, KV state and scheduling shape an inference path.

Read the systems notes in Resources
GEMM / KERNELS

From operators to serving: cuBLAS & cuDNN

Identify the limiting resource before selecting the implementation.

STATE / CAPACITY

KV cache: from tensor shapes to capacity budgets

Context length is a state budget that grows with concurrency.

TIERS / TRANSFER

LMCache: placing KV state in a storage hierarchy

Avoiding prefill requires paying for state access.

Let’s talk infrastructure.

Connect to the platform, or discuss models, performance and deployment.