The plain definition
A router is the layer that decides which model gets a request. Instead of the agent talking to exactly one endpoint, it talks to a router that can send different requests to different models or providers — by your choice, by cost, or by availability. Think of it like a switchboard between your agent and every model you might use.
Why people want one
Cost control
Route routine edits to a cheap model (GLM/DeepSeek), save the premium model for hard problems.
Reliability / failover
When one provider is rate-limited or down, the router falls back to another so you're not stuck.
No lock-in
You're not married to one vendor's pricing, limits, or outages.
Right tool per task
Different models are better at different things; a router lets you match them.
Two ways to get routing
Standalone router vs built-in
| Approach | How it works | Tradeoff |
|---|---|---|
| Bolt on a standalone router | Run a proxy in front of the agent that speaks the agent's API and fans out to models | Extra service to run/configure; another moving part |
| Built-in routing | The agent itself lets you pick provider/model as a setting | Simplest — pick in the UI, no proxy to maintain |
Built-in routing, done simply
For most people, "routing" really means: I want to pick which model runs this, and switch easily. An agent with built-in multi-model support gives you that directly — Claude, OpenAI-compatible endpoints, GLM, Kimi, DeepSeek, or a local model, chosen in the UI. No standalone router to deploy, no translating proxy, no restart-to-switch.
FAQ
What does an agent router do?
It sits between your coding agent and the models, and decides which model/provider handles each request — letting you control cost, fail over when a provider is down, and avoid lock-in to one vendor.
Do I need a separate router service?
Not if your agent has multi-model support built in. A standalone router (a proxy in front of the agent) works but adds a moving part. Built-in routing lets you just pick the model in the UI.
Which agent has built-in multi-model routing?
Wuwei has it built in: pick Claude, OpenAI-compatible endpoints, GLM/Kimi/DeepSeek, or a local model from a list and switch in a click — no separate router or proxy to run.